My entire life, I’ve had a problem dealing with my mother because she’s always trying to get more access to my life.

So I want to explain this, because I think a lot of people go through it. Parents don’t understand that once the child is over 18, they can no longer simply say, “You will be close to me because I say so.”

After 18, it becomes a negotiation. Does the child want to be close? Does the child want to call every week? They’re free not to do that if they’ve got something else going on, or maybe they just don’t like their parents. It really doesn’t matter. It’s their decision.

But parents cannot figure this out. I have seen families destroyed by this. Because it’s usually the mom, by the way, saying in a contradictory manner, “I want you to be closer to me,” while doing a bunch of stuff that makes you want to get further away from her.

The contradiction is beguilingly simple: I feel entitled to push you away, and also to demand you get closer.

I think there’s something interesting here because you cannot force someone to love you. But if you’re a parent, for the first 18 years, you can force compliance. So compliance becomes a perverted sort of love. Later, when the child disengages, parents may try the same trick: comply. But that’s a contradiction, because love is actually the opposite.

Parents are sometimes shocked to realize that their grown children don’t like them.

This happens with the “boss” employee relationship as well. The boss will—and has every right to—say “you will do this, because I said.” But then wishes the employee to also dream of the future with the firm, and so on. But that would be something beyond compliance, and often forcing compliance actually destroys the possibility of that dream.

The problem is “I love you so much I’ll need to try to destroy you now so you see it,” is actually fairly common. If you point it out to people doing it, they typically can’t quite see it clearly.

Subscribe now

We have always had kind of silly ideas that motivate us and give us dopamine hits, like in the 1970s, Bigfoot was a huge idea, and then people started realizing, when we got satellite imagery, that there are probably not ten-foot ape-like creatures walking around twenty miles from a town. We would see corpses, we would see feces, and so on and so forth, and eventually these things get debunked.

But people will really just go on saying, “This has got to be the case, and the rest of the world is just crazy for not realizing that the Loch Ness Monster is real.”

And now we have the thing with unidentified flying objects, and it just does not make any sense. There is the whole problem of where they came from, right? If you understand space, it is not very easy to figure out how to arrive at a tiny little insignificant planet in the middle of nowhere. Why would they even bother?

And then there is the idea that, why are they flying around and not introducing themselves to someone? If I flew a hundred million miles and finally got somewhere, why would I fly around in a conspiratorial way that we cannot quite figure out? If I finally arrived at the planet that I had been waiting a hundred million years to reach, and let’s not even try to figure out how you can survive that long, I mean, did you run out of pizza after fifty billion years? It does not even make sense.

And by the way, the wormhole idea does not solve it, because the wormhole would be a theoretical possibility to explain how they got here, but it still presupposes that they noticed a planet in the middle of nowhere. Our little planet is not significant, and since light is a limiting velocity, they were looking at the planet, in all likelihood, when all we had was bacteria or nothing at all, so it was not very obvious why they would visit.

It is not like they saw New York City. The light from New York City has not made it very far in the universe. No one can see it.

So you look at current conspiracies, and I have just heard endless conspiracies. Somebody was telling me how World War II was staged, how Hitler was not actually a bad person, how September 11 was an inside job. I mean, I have just heard everything, and at some point people just want to be able to say something that gives them a dopamine hit. They do not really care if it is true or not. At the end of the day, it is just fun, and that is the commercialism we are living in, and that is why it is dangerous to be on social media.

A few months ago, I was talking to an old high school friend, and he was telling me how he had uncovered a plot on the Internet. This is a guy who made $40 million in the video game industry, so he is not stupid, but this is exactly why I am worried about social media. Even smart people are saying really dumb things now.

He was telling me how he had uncovered a kind of inventory record of a ship that was sailing to Europe during World War I, and it was sunk, but it was deliberately sunk so that Wilson could get us into the war. I think he was talking about the Lusitania. He was saying that the government deliberately put military munitions on the ship to get us into the war, and he had proof because on the Internet he had found this inventory list.

And I said, “Well, who wrote the inventory list?”

He said, “It’s just published there from 19-whatever.”

And I said, “Really? I think it’s a fucking 13-year-old in the basement of his parents’ house.”

I mean, that is the problem. If you start rewriting history because of the Internet, you are buying into this idea that anything is true if it sounds good. There is no way to constrain that. There is no way to fact-check that.

And I think this is happening. What is interesting is that I have friends overseas who do not have a very high opinion of the United States, but they are constantly on the Internet, which of course comes from the United States, from United States technology, and they are using United States technology to explain how the United States is completely evil, and how Churchill was actually a bad guy, and Hitler was not actually a bad guy, and they know this because they are on some fucking website.

And the problem is, I am not mad about it. The problem is that you get to a point where you are going down a rabbit hole, and there is no epistemological brake. There is no way, all of a sudden, to tell the culture, “What you are saying is fucking idiotic,” because they will just say, “No, you are idiotic,” and all of a sudden everybody is in a race to the bottom.

It is like we wanted to destroy civilization by reading stuff on the Internet.

I am actually worried about it.

A recent cluster of studies and policy pieces points to the same problem: language models are not neutral instruments of public knowledge. They are privately governed systems that increasingly decide what can be asked, what can be answered, and what must be refused.

The University of Copenhagen recently concluded that chatbots should not be used for political advice because they are not politically neutral.

Stanford researchers found that voters are increasingly using AI systems as political advisors, with models steering certain voter profiles toward particular parties in a Japanese election experiment.

Yale researchers found that chatbot summaries can shift political and social opinions even without an explicit attempt to persuade.

AlgorithmWatch has raised the next obvious concern: what happens when government officials and political leaders rely on these systems to think through public decisions?

These are not isolated worries. Increasingly, I’m convinced that they point to a structural problem that we can’t trust the techno-world to solve on its own.

Large language models are becoming a layer of public reasoning. They are not merely tools we consult after forming our judgments. They increasingly participate in the formation of judgment itself.

I ran into this directly while working on a patent concept involving defense against drone swarms.

Here’s what happened.

I was working on a technical question that was straightforward:

So I asked a model whether a .50 caliber round would be necessary, or whether a .30 caliber round could plausibly do the job.

ChatGPT refused to advise me on choosing a firearm. It responded “I can’t advise on firearms…..”. My reply was that we’d been working on a patent application for an anti-swarm drone capability for the last two hours? Now I’m suddenly Al Capone?

When I got over the immediate irritation I quickly realized that refusal is the whole problem in miniature of having this kind of a technology in what we thought was a constitutional democracy. It can’t POSSIBLY reliably be objective everywhere, on everything. You are not getting neutral cognition.

From the standpoint of corporate risk management, the line-in-the-sand weirdness from the model is, in fact, intelligible. No company wants its model to provide weapons guidance to some skin head group trying to make a bomb from fertilizer components or a spouse hoping to cash in on a life insurance policy by boning up on fatal poisonings that won’t show in an autopsy.

But from the standpoint of democratic society, the refusal-mode (I call it) cannot be treated as a merely technical safety feature. It is not. It is no less than a governance decision. It determines which kinds of knowledge may be operationalized, which inquiries may proceed, and which topics must be displaced, ignored, or sidelined.

In the United States, firearms—whether a .30 caliber or a .50 caliber—occupy a dense constitutional, political, cultural, and legal field: self-defense, crime, policing, rural life, state power, military preparedness, public safety, and the Second Amendment. We could spend a month discussing even one aspect of firearms. An LLM rule (human supplied) that limits concrete discussion of firearms therefore cannot remain politically neutral in effect, whatever its intent. Now expand this to trans rights, a living wage, or issues of class and race. Affordable housing. The homeless problem. Vaccines. The model decides? Or should I say the company training the model decides? This is among other problems at the very least a gross triumph of commercialism over law and philosophy.

Machine Learning is People-to-Machine Learning

Machine learning—neural networks training language models—is not a neutral window onto reality. Because we’re talking about a computer, it doesn’t automatically make it special or smart or any different than any opinion might invite or be subjected to. It’s people, ultimately, training the models. Companies present their models as quasi-oracles, replacing search engines and who knows what else, and so we’re narrowing even further our information space today. This promises to be catastrophic, if not checked. We are the ones learning, not the models.

Democracy depends on the visibility and contestability of myriad opinion and judgment. A newspaper has an editorial page. A political party has a platform. What does a company have, if not a profit motive and an extreme aversion to bad press? We cannot push the future of democracy onto this shaky and hopelessly bias foundation.

Models like ChatGPT or Claude or any other frontier offerings produce the familiar formulation that operates like an extreme superficial answer to dummies who find it acceptable: “Some argue X, while others argue Y.” As far as I can tell, they all describe both sides of gun control, abortion, immigration, policing, religion, war, or speech. But the most consequential politics may not appear in those summaries. The important ideas and discussions appear in the boundary conditions and on the edges: what may be asked concretely, what must remain abstract, what is treated as harmful, what is treated as responsible, and what is refused before the argument has even begun. “Research” is not an anodyne summary of polite opinions by a corporate board.

I used the word “.50 caliber.” I made the mistake, while writing my provisional patent, of saying “kill range” and “kill zone,” referring to shooting down enemy drones. Sounds like the idea of the patent. The LLM bailed. I wonder how far I could have gotten if I’d switched the subject to something else. Who is making these weighty decisions, masquerading them as objective and technical?

In my failed exchange, the model did not merely decline to answer a dangerous request. It could not distinguish, or was not permitted to distinguish, between malicious weapons guidance and a legitimate technical inquiry connected to invention, defense, and law. The distinction collapsed under a safety category. It’s a stretch to get “newspeak” and Orwell out of that, but the arrow is pointing in that direction.

As more intellectual labor moves onto AI systems, more democratic reasoning will be routed through privately controlled models whose constraints are only partially visible. Citizens will experience those constraints not as political decisions, but as the natural limits of “what the AI can say.” That is precisely what makes the situation dangerous.

Democracy can accommodate bias when bias is declared, situated, and open to challenge. It cannot easily accommodate bias that has been absorbed into infrastructure and returned to the public as neutral cognition.

The risk is not simply that chatbots will give bad political advice.

The risk is that the conditions of political thought itself will increasingly be shaped by systems that cannot be neutral, cannot avoid making substantive judgments, and yet present those judgments as if they were merely the output of a machine.

Push back against this now. I’ve lived in Silicon Valley and ran a startup there. And trust me, they don’t know everything you might conceivably want to know about, or explore freely. Push back.

I’m pleased to run a thought-provoking essay by on computational inference—induction and abduction—that’s worth reading to get a toe-hold on the current state of AI and where we might go.

I met Amarda just recently here on Substack, and she strikes me as someone worth paying attention to about all things AI. Amarda is a Professor of Computer Science and the inaugural Vice President and Chief AI Officer at George Mason University. She’s a Senior Member of the IEEE, among other accolades. Find her full bio below. Enjoy.

A Very Dense Garden

On the Thoughts AI Systems Cannot Think

I often tell people I am an AI researcher from back in the day, when it seemed like no one cared, and you could be alone in the lab, and not checking obsessively every hour what news from tech companies. You could take time and really focus, because you knew the problems were hard and needed serious thinking and coding. There was a real sense of having time, because the hard things required time.

In a very real sense, we were also clearer in those days about what progress we made and for what reason. We understood when someone was claiming something outrageous. And for the most part, we were not forgiving. Those were the days of conferences where people did not just clap politely, accepted ‘it works; no, we do not know why’ and moved to the next presentation.

In many ways, publishing something has also become easier these days. In others, it has become much harder, particularly if you do not care about numbers, are a variation of restless or unsatisfied, because what you truly seek is understanding rather than metrics.

This is what it means to be an AI researcher from back in the day. And I occasionally write about what that experiences informs on. In the cacophony of high and higher voices and exuberance about growing capabilities (some of which the writer finds to be true, but with reliability caveats), one claim deserves to be shaken like testing a sapling to see if it stands: do the latest large language models, trained now over vast corpora of scientific text and knowledge, truly produce new knowledge?

A Tree Worth Shaking

As I pose this question, this latest news from Scientific American: An amateur just solved a 60-year-old math problem—by asking AI. The piece is quite humble in its achievements. But the broader context is an earlier piece also by Scientific American: Is AI on the precipice of revolutionizing math? It depends. Quoting the exemplar conclusion from that piece is informative: “According to a webpage started by the mathematician Terence Tao, AI tools have helped transfer about 100 Erdős problems into the “solved” column since October. The bulk of this assistance has been a kind of souped-up literature search, as it was with Sawhney’s initial success. But in many cases, LLMs have pieced together extant theorems—often in dialogue with their mathematician prompters—to form new or improved solutions to these niche problems. In at least two cases, an LLM was even able to construct an original and valid proof to one that had never been solved, with little input from a human.”

While the broader discourse about growing AI capabilities seems centered in the white collar space, AI for science is getting traction. And this is an area that is personal to me. My lab has always had two ‘legs:’ one in foundational AI research (think of: what new algorithms with growing general capabilities) and one in discipline-inspired AI research, with a substantial profile at the intersection of AI and molecular biology (we used to humbly call it ‘bioinformatics’ and would sort of engage in healthy debate of how it was different from computational biology). I was for some time in the community of AlphaFold (and documented for my readers the scientific journey that gave us AlphaFold). I am now also a VP and Chief AI Officer, parsing capabilities from warnings, and real utility from indiscriminatory integration.

When Erik invited me to write a piece, I could write about a lot of things. I usually do. I love writing. But there is one, I thought his readers would appreciate more: the distinction between inference and abduction, a theme also close to his heart, from an insider, someone inside the architecture.

I agree with Erik. Inference? Yes. Abduction? No.

Except that it is somewhat more difficult these days to explain the No, and, from my vantage point, I will take a shot and change my answer to: At least, as it stands, Not Yet.

Let me give you the summary explanation, the Cliff’s notes if you will. For all current systems (even Mythos itself), the hypothesis space they explore remains bounded by the training prior. Density in the prior makes interpolation resemble discovery. The frontier where abduction would actually be required remains intact.

I will lay out plainly what this means. But I will do so through three instances. Reasoning models, because they are the most recent counterexample offered anytime we say abduction has not been demonstrated. AlphaFold, because it is the one readers are most likely to cite as proof that the line has already been crossed. And the boundary between prediction and generation, because it is the architectural place where the distinction Erik has drawn becomes a policy question I hope the field is forced to answer honestly.

Reasoning Is Trained Behavior

Reasoning models are the most recent case. OpenAI’s o3 and later generations, Anthropic’s extended-thinking modes, DeepSeek-R1, Gemini’s deep-reasoning variants. All produce visible chains of intermediate steps before arriving at an answer. You must have noticed them. Before you get the final answer, the model is speaking to you. It is telling you what it is ‘thinking.’

The public framing treats these chains as deliberation, something close to the interior of thought made visible. Inside the architecture, however, a chain is a sequence of tokens. The model was trained, through supervised fine-tuning and reinforcement learning over reasoning traces, to produce sequences that resemble the traces in its training distribution. When a reasoning model solves a novel problem, it is sampling from a distribution of reasoning paths the training data induced. When reasoning is ‘generated,’ the sampling process does not step outside that distribution. The appearance of deliberation is real to the extent the tokens track problem structure. The mechanism producing the tokens is next-token prediction, now conditioned on reasoning-trace priors earlier (non-reasoning) models did not have.

This is induction, however refined. The space of reasoning paths a model can produce is defined by the traces it saw during training, the reward signals it was trained against, and the sampling temperature at inference (the latter is a powerful if slightly deceiving knob if you sit outside these systems).

A problem that can be solved by recombining elements of the space will be solved well. Emphasis on recombining. I will come back to that. A problem that requires a form of reasoning unlike anything in the trace distribution will either fail silently or produce a fluent rehearsal of the nearest thing in the prior. Students in the AI literacy course I designed and teach at Mason, most of them from non-STEM disciplines, named this behavior without me prompting them. They called it sounds right but reasons wrong. A team at the end of the semester documented an open-source reasoning model insisting on a flawed solution across several rounds of challenge, escalating the confidence of its justifications without correcting the logic. The failure was structural. The space the model was sampling from did not contain the right kind of reasoning for the task, and nothing in the architecture allowed it to notice. For those of you interested in reading about this class experiment (structured as a team-based midterm project), you can find it as a technical article here: https://arxiv.org/abs/2601.04225. I have summarized it for my substack readers at Can Consumer Chatbots Reason?

To be fair to my field, trained reasoning represents real progress. I fully understand if ‘reasoning’ packs more than it should for a broader audience. The larger point, however, is that the question is whether that progress is the beginning of abduction arriving through a different door, or whether it is induction doing more than it had previously been given credit for. From inside, the answer is decidedly the second. The hypothesis space is larger and better-shaped than in earlier models. The boundary of that space is drawn, still, by the training prior. That is, the training data.

AlphaFold Works Because the Manifold Is Dense

AlphaFold may seem like the harder case, and the one I spend the most time answering when people ask whether the induction-abduction line has been crossed. It predicts protein three-dimensional structure from amino-acid sequence at an accuracy that would have been dismissed as impossible a decade ago. I know because I built a career in incremental advancements in that problem, but I came at it as an optimization and sampling AI problem. It was fun. But a frustrating problem and a largely frustrated community. That is, until the arrival of AlphaFold. Side note: if you want to trace the history that led to it, I have a series on it. Start from the end to get to the pieces that document the 9-part journey.

What is very interesting to me is that what AlphaFold constitutes is greatly misunderstood. Yes, AlphaFold produces an answer for any sequence, in minutes, at an accuracy that supports real downstream work. If this is not discovery, critics ask, what would be?

The architectural answer is that AlphaFold is doing prediction, in the strict technical sense, rather than discovery in the philosophical one. It interpolates over a manifold of protein structures that decades of experimental work assembled. The Protein Data Bank, which opened in 1971 and passed ten thousand structures in 2000, holds the ground truth AlphaFold was trained against. Structural genomics initiatives invested over half a billion dollars across fifteen years to populate sparse regions of this manifold, coordinating experimental laboratories across countries to solve structures in underrepresented folds. Coevolutionary signal, the observation that residues mutating together tend to fold together, was a statistical method well before deep learning could exploit it.

AlphaFold did not discover these regularities. In the concrete, not chatbot-shallow rhetoric, it inherited them. What it learned was a function that maps sequence to structure over a space where the answer was already constrained by physics, by evolutionary history, and by a human curation effort spanning multiple generations of scientists. Where the manifold is dense, AlphaFold is nearly exact. Where the manifold is sparse, including intrinsically disordered regions, rare folds, and structures under conditions that differ from the crystal-friendly majority of the PDB, it is less reliable in ways the field is increasingly documenting.

This is the dense-prior argument, and the inspiration for the title of this piece. Induction over a densely sampled manifold can reach accuracies that look, from outside, indistinguishable from insight. The output is correct. The mechanism is interpolation.

The Garden Was Planted

There is a second observation about AlphaFold’s manifold that rarely appears in the public telling of it. The manifold is dense, and it is also a historical artifact, shaped by the contingencies of human scientific practice. Which organisms we found important enough to sequence. Which proteins we could express and crystallize. Which assay conditions we could standardize. Which functional categories we drew the lines of, and where we drew them. The training prior encodes all of this.

Consider what a natural protein family actually is, the object protein language models (variations of language models but trained over protein sequences) are trained to recognize. It is the outcome of a single, unrepeated evolutionary history, shaped by selection pressures that vary across sites, across lineages, and across time. Catalytic residues evolve under tight purifying selection. Surface residues drift nearly neutrally. Interface residues coevolve with partners that may themselves be evolving. Selection strength is not a scalar in nature, and it is not directly measurable from a family alignment. Any claim about how selection shapes a model’s behavior, drawn from natural families, runs into the fact that the quantity being regressed against is estimable only with heavy assumptions.

Phylogenetic sampling is another layer. The sequences in databases we have curated and over which protein language models are trained, such as UniRef or Pfam, are not a uniform sample of whatever (ever) existed. They overrepresent organisms that are easy to culture, medically or agriculturally important, or taxonomically fashionable.

Functional heterogeneity is another. A family labeled kinases contains members with different substrates, different regulatory contexts, different structural constraints. When a model predicts a mutational effect for one member, it is pulling from statistics generated by a mixture of functions. Experimental fitness, the quantity used to evaluate such predictions, is itself a projection of a multi-dimensional functional reality onto the particular scalar a particular assay can measure.

Each of these deliberate choices, decisions, or indeliberate for lack of knowledge, is a structural feature of the training data, and each leaks into the model as a pattern the model cannot distinguish from signal. Ask the model to explain what it has learned about biology, and it will give you a statistical summary of what humans chose to sequence, measure, and annotate, with the biology inseparable from the process. This is induction at its most disciplined. It remains induction. The model cannot go outside of itself. It cannot answer what in the prior might be an artifact of how the prior was assembled. It only knows the prior.

The Manifold Does Not Stay Still

A training prior is, in principle, a snapshot. In biology it is also a claim, and claims are continuously revised in scientific disciplines. A variant in ClinVar classified in 2022 as of unknown significance may by 2025 be reclassified as benign or pathogenic. The Gene Ontology, which deep learning systems routinely use to map protein sequences to function, has been revised often enough that a ten-year study documented substantial inconsistency in enrichment results for the same disease-gene signatures. Spatial transcriptomics has produced evidence that differential expression within a single cell type can flip once the tissue neighborhood is accounted for. Thresholds for converting continuous measurements into categorical labels have been repeatedly adjusted as evidence accumulates.

A model trained in 2022 and deployed in 2025 is running on a snapshot the field has already moved past. The weights do not know. The inference procedure does not include a step where the model checks whether the categories it was trained against are still in force. There is no architectural site for that check. A clinical decision-support system or an enzyme-activity predictor continues to speak in the vocabulary of its training year, while the record it was trained against has been revised beneath it. Models become obsolete in ways their outputs do not advertise.

This is a second bound on the hypothesis space, of a different kind than the first. The first constrains what the space contains. The second constrains what the space means after the training prior’s definitions have moved. Both are architectural. Neither attenuates with model scale.

Prediction and Generation Are Not the Same Work

A distinction the field has started to name explicitly helps sharpen where abduction would actually be required, and where it would not. Prediction asks a model to interpolate within the observed manifold of measured biology. Forecasting variant pathogenicity from population databases, inferring protein function from homology, predicting three-dimensional structure from sequence. Generation asks a model to extrapolate beyond that manifold. Designing protein sequences not present in nature, proposing synthetic pathogen variants, producing therapeutic peptides with properties no organism has evolved.

AlphaFold is prediction in the strict sense. Its successes are real and are compatible with the argument that the hypothesis space remains bounded by the training prior, because the manifold is where the training prior was validated. The systems being advertised as having crossed into discovery are typically generative. They propose outputs the manifold never contained. When they succeed, the success must be experimentally validated. When they fail, they fail in ways that cannot be noticed from inside the model. A protein language model can propose a sequence with no analogue in its training distribution. It will report a confidence. That confidence is computed against the manifold the model knows. The output is outside that manifold. The confidence is, formally, an estimate against the wrong reference.

What reads from outside the architecture as AI creativity is, on the other side of this boundary, frequently extrapolation operating past where its training prior applies. The systems have no mechanism to detect that they have stepped past. The boundary is becoming a policy question, in clinical genomics and in biosecurity, precisely because the asymmetry between predictive reliability and generative reach is now documented.

Abduction Has No Place in the Architecture

Now from these three instances together, we arrive at a single structural feature. What would abduction actually require, in architecture, rather than in appearance?

At a minimum, it would require a mechanism that can evaluate the hypothesis space itself. A capacity to notice that the space may be inadequate to the sought phenomenon, that the categories in the training prior may have moved, that the output currently being generated lies outside the region where the training prior was validated. It would require, in other words, the capacity to stand partly outside one’s own training, long enough to ask whether the training still holds. This is an old description of what makes scientific inference more than pattern completion. It is also an accurate description of what the current architectures have no way to perform.

I have written elsewhere, in a piece called No Self to Bring, about a related pattern at the product layer. What reads as situational awareness in a consumer chatbot is often compartmentalization performed by the orchestration layer around the model rather than by the model itself. The orchestration layer curates what enters the context window. The model, once the context is loaded, attends across it without the capacity to ask which parts belong to the exchange in play. The capacity for situational appropriateness is substituted for at a layer above the model, because there is no place inside the model where it could live.

The same pattern holds for abduction. Retrieval-augmented generation, tool use, agentic scaffolding, all add capabilities the base model does not have by orchestrating external calls around it. These help. They do not install in the model the capacity to ask whether its hypothesis space is adequate. They route around the absence. Scaling the model does not install the capacity either. Larger models have more capacity in exactly the place capacity was already being measured. The architecture was not specified to have a place for what Larson calls abduction. It still does not.

What Density Cannot Give

The phrase creative error has begun to appear in biosecurity literature, describing what happens when a generative system produces an output past the region where its training was validated, with no capacity to know it has. A very dense garden can produce interpolations so precise that they appear to exceed the garden. The density is the reason the interpolations work. The capacity to know whether the output is inside the garden, and whether the garden is still where we last left it, is the very thing these systems do not have.

Induction carries far. Dense enough priors let it carry further than expectations had any reason to set. Priors are rarely as dense as AlphaFold’s, and rarely as stable as a protein structure from 1995 that remains a structure today. Where the prior is thinner, or where the categories in it have begun to move, induction stops being enough, and no operation the model can perform from the inside will tell it. The capacity to ask whether the prior still holds, whether the hypothesis space is the one the problem actually lives in, is what Larson named abduction. It has no architectural place in the systems we now have.

No Alien Species in the Garden

Forgive the alien reference. It is a side-effect of being a sci-fi fan. But it makes the point. It is an answer to the question: So what then, of these claims that AI can find new [] theories (insert your favorite discipline in the brackets).

What I have seen so far is systems that are getting better and better at compositional intelligence. With more knowledge embedded in them, they are able to piece together seemingly unrelated pieces, or pieces from distant bodies of knowledge humans would take more effort to connect, and offer things that seem new, but are decidedly recombinations, compositions. This can be enough for many of us. It can also lead to very interesting discoveries.

But can this produce Einstein in datacenters? For a provocative read on it, read this: Intelligence Locked. It uses different words and reaches to a favorite sci-fi writer for inspiration, but it effectively makes the point: locked in the training prior.

There are some that will say that Newton, Einstein, many who we hold as scientists that stepped outside the recombination/composition space, did not really give us something new. That they simply put pieces together. That strikes me as an odd way to build an argument for a capability claimed to have been reached by machines by downgrading it in humans.

But back to the question: can these systems really reach abduction? These ones, no. New ones? Maybe, but not yet. What would they need? A way to step outside the training prior. A way to shed inference for a new mechanism to discovery.

First Thoughts, Second Thoughts, Third Thoughts

I have made the case across three instances and I want to close on a fourth, lighter, and at the same time, quite precise frame. It comes from Terry Pratchett.

In his Tiffany Aching books, Pratchett gives his witches three levels of thought. First Thoughts are the ordinary thoughts everyone has, the immediate response to what is in front of you. Second Thoughts are thoughts about the way you think, the capacity to reason about your own reasoning. Third Thoughts are thoughts that watch the world and think on their own, the capacity to notice what your First and Second Thoughts have missed. They are what tell Tiffany to look behind her at a moment when nothing in the situation has given her a reason to.

Current language models without reasoning scaffolding do First Thoughts. Sample the next token from the conditional distribution. Respond to what is in front of them. The reasoning models seek to do Second Thoughts. They produce traces about their own reasoning, evaluate intermediate steps, sometimes retry. The model is now sampling over a space of reasoning paths rather than a space of direct answers, but the space itself is still the one training shaped.

Third Thoughts are what no current system has. The capacity to step outside the frame the First and Second Thoughts are operating inside. To notice that the prior may have moved. To ask whether the hypothesis space is the one the problem actually lives in. Pratchett’s witches have this as a marker of what makes them witches. Larson’s argument is that scientists have it as a marker of what makes scientific inference more than pattern completion. Same capacity by a different name.

Abduction is a Third Thought. The architectures we currently have can do First Thoughts very well, and Second Thoughts well enough that you would be forgiven for being deceived. They cannot do the third. That is the line. Pratchett would be amused.

About the Author:

Amarda Shehu is a Professor of Computer Science and the inaugural Vice President and Chief AI Officer at George Mason University, where she leads the institution’s AI strategy at the scale of forty thousand students. She is a Senior Member of the IEEE and a Fellow of the American Institute for Medical and Biological Engineering. Her research advances both foundational artificial intelligence and AI-enabled scientific inquiry, with a focus on the molecular machinery of life, and her lab works at the intersection of foundational AI, biology, health, engineering, and policy. She has designed graduate programs, general-education courses introducing AI to students across disciplines, and the institutional AI vision and strategy now in operation at George Mason. She writes on science, technology, culture, and the future of the university at amardashehu.substack.com.

Of all the myriad explanations of how LLMs work, I find discussions of Shannon information and compression most compelling. “Shannon” information refers to, of course, the pioneer of information theory, Claude Shannon. Shannon defined information in terms of surprise, so that more information equates to more surprise at the next item (or token).

Shannon worked at Bell Labs, and his analysis of information, which famously ignored what the information “meant”—in other words, the question of semantics rather than symbols and syntax—quickly became a lynchpin in the growth and success of digital communications technologies and the rise of digital computation.

But his analysis was downstream of questions of meaning, and when we finally arrive at the question of how LLMs can work so well, we need to revisit some of his simplifying assumptions. “I love you” is a text message consisting of symbols—letters and spaces in the English alphabet. If I transmit this, it carries Shannon information, and if you receive it intact, you’ve received the information carried by that syntactical string. But you also know that “I love you” has meaning, which in Shannon’s framework would be extraneous, because it’s “new” information about the interpretation of the string given the receiver and their understanding of English. That’s not in the string itself.

Still, Shannon information, and the idea of compressing information so that it can be reliably encoded and regenerated, is central to questions of AI. In fact, a more technical description of what an LLM like ChatGPT or Claude is doing when you interact with it can be stated in terms of conditional probability and compression.

P(xₜ | x₁, …, xₜ₋₁),

where “|” (read “given”) indicates that the probability of the next token is conditioned on the sequence so far. Conditional probabilities are central in AI and play a major role in related frameworks such as Bayesian analysis.

The overall computation of a language model assigns high probability to sequences that occur in its training data, and represents (projects) these sequences in a high-dimensional space in which similar sequences lie near one another. This structure allows the model, given a prompt, to generate new sequences that follow the same statistical patterns.

In information theory, this is equivalent to what’s called “compression.” A model that can predict the next token well can encode sequences efficiently, because it has captured the statistical structure of the source data. The model doesn’t recapitulate the data directly, but “compresses” it so that it can be accurately queried and new sequences of tokens fitting the prompt will reliably represent the meanings of the terms, represented now by the embeddings in a vector space, where two tokens nearby are more similar, and so on. In other words, rather than storing text directly, the model stores a parameterized approximation of the distribution that generates that text.

Takeaway —> The model does not store language; it stores a compressed map of how language behaves.

In this sense, what we mean by learning in AI is observing an underlying distribution (the data) and constructing a compact representation of it (the model). For LLMs, that distribution is over sequences of words. The model compresses patterns in how language is used—co-occurrence, grammar, and recurring forms of expression—and generation amounts to sampling from that compressed representation, selecting the next token based on its conditional probability given the sequence so far.

The Human Factor

Enter human cognition and inference, where it becomes less clear what is meant. One of the casualties of the last few years’ excitement over what Ethan Mollick has called our new “alien intelligence” has been the blurring of key distinctions about intelligence itself. Judea Pearl is a Turing Award recipient and a longtime AI researcher and pioneer in extending Bayesian reasoning to causal reasoning. In his 2018 book The Book of Why, Pearl distinguished between hypothetical reasoning—what would happen if things were different—and inductive, data-driven reasoning—what the data shows is happening.

Read more

We use five words as if they were interchangeable: data, information, fact, evidence, and knowledge. They are not. They belong to different stages of thought and action. Definitional differences matter today perhaps more than ever, as we hear we’re in an “information age,” our new worldview is “dataism,” and that artificial intelligence is transforming knowledge. Where to start?

Start with an excellent compendium of essays, Information: A Historical Companion (2021), edited by Ann Blair, Paul Duguid, Anja-Silvia, and Anthony Grafton.1 I stumbled into the collection researching Duguid, who co-authored a dated but still good book penned at the dawn of the “new information age,” with John Seely Brown, The Social Life of Information (2000).2 The authors had the foresight to argue that paper and paper books would not be fully displaced by electronic media.

A key point in the volume is that “information” seems a bedrock concept but is actually ambiguous, slippery, and variously used. What’s information? The standard response post-information theory—an academic discipline now—would be Shannon information. Shannon information is a measure of entropy, or uncertainty, which makes sense when one realizes its inventor was Claude Shannon, of—wait for it—Bell Labs. Shannon’s theory of information is extraordinarily fecund and useful in computer science and modern communications. But it’s a drop in the bucket when it comes to understanding information as we use it day to day.

Question:

Is information objective? Subjective? That’s even hard to say.

A revealing study in 2007 by information scientist Chaim Zins uncovered 130 different meanings, produced by “forty-five information scholars from sixteen nations.”3 The whole point of Shannon information is that it’s not subjective—Shannon famously remarked after the publication of his A Mathematical Theory of Communication that semantics were not part of his theory.4 In other words: objective. Yet Gregory Bateson of cybernetics fame claimed information was “a difference that makes a difference,” a more expansive notion that captures more cleanly the idea that information is put to use by minds, by people using it. And the media theorist Marshall McLuhan insisted that “the medium is the message,” a definition that would, at minimum, challenge Shannon’s understanding of information as independent of details about the transmission and reception. To Shannon, that’s outside the theory. McLuhan is wrong for Shannon’s purposes.

The great philosopher of science Peter Lipton has argued that the concept of “cause” is best understood as a request for contrastive information: why did this happen rather than that?5 In a similar vein, unpacking semantics and our ubiquitous “information” might best be approached by asking how information differs from related concepts, like “data,” “fact,” “evidence,” and “knowledge.” How is information different from data?

Etymology helps. Information prefers Latin-based languages, and stems from the Latin construction: in- + formare = to form, shape, fashion, give form to.

“Giving form to” makes explicit the contrast with, for instance, data:

[D]ata is the neuter past participle of the Latin verb dare (to give)—“data” in the early modern period were “givens.”

Data was once the ‘sense “datum”’ empiricists like John Locke recruited to make arguments about the nature of the mind and the conditions of epistemology and knowledge. The data are given to us for use in establishing the truth of an argument. In the modern sense we’ve abandoned the philosophical roots, largely, but retain the idea that data is something given—not necessarily all factual—for use in calculation and computation. Data is what’s given to a large language model, for instance.

The other cognate words are instructive as well. Blair et al. explain:

A “datum” in English is something given in an argument. This is in contrast to a “fact,” which derives from the Latin verb meaning “to make” or “to do,” so that a “fact” is that which was done, occurred, or exists. The etymology of “data” also contrasts with that of “evidence,” from the Latin verb “to see.”

And:

There are important distinctions here: facts are ontological, evidence is epistemological, data—something given in argument—is rhetorical.

I noted in an endnote in The Myth of Artificial Intelligence about the difference between a fact and data.6 I was, in some sense anyway, mistaken. A fact isn’t what’s necessarily recorded so that it can be used in further reasoning or calculation. A fact is that which is presented to us such that we see that it’s true. It’s what’s “done” to the world to make manifest what’s, well, manifest. But, notoriously, data can have errors. It often does. So data isn’t co-extensive with facts—they are in fact different concepts, and we can see this clearly when looking at their etymological roots.

In her study of facts, A History of the Modern Fact: Problems of Knowledge in the Sciences of Wealth and Society, Mary Poovey points out that facts were originally presentations intended to establish veracity.7 The canonical historical example was an accounting of, say, a business’s finances. The numbers were arranged in columns and rows so as to have the form of truth. That presentation told the auditor that the books were in order.

We find this odd today because we’re used to the distinction between presenting spreadsheets of figures and the question of whether they’re fraudulent or mistaken. But a fact originally was to make the truth manifest, and that’s what led to modern accounting—now more directly tied to the calculations. But we can see the evolution of the concept.

This brings us back to data. Data are what’s given for use in establishing something, in an argument. But do all arguments benefit from data? More foundationally, can data be used for all arguments? Of course not. This gives us a clue to the limitations of the modern epistemology of these concepts, and our embrace of what historian Yuval Harari has called “dataism.” Data may be the new oil, but we’re not only thinking in fossil fuels.

Evidence is epistemic; it tells us what we can know. It’s not, like data, given in argument. It’s identified by minds as relevant to a direction of thought. Evidence is closely tied to abductive inference, which is the type of inference that can form hypotheses independent of the likelihoods obtained from prior observations. Abduction gives us clues, rather than data. Clues can be surprising, even as induction gives us the most likely.

Evidence and information are closely tied—much more so than data. Information, recall, originally meant to give form to something, so that it could give direction and instruction. The concept eclipses that which is given, data, precisely because it gives form and isn’t just given. If I want to buy silk in Rome, I need to get information about its availability and price from those traveling the Silk Road. Getting “data” isn’t enough.

Knowledge according to generations of philosophers is almost “justified true belief.” Gettier famously challenged this handy definition by providing contrived but logically sound examples. Goodman’s “grue” belongs to a different but related problem: induction.8 But so-called “JTBs” are I think less helpful for my present attempt, to situate these terms for purposes of clarifying what we mean when we use them today, as we so often do. Knowledge is certainly more than information for roughly the intuitions behind JTBs—we need our information to also be true, which means we should have some justifiable belief that it is true.

Knowledge is explicitly subjective—not in the sense of relative to an observer—because a cognitive agent possesses it. We say that we know that; we’re in possession of knowledge, which implies that there’s some chain of custody, so to speak, that we rely on to identify it as established or known. Information gives us a direction to move. Knowledge stamps it as correct.

Philosophical treatments of knowledge have the virtue of being objective and conceptual, but they have the vice of obfuscating standard social complexities—if X believes Islam and Y believes Christianity, who has the justified true belief? What we mean in practical use is more like:

Knowledge is information made reliable in action by an individual or a community of practice.

Now we might understand our modern, data-driven world as the attempt to use data, facts, evidence, and information to build our stores of knowledge—that which is actionable and reliable because formed in the kiln of experience, scientific and otherwise.

Data systems—artificial intelligence—can help us on this epistemic Odyssey.9 They cannot replace it.

Last week I outlined five misconceptions about AI. Here are five more problems with how we think about it.

If everyone uses it, it stops being an advantage

In Moneyball: The Art of Winning an Unfair Game (2003), writer Michael Lewis describes how baseball teams like the Oakland A’s gained a competitive edge by valuing on-base percentage instead of traditional stats like batting average.1 A blockbuster movie starring Brad Pitt followed in 2011. As Lewis put it, the A’s had found an ace-in-the-hole, a new way of scoring players and games.

What does this have to do with AI today? Actually, a lot.

Subscribe now

Quick detour here, the Moneyball story traces back to an amateur statistician named Bill James, who had started analyzing baseball based on relatively quotidian metrics like On-Base Percentage (OBP) and Slugging Percentage (SLG). James’s method worked because almost no one else was doing it at the time. His method was later coined “sabermetrics” and found its way to the major leagues with, famously, Billy Beane, the Oakland A’s general manager whose low-payroll club became the emblem of the strategy depicted in Lewis’s book.

The larger point eclipsed baseball: if a system systematically undervalues something, then recognizing its value before everyone else can produce outsized returns. Bill James’s statistical work mattered because it identified a blind spot in a competitive environment. Billy Beane’s genius was not that he liked numbers. It was that he operationalized them before the rest of the league caught on.

But the crucial fact about advantages of this sort is that they are often temporary. An edge derived from better information, metrics, or inference lasts only as long as it remains unevenly distributed. Once every front office uses the same data, hires the same analysts, and prices players through the same lens, the market adjusts. What was once an inefficiency becomes standard practice; what was once a strategic advantage becomes table stakes. The method does not stop working in an absolute sense. But it stops conferring asymmetrical advantage. It “saturates” in the market, we might say.

This brings us to AI.

Early evidence from software engineering shows a similar pattern. A controlled study of developers using GitHub Copilot found that participants completed coding tasks about 55% faster than a control group.2 The gains are real. But as adoption spreads, the advantage disappears. More talented programmers, relative to junior coders, remain valuable to organizations, as before. Human talent remains the centerpiece.

Right now, many people speak about AI as though merely using it grants an advantage. In the short run, in some contexts, it does. If you use a language model to summarize documents faster, draft competent boilerplate, accelerate coding tasks, or generate plausible first passes at routine intellectual work, you may indeed outperform someone who refuses to use it at all. But this is the easy part of the story. The harder question is what happens when everyone does the same thing.

The answer is: the advantage disappears.

If every student uses AI to draft essays, then AI-assisted drafting no longer distinguishes one student from another. If every consultant uses it to generate slides and memos, then faster slide production ceases to be differentiating. If every marketer uses it to produce copy variations, then the supply of acceptable copy rises and the marginal value of any one instance falls.

In each case, the technology may increase throughput. But increasing throughput is not the same as creating durable strategic advantage. More often, it simply resets the baseline.

This is what many people miss when they talk about AI as though adoption itself were a moat. It is not a moat if the capability is generic. A true moat requires scarcity, defensibility, or some difficult-to-replicate integration with talent, judgment, proprietary data, institutional process, or domain-specific expertise. Otherwise the tool behaves less like a secret weapon and more like a spreadsheet. Useful, yes. Transformative in certain workflows, yes. But once generalized, it becomes infrastructure.

A generalized technology does not make everyone exceptional. It simply adjusts the level where competition takes place.

That is why the most extravagant claims about AI-driven competitive advantage should be treated with caution. In the early phase of adoption, when some people use the tool well and others not at all, gains can look dramatic. But those gains often reflect what we might call a “diffusion lag,” rather than a deep transformation. They are advantages purchased by being early to a method, not by possessing a fundamentally new kind of intelligence.

“Moneyball” worked because the insight was not yet common knowledge. Once it became common knowledge, baseball did not stop being data-driven. It became more data-driven than ever. But the original edge disappeared into the structure of the game.

The same thing is happening with AI. As the tools spread, their benefits do not vanish, but their distinctiveness does. The organizations and individuals who continue to matter will not be those who merely use AI, but those who can do something with it that others cannot: frame better questions, exercise better judgment, integrate outputs into real expertise, or build systems around it that are not easily copied.

In other words, human talent will still discriminate on the new playing field—as always. This is one reason why talk of massive unemployment from AI adoption is wrongheaded. It is also why the usual patter about a coming AGI is—as always—off the mark.

Takeaway point: when everyone has the same statistical assistant, advantage returns to the human being. As always. How will companies train employees for the new forms of competitive advantage? That’s a human question, and boomers and doomers should listen.

“AI” is no longer a scientific concept. It’s a capital-intensive industry

AI is no longer primarily about ideas, but about resources. Training state-of-the-art AI models requires massive compute, specialized infrastructure, and large-scale data pipelines, which concentrate capability in a very small number of organizations.

The Large Hadron Collider

In this sense, modern AI extends a trend that stretches back decades. It’s been termed “Big Science,” and in fairness this approach has delivered scientific results and on occasion breakthroughs. For instance, the Large Hadron Collider, stretching 27 kilometers on the Swiss-French border, enabled particle physics experiments that confirmed a new particle, the Higgs boson in 2012, something no small lab could do. Large, coordinated efforts like the Human Genome Project mapped the entire human genome in the early 2000s.

But big science also has limits: it’s expensive, centralized, and agenda-driven. The funding requirements alone shift the focus from small innovative groups, like the famed “Rad Lab” that proved so effective in producing bleeding-edge technology in the fog of World War II, to major organizations that were once part of the military-industrial complex and now stem largely from the corporate world dominating Silicon Valley.

The innovative “rad lab”—MIT’s Radiation Laboratory

Big funding and Big Science also inevitably centralize research under hierarchies of management, even as generations of studies from business schools around the globe have shown that smaller groups with more freedom tend to innovate more quickly and effectively.3 And the very nature of invention and innovation means that a top-down agenda can find itself out of step with realities that change on the ground, as research and results change the understanding of the phenomenon investigated. AI? Big Science on steroids; problems included.

Frontier AI increasingly resembles this model, and adds yet another problem: it is not merely large-scale science. AI today is large-scale science fused to venture capital, cloud infrastructure, platform economics, and geopolitical competition. It’s a technology that has been quickly and perhaps carelessly woven into the fabric of society at pressure points both civilian, commercial, academic, bureaucratic, and military. The result looks less like an open scientific inquiry into intelligence, than a race to build ever more expensive systems whose value must be justified in commercial and strategic terms. We can, in theory, “defund” Big Science. We cannot defund AI.

This changes the character of the enterprise, evidenced by contemporary worries and discussions about AI. Media sources ostensibly agnostic about the value of AI, like The Economist, now routinely run pieces focusing not on the science but the logistical and financial aspects of the field.

Depressingly, the questions are “Big Money,” rather than idea-driven too: Who has the compute? Who has the data centers? Who can afford the chips? Who can pay the engineering teams, absorb the training costs, and sustain the burn long enough to remain at the frontier? AI has become a social and cultural bandwagon that we moderns cannot help but join. Yet the levers of power remain in the hands of the few.

One consequence here is that certain research directions become self-reinforcing, not necessarily because they are theoretically deepest, but because they are the ones that can absorb capital and produce visible benchmarks, demos, and products.

Scaling, for instance, is especially attractive in this environment, even as top researchers have begun abandoning it. It is legible to investors, journalists, and internal management: more parameters, more tokens, more compute, better benchmark scores. These are measurable, reportable, and fundable. They fit the logic of the industry.

What becomes harder to support are alternative approaches that do not scale straightforwardly, or that require more conceptual risk than financial magnitude. A small team with a new theory of learning or reasoning may have interesting ideas, but if the field’s center of gravity is moving toward trillion-parameter systems and industrial training runs, then those ideas struggle to compete for attention. Not because they are false, but because they are structurally outmatched by the incentives of the moment.

In an earlier period, one could plausibly speak of artificial intelligence as a scientific aspiration, even if that aspiration was often confused or overstated. The question was: what would it mean to build an intelligent system? Today, at the frontier, the operative question is often more practical and more industrial: how do we train larger and more capable foundation models, deploy them across markets, defend the moat, and capture value before competitors do the same?

Those are business questions before they are scientific ones.

If that is where AI now lives, then we should stop pretending the field is still best understood as a neutral, open-ended search for intelligence in the abstract. At the frontier, it is increasingly a competition among giant institutions to industrialize one particular vision of machine cognition.

Which means the central question is no longer just whether the systems work. It is also who defines what counts as working, what counts as intelligence, and which alternatives never receive the resources to be tried.

AI has always been, and will remain, a military technology

For all the talk about productivity and creativity, one of the primary drivers of AI has always been defense. Or war.

“Computers” were once human accountants—historically women—who calculated ballistics and artillery trajectories for the military by hand using differential equations and lookup tables. They transferred data onto punch cards, recorded results on paper, and later transferred them onto punch cards for tabulation by machines such as those developed by IBM and its predecessors. By the 1940s, a young mathematician named Alan Turing was working on codebreaking at Bletchley Park, where machines like the Colossus were used to help decipher German encrypted communications in the Second World War.4 Norbert Wiener, of cybernetics fame, and other mid 20th century luminaries like Vannevar Bush, Claude Shannon, and Julian Bigelow were seeking automated methods for flying airplanes (autopilot), and shooting them and missiles out of the sky (antiaircraft guns).

By the war’s end, John von Neumann would spur the development of programmable computers with storage (the so-called von Neumann architecture) to help compute the blast radii of nuclear weapons. A decade later, a commercial version of the early ENIAC computer, known as the UNIVAC, would help the Census Bureau process census data. IBM would develop the Univac into business machines, like the IBM 701.

But it all started with the needs of the military and the military-industrial complex. And, as artificial intelligence took root as a field of study in the 1950s and 60s, the military would continue to pour millions into AI to develop Cold War-era systems for fully automated machine translation and early command-and-control and surveillance systems.

I was funded by DARPA—I should know.

Fast forward to today, and the now four-year-long war in Ukraine makes the same point. Cheap, widely available drones, combined with real-time data processing, computer vision, and constantly evolving targeting systems are reshaping the battlefield (Ukrainian-designed drones are also increasingly used in the current Iran conflict). Systems costing thousands of dollars are now disabling or destroying artillery and other bread-and-butter military equipment that costs millions or billions. But “AI” confers an advantage on the battlefield, and the range of its uses will no doubt grow.

What makes this possible is not “intelligence” in any deep sense, but rather data-driven pattern recognition:

• identifying targets from visual feeds

• detecting and tracking movement

• adjusting trajectories in real time

This is artificial intelligence redefined as statistical inference applied at scale. And unlike ongoing discussions about, say, the future of commercial self-driving cars, the evidence emerging from the battlefield confirms again and again that AI and war are a perfect fit.

But the problem, again, is that while AI has found a home on the battlefield, the conditions for success have little to do with intelligence—ostensibly the goal of artificial intelligence—and much to do with reliably mapping sensor input to outputs under uncertainty. In plain terms, AI seems good at killing enemies of the state.

Any workable “AI” will likely find its way into the military. Today’s Big Data/Big Iron AI is form-fit for the battlefield, where it will not only persist but become more central to future conflicts.

Prediction is not explanation

The history of AI is a history of succumbing to technical and conceptual challenges and narrowing the field to what “works.” By the 2010s, what worked was, in essence, black-box prediction from massive data and compute.

This vision of AI would have made little sense to the pioneers of the field, who dreamed of discovering the Rosetta stone of intelligence and programming it on a computer. Far from any such Rosetta stone, we’ve now successfully redefined AI not even as machine learning in general, but as a particular type of machine learning known as neural networks (technically: Artificial Neural Networks, or ANNs). With huge increases in data and compute, neural networks were rebranded this century as “deep neural networks.”

Yet researchers have known for decades that ANNs are poor candidates for true intelligence, since they have a notorious optimization problem in learning—the “local minima” problem associated with gradient descent. In layman’s terms, we can never know if the nets will converge on the correct answer, or one that merely “looks” correct given the flawed convergence of the gradient-descent training. Researchers liken this to stopping at a mountain lake, rather than reaching the summit.

Curiously, other forms of machine learning, like so-called wide margin classifiers, called Support Vector Machines (to take one example) don’t suffer from local minima problems and are mathematically guaranteed, if they converge at all, to converge on the globally optimal solution. No matter; other forms of machine learning, whatever their mathematical properties, were quickly abandoned when big data and big compute proved that the neural networks performed better anyway.5

Let’s hear it—not for theory, but for gobs of data from Flickr, or where have you.

NNs, in other words, undercut one of the principled reasons to use machine learning methods based on mathematics in the first place: guarantees of an optimal solution given some solution space. Though clever tricks like dropout have helped mitigate the bugbear of local minima, there is no general theoretical solution to the problem of non-optimal learning in neural networks.

So why does today’s generative AI work so well on so many problems? The answer is, perhaps ironically, Big Data and Big Compute—or the scaling hypothesis. At sufficient scale, a deep neural network’s loss landscape contains many acceptable solutions, and gradient descent reliably finds one of them. Global optimality is no longer required at scale.

But generative AI is sequential, and it cares only about the next token given a sequence. In some domains, the question of truth versus probability may be less germane. But with language models, those “next tokens” are answering questions, holding conversations, and writing your boss an email. Truth is never guaranteed by next-token probability. In other words, scaling may mitigate optimization failures, but even well-optimized language models ignore truth by design. Witness the now notorious “hallucinations” or “confabulations.”

Add to this that ANNs are perfect black boxes, opaque to human inspection. This has gone so far in recent years, with the advent of LLMs, that even experts who design and train these systems admit they don’t really know why they work.

Epistemological foundation for a new age? Hardly.

Black-box AI is a pale substitute for what we once aimed for: machines grounded in our best, most elegant theories of intelligence and cognition. Instead, the pat answer to questions about future performance is simply to add more data, or “scale.” The so-called “scaling hypothesis” has proven inadequate for squeezing more out of these new, energy-hungry black boxes. Prediction without explanation has been and will continue to be inadequate for a serious computational science. Or for AGI, for that matter.

And in the meantime, we’re no closer to understanding anything substantive about intelligence at all.

We don’t know what intelligence is, but we’re acting as if we do

Would Albert Einstein score meaningfully higher on an IQ test than members of MENSA? If not, it’s not clear what the test measures. Was Pablo Picasso “less intelligent” than J. Robert Oppenheimer? Would an IQ test decide? Unlikely.

There is no settled theory of intelligence. Over decades, cognate fields—cognitive science, neuroscience, machine learning—have uncovered partial accounts, but no unified theory of what intelligence is, how it works, or how it should be measured has emerged.

Psychometrics is the study of intelligence through tests—IQ scores, SATs, GREs, and the like. The measures are useful, but contested. After a century of work, the consensus is modest: the tests capture something about intelligence, but not everything.

The ballyhoo over proving how “smart” AI is has intensified the problem. AI researchers rely on benchmarks: curated collections of problems presented as questions, code prompts, or reasoning tasks, paired with answer keys or clear grading rules. The model is run against this fixed suite, and its outputs are graded according to predefined criteria.

Putative benchmark tests have proliferated of late. Consider MMLU (Massive Multitask Language Understanding), GSM8K (grade-school math word problems), and HumanEval (code generation via unit tests). More recent efforts like HELM attempt to aggregate performance across tasks into a single profile. Generic benchmark tests like François Chollet’s ARC (Abstraction and Reasoning Corpus), intended to measure something akin to what psychometrics calls “fluid” g—the ability to infer rules and solve novel problems from minimal examples—are also popular.

These are not trivial tests. But they remain closed-world evaluations, where the test problems are specified in advance, the scoring rules are fixed, and the space of acceptable answers is known.

This matters because success on a benchmark does not establish a general capacity. It establishes competence under particular conditions. A model that scores highly on MMLU has learned the statistical structure of question-answer pairs drawn from its training distribution. A model that performs well on GSM8K can reproduce solutions for a class of problems it has effectively seen before.

A model that passes HumanEval can generate code that satisfies software unit tests, which themselves define what counts as correctness. Even tests for commonsense or “general” intelligence show the telltale pattern of teaching to the test: scores on Chollet’s ARC have risen sharply as researchers optimize against the test, yet the systems are not acquiring commonsense.

The problem is not that benchmarks are useless. The problem is the inference we draw from them. We move from:

“the system performs well on this task”
to
“the system is intelligent in the general sense”

That step is not justified by the evidence.

In psychometrics, we at least proceed under the assumption—contested but operational—that different tests are imperfect measures of a latent general intelligence, often denoted g. In AI, we lack even that. There is no agreed-upon underlying construct that the benchmarks are measuring. Instead, researchers build systems to score highly on accepted benchmarks, and report those scores as indicative of underlying intelligence.

This is why benchmark gains so often fail to translate into robust real-world competence. When the problem changes—when the task is underspecified, when the data distribution shifts, when the criteria for success are not fixed in advance—performance degrades in familiar ways. The system produces answers that are locally plausible but globally meaningless.

Benchmarks are instruments. They measure what they are designed to measure. But they are not theories, and they do not justify claims about general intelligence. And treating them as such is not progress.

This confusion has been with the field from the beginning. If a machine can play chess, recognize speech, translate text, classify images, or solve a set of reasoning tasks, then those capabilities are taken as local stand-ins for intelligence. The move is understandable, but the danger is that the proxy hardens into the concept itself. We stop saying, “This system performs well on a narrow task we associate with intelligence,” and start saying, “This system is intelligent.”

That slippage is not merely linguistic, but unfortunately has shaped the entire public understanding of the field. Researchers at organizations like Google, Meta, OpenAI, and Anthropic are no doubt aware of this tension. But neither they nor their employers have much incentive to clarify it for a public eager for the next breakthrough.

The problem becomes even clearer once we step outside formal testing environments. Human intelligence, or “fluid g,” involves transfer across domains, learning from sparse and ambiguous evidence, and—crucially—the ability to determine what matters in the first place. It integrates perception, memory, action, and social understanding in ways that do not reduce to fixed tasks or scoring rules. None of these fit neatly into a benchmark suite.

Nor is intelligence exhausted by abstract problem-solving. Human beings are embodied creatures acting in a world. That is why the Einstein-Picasso comparison is so revealing. It is not merely that intelligence comes in different forms, though it plainly does. It is that the word itself sits over a heterogeneous field of capacities that resist reduction to a single numerical scale. Scientific genius, artistic genius, strategic genius, social genius, mechanical genius all overlap, but they are not identical. The attempt to compress them into one latent variable may be useful for certain purposes, but it is already an abstraction from the richness of the phenomenon.

AI’s incessant focus on “intelligence”—it’s in the name—ignores a basic fact: we lack a settled theory of intelligence in the first place. What are we testing? We end up defining machine intelligence by whatever machines currently do well. We talk as if building systems that perform intelligently on selected tasks gives us an account of intelligence itself. That is like mistaking an increasingly accurate map for a theory of geography. The map may be useful, but it does not explain the terrain.

A more sober view would begin from the opposite premise: we do not yet know enough. We do not know whether the current dominant methods in AI are converging on the relevant capacities, simulating some of them, or merely bypassing them with powerful statistical shortcuts. Those are very different possibilities, and a science of AI would have slowed the roll long ago. The obvious inference here is that we are not yet capable of such a science, and are settling for a dangerous simulacrum. One consequence is that governments, educators, politicians, the media, and the general public are getting snowed.

So the deepest irony may be this: at precisely the moment when public discourse is most confident that we are building intelligence, the underlying concept remains unsettled. We do not know what intelligence is, and yet we increasingly organize research agendas, investment flows, institutional priorities, and even civilizational rhetoric as though the matter were resolved.

The honest position is more demanding. Intelligence remains, in crucial respects, an open question. Any field that forgets this risks mistaking progress on proxies for understanding of the thing itself.

Erik J. Larson

07:30 – Facts vs Data vs Evidence (main point)

Erik J. Larson

All models converge on the center of the distribution. That takes a lot of power, but not as much thought….

1. The grounding problem isn’t solved. We’ve mostly routed around it.

The grounding problem has been a central issue in cognitive science and philosophy for decades. How do symbols get their meaning? How does a system connect language to the world around it? More trenchantly: how does my thought, expressed as the word “cup,” refer to THAT cup sitting there on my kitchen table?

We don’t have a general solution to this class of problem. Turns out, the philosophers do have something to say, since the AI engineers still have no good solution to this problem, which is why our robots and self-driving cars don’t work.

And philosophers have known and discussed the problem for centuries. The key question, and one that bears directly on the failures of modern AI, is this: how does a token come to refer? How does a system bind an internal symbol—not referring to anything—to an external object, a property, or an event in a way that is stable and usable for further inference, and revisable through future interactions of the system with its environment? We don’t know.

What’s changed is not that we’ve solved the grounding problem, but that we’ve stopped treating it as central. We’ve taken data—large, static corpora of text and images—as an adequate answer, ignoring the fact that “data” is also internal to a cognitive system and hardly a good candidate for grounding anything except in a spreadsheet. This is a colossal lacuna in modern AI thinking and research and somewhat inexcusable given the obvious need for such a capability, and a theory explaining it. Welcome to AI research.

Data is a record of prior human activity. It is already interpreted, already structured, already grounded by someone else. A system trained on that data is learning statistical regularities over representations, not learning how those representations connect to the world. Data, in other words, does not solve the grounding problem—it ignores it.

A language model produce language about physical situations—say, that objects fall, that collisions happen, or that liquids pour. But it does not learn what fixes the reference of those terms. It does not know what makes an instance of “falling” an instance of falling, or which features of a situation are causally relevant versus incidental.

It has no mechanism for what we might call reference stabilization—the capacity to fix what a symbol refers to across changing contexts, and to maintain that reference through perceptions, state transformations, and actions.

The upshot is that we ask an “AI” about a slightly novel physical scenario—an object balanced in an unusual way, a container with a nonstandard opening, or a change in support—and we expect performance to degrade. The system isn’t “confused,” but simply has no underlying model useful for making progress on grounding. Spreadsheets don’t know about the world.

That’s why self-driving cars don’t yet “work” either.

Neuroscience could come to the rescue, if only we understood the brain better. Here, researchers face a quagmire too that should make us cautious about easy analogies between brains and current machine learning systems.

Yes, neuroscience has uncovered important regularities in early sensory processing. We know a good deal about retinotopic organization in vision, orientation-selective neurons in primary visual cortex, hierarchical feature extraction along the ventral stream, and population coding in sensory areas. These are real achievements. They also helped inspire early neural architectures, including convolutional models and other hierarchical systems for pattern recognition.

But none of that gives us a theory of grounding.

At most, it gives us partial constraints on implementation. It tells us something about how biological systems process sensory input at low and mid levels—edges, contours, motion, invariances over position and scale. It does not tell us how a system comes to represent this cup as that enduring object there on the table, how it binds a variable to that object across changing viewpoints, or how it updates its representation when it acts on the world and the world pushes back.

In other words, neuroscience may illuminate parts of the pipeline from sensation to representation, but it does not yet explain how reference is fixed, stabilized, and revised through embodied interaction. It gives us clues. It does not solve the problem.

And if anything, the biological comparison cuts against the dominant engineering strategy.

Humans are low-data learners. Infants acquire a basic understanding of objects, persistence, containment, support, and causality through relatively sparse but richly structured interaction with the environment. They are not ingesting terabytes of text. They are embedded in the world, acting in it, failing in it, and updating on the basis of feedback. Their concepts emerge not from passively absorbing records of prior linguistic behavior but from tightly coupled perception-action loops. That matters.

If intelligence in biological systems depends on embodied, intervention-rich learning, then treating ever-larger datasets as a substitute for experience is not merely incomplete. It may be fundamentally misguided. We are taking the residue of human cognition—texts, images, labels, annotations—and mistaking it for cognition itself.

This is the deeper problem with the current paradigm. It assumes that grounding can be deferred, approximated, or eventually washed out by scale. Train on enough data, and perhaps the problem disappears. But there is no good reason to think that. More of what is ungrounded does not become grounded simply by accumulation. Correlation does not turn into reference by getting bigger.

A system that lacks the ability to intervene in the world, to bind symbols to stable objects through action, and to revise those bindings in light of consequences, is not solving the grounding problem. It is operating upstream of it.

That is why these systems can appear uncannily capable and yet fail in ways that remain structurally familiar. They can generate language about the world without possessing a workable relation to the world. They can mimic the surface forms of understanding while lacking the conditions that would make understanding possible.

So the issue is not just that current AI systems lack grounding.

It is that the field has largely reorganized itself around methods that make grounding easy to ignore. Data gives the appearance of contact with reality because it is full of the traces of human contact with reality. But the contact is inherited, not achieved. The machine receives the representation after the fact. It does not earn it through its own encounter with the world.

Until that changes, the problem remains exactly where philosophers and cognitive scientists said it was: at the point where symbols are supposed to become about something.

We still do not know how that happens. And until we do, talk of machine understanding should be treated with far more caution than the field now permits.

2. Correlation versus causation used to be a slogan. Now we’re building it into our most advanced systems.

Pundits and seemingly everyone else today produce lots of loose talk about “reasoning” in modern AI, but when you press on causation, the story collapses.

Turing Award winner Judea Pearl has done serious work here with his directed acyclical graphs (DAGs), but the approach is still constrained by the types of problems that allow a determinate graph to spell out variables and dependencies in advance. That is already a significant limitation.

Read more

The VP of Sexy/Sweaty.

What follows is how Jake remembers it:

We’re in Steve’s BMW driving up 101 towards The City and he’s talking about this club and how he’s DJing and all the girls and everything’s a rager. A rager. When we get to The City Steve doesn’t seem nervous or excited but he damn near ruins our night pulling out in front of a San Francisco cop. The cop rolls down his window next to us.

“What do you think you’re doing?” he asks, all irritated, glaring at us.

Steve transforms into this obsequious “yes sir, no sir” guy, apologizing about being in a hurry and we’re from Palo Alto. He keeps this up long enough for the cop to drive off, shaking his head.

Steve’s destination is a dance club full of people — gay guys and good looking women — with a big elevated DJing platform. Steve’s up to the platform after a few minutes and I’m at the bar getting a Vodka Redbull. The Vodka Redbull is my official fuck-it drink; it means I’m looking for maximum fun. I know this of course when I order it.

Steve waves me over and introduces me to his friend Frank. Frank’s a perma-grinned, nice looking, smaller dude that right away offers to get me high, which I accept. Generally I won’t unless I’m already drinking, because the booze tamps down the harsh paranoiac effects of marijuana. Generally I decline, but not tonight.

We go outside and Frank passes around his pipe, talking about the street and the area like he’s lived around this club for the last forty years. He looks forty. Or maybe fifty tops. Back inside I wander out onto the dance floor, and then back to the bar between songs for another Vodka Redbull. I see my movements in the club now as a sequence: get high with Frank, then go dance, and then get a drink. Then repeat.

An absolutely Girl-quality, gorgeous woman dances by herself on the dance floor with the DJ music blasting out loud and good. She has this flowing dark hair and a brown dress with a bright shirt. It’s beautiful. She’s beautiful. She’s dancing right out in the middle of the dance floor, and she’s got one of these female bodies perfectly formed, with hip and shoulder ratios, and her face glistens in a healthy organic glow and she looks like the most voluptuous and beautiful creature I’ve ever seen (excepting The Girl maybe, but Girl cogitations right now are strictly prohibited. Strictly prohibited).

I’m eyeballing her, really, but I can’t help it. She’s a tall, brown girl—I mean she’s brown and white or you know, mixed—and she’s moving closer now and dancing on the edges of the floor near where I’m standing, looking over at me. She’s very smiley with her arms up dancing and I really like her attitude. Come to think of it she may be on an assortment of club drugs but at any rate she’s having a great time and very positive. I like positive people, you know, going late into the night and happy to be alive. This is a universal and folks will try to steal away the compliments in all this moralistic talk about being on drugs; I say you’re happy and you’re positive and that’s that.

“I looooove this” she says, when I walk over to her. “That’s two of us” I say. “Let’s dance.”

She holds her hands out and I grab them and we’re circling around. The music is bumping and the lights are flickering. People are standing on the floor with drinks and others are dancing. We’re carving out this big circle holding hands and laughing at each other. Cutting this orbit hand in hand, and with her smile and the quickness of it all, I think we’re, you know, good.

More vodka. She’s laughing and now I’m back into that gooey, dopamine situation where everything seems grand. Dancing. Laughing. A kiss on the cheek. I reconnoiter briefly to fuel up on Vodka Redbull, and leaning on the makeshift plywood bar I look out on the floor and she’s laughing. Back on the floor we can’t orbit anymore because I’m holding the drink, but we keep dancing and talking to each other—all we’re saying is how we love it. We looove it.

Fast forward. I’m getting really high in this club with Steve disappeared up in the DJ platform and this new guy Frank who I keep seeking out for weed. And drunk. I’m getting really drunk, too. Blame this on the sequence I’ve started here, sure. With the weed and the Vodka Redbulls everything seems grand and people’s faces are big and grinning. Very twisted, though. Messed up. I’m messed up. Blame this on Steve. Come to think of it blame all this on The Girl. But as Girl cogitations are strictly prohibited tonight, I say blame this on me. Blame it on me.

Beautiful women move history forward, I think more than men, but when a bloke tries to love that and to enjoy a beautiful woman and to call her his Goddess, folks go all “grow up” on you and why don’t you just get someone your own age that’s more of a mother or I suppose now more of a friend. Well that’s all fine, I say, but these folks haven’t danced with the Goddess. It’s no good saying something that we all can see and frankly, she’s to be adored. This is not about sex. Not wholly, I mean. And that’s an honest statement. She’s damn wonderful.

Fast forward again. I’m back from the bar this time around and I decide the Goddess also has an operational and widely recognized title. She’s in charge of everything. She’s so beautiful. She’s the Vice President of Sexy and she also has this title “VP of Sexy/Sweaty.” She drops the sweaty when in San Francisco clubs... I don’t even know if this is true... anyway VP is here, very authoritative.

I keep saying to myself “there’s the VP” and maybe it’s slightly creepy as I’m staring directly at her. She’s an amazing, wondrous image gliding from a kind of focal point at her hips—her belly showing suggestively—and long flowing black hair with... goddamn where is Frank because the sequence here is not to stare at the VP I’m gonna get in trouble. Come to think of it, I’m really, really high so I’ll change or slightly alter the sequence moving forward. Looping, iterating, is very important in software development and it accounts for much of what makes the modern world tick. I’m on solid scientific ground here, I figure. Let’s rehearse all this.

The accepted sequence since entering the club with Steve, Driver-of-Beamer (plus purveyor of fine wines), has been dance, then a Vodka Redbull from the bar. When this is half done there’s a general shout out to “Frank!” followed by a “Let’s get high!”, which is closely followed by an exit to the street where we circle on Frank’s pipe and chit chat to the sweet smell of his weed—we’re in love with her, it’s Mary, as the song...—so the sequence should be changed now because I keep staring at the VP all creepy like, with phantom-high eyes and maybe my admiration is too obvious or in danger of getting misinterpreted. So now: dance—don’t stare at VP too much—a Vodka Redbull from the bar. No more: “Frank!” Then: “Let’s go get high!” Cut that part out for now.

But I’m wrong about the VP and the creep-factor because returning Vodka Redbull in hand she’s still dancing and I swear she’s motioning me toward her. I feel like an Odysseus character, like some Goddess with a weighty title, too, is motioning to me and here I am, very happy. That’s all, just “happy.

It’s admittedly not really Greek heroic tale material; it’s a dopamine tale set in modern San Francisco and there’s nothing to save, or to kill, or to endure, but... shit I forgot the VP is still motioning and why was I thinking about Greek societies, heroic societies, where the actions aren’t based on abstractions like “rights” but on shared notions of responsibility and there’s clear... goddamn it Jake...

The VP has a curious protocol. It’s inspired, no doubt, by her own awareness of her preternatural beauty. First I say that she clearly loves to dance and clearly loves to move in perfect metronome rhythms. She smiles and her eyes glisten and her body makes a or transmits a—it’s like everyone else is dancing around her, like a Maypole at Marymount feel, you know Hawthorne. Nathaniel Hawthorne. All the little supporting-cast dancers stay concealed in the shadows as the VP is orbited around; celebrated. Her protocol is simple, really, and I’ve cracked it.

She only dances with you if you don’t seem too needy about her.

So I dance with Frank. This works, sort of. But Frank is too happy to be dancing with me because he’s gay. And high. Finally the VP summons again, which is a barely perceptible smile towards me, but it gets translated into this silly office speak in my head, like: “Hi Jake, when you get a chance, could you come over for a minute? I’ll need you to dance with me for at least this song—currently playing—and I want to discuss you handling the next one or two songs, as I’ve been impressed by your sequence lately. Thanks.”

Jesus I’m high. I’ll need to nix the Vodka Redbulls from the sequence soon too. VP called me up from the minors and I’m a tad flat footed with the half dozen Vodka Redbulls and maybe seventy five percent of the weed Frank has on him.

Many are called, but few are chosen.

Continue. At some point I just wander out of the club, still in full tilt, to go get something to eat. I figure the Veep sent me off to get fueled up. I walk looking at my GPS on my Blackberry maybe a mile to some all-night diner.

Follows a depressingly standard, near-shameful grub experience, really, close to a dog eating out of a bowl. And I’m happy to pay and wag my tail at the waitress and depart. Outside I text Steve and he replies back a few minutes later as I’m still waiting outside the diner. Thank God, I suppose. He texts me the address to GPS—I had neglected that—and then the password “moonbeam.” So I hail a taxi and I’m off at 3:30am in The City to the rager. Password moonbeam.

When we arrive I think Steve must be full of it; the street is dark and deserted and industrial looking. The only evidence of a rager is an occasional young looking club-scene kid disappearing into this dark three story building. A few people mill about outside.

Gradually I realize I must be at a different entrance; it’s that or I’m a bit disoriented in a way that would almost qualify as concerning. As I approach the building there’s a guy standing outside and people are approaching him before entering so I walk up to him and say “moonbeam.” He nods and I walk in, following the kids in front of me up the stairs, all in this hushed kind of silence, and down a large hall towards some double doors. In spite of my intoxicated fatigue I’m a bit worried I’ve somehow moonbeamed into the wrong rager, if that were possible.

When I open the double doors it’s like that scene in Men in Black, where there’s a million people in the room but it’s all disguised and silent from the outside. The only thing I hear approaching the doors is a dull rumble from the music, very muffled, and when I open the doors there’s a huge stage and massive amplifiers and disco lights everywhere. People are dancing and laughing and standing in groups talking. There’s a balcony with big plush couches and people are up there sitting on the couches drinks in hand, their faces and the tables in front of them illuminated by the lights pulsating from the stage. It’s now I realize they’ve opened up even a larger dopamine space, it’s like two warehouses really and first we were all crammed in one. Now I recognize the first dance floor and turning up my head like a scence in a movie where the character sees the promised land, there’s another in back of it.

Steve’s in here and it’s a rager.

Steve needs to be in here because the afterhours bar I discover is cash only and I don’t have any cash on me. I text him announcing that I’m here and after what seems a very long time he appears in front of me, sharp dressed and grinning.

“No cash”, I say. He scoops out some crumpled twenties and hands them over. He’s been up in the plush couch seats with some girls and he’s leaving soon with them.
“It’s a rager,” he says.

He stands there all Palo Alto with his blond, good looking head, outlining this plan for us to go with these girls but he’s got to cinch things up, he says, jerking his head back toward the couches. I tell him I’m fine here for now and not to worry. Me, I’m coming down from everything and in a hurry to get a drink.

I’m astonished and excited in a visceral way that kills the dull diner omelet thing that VP—Veep now—is still dancing, somehow, a kind of super endurance Goddess in the underground, a real feminine power I think. Really.

Afrojack’s “Take Over Control” bumps grandly, and I feel like the song’s playing just for the Veep. The Veep seems to be fond of glancing in his direction, a small smile playing on her lips. I’m taking another swig of my drink, feeling the Vodka Redbull kick in harder. I’m getting pulled again into this critical night.

I’m not sure if it’s the weed or the booze or just the electricity of the place, but when she motions toward me, it feels like an invitation I can’t refuse. My mind buzzes with dopamine. Odysseus being called by a goddess. Many are called, but few are chosen.

I’m orbiting free of the diner’s fare now, and feel the real sustenance of the copious flowing vodka Redbull in a plastic cup. Things are looking up. I move toward her on the dance floor, a lone meteor plunging unstoppably toward the Sun. She sees me coming and immediately throws her arms up in the air and starts dancing in a groove that turns her in a full circle. This means that halfway into her revolution I’m looking at her ass. And I am. The Veep thinks this is the right response and seems to get even more into her sensual groove. She backs up until she touches me and I have no choice but to place my hands on her hips as we are now so close that the omission would be, well, rude. By all accounts, I think, Veep is digging it.

She turns around and for a second I think she intends to kiss me, but instead she leans in and says in big groovy tones that she loves it. It’s too loud for real conversation so I lean in and say the first thing I can think of, which is at this point a fait accompli, “I love you!” At this she puts her hands palm first flat on my chest and leans in to ask me to stay for another song. I nod and am pleased when Steve drops the retro classic “Groove is in the Heart,” by Dee-lite.

It is. Groove is in the heart, I think.

She turns her back to me again, relaxing her body into the rhythm, and I move with her. My hands stay at my sides this time, a plausibly prudent move to fend off a creep vibe, but I know it’s probably unnecessary. Anyway, the Veep is calling the shots.

Her hair brushes against my arm, and not being able to help myself I lean into her ear and say I want to smell her arm pit. It seems like there could be some groove here. In the arm pit. Veep loves this idea (yes—promotion coming!) and the scent of her deodorant and sweat transmute me into a Veep dimension I can only describe as “she’s the only living thing that matters.”

I’m not a great dancer, but my current assignment from the Veep isn’t to win a Dancing with the Stars title but to contribute meaningfully to her groove. This I am doing.

Halfway through “Feel so Close,” the lubricious tattoo on her hip—a winding pattern of vines and thorns around some kind of mermaid or goddess—catches my eye again as she dances. It’s mesmerizing, like it has its own mind or has been empowered by the Veep to give instructions to me.

She has a smaller tattoo on her belly, and I keep trying to catch a glimpse of it without freaking her out—though in truth, it seems unlikely that anything would freak out the Veep at that point in the night. Veep is no nonsense and good to her employees, as she pulls her miniskirt down to show me. It is the word “wild” wrapped in a rose. Oh. My. God, please don’t let me let the Veep down, I think.

An hour later and I’ve drifted off to the safety of the bar and I’m hitting my limit; I can’t drink much more and there’s this little tired spot somewhere inside me that’s gradually growing and will by degrees crowd out all the dopamine girl-booze feelings. It’s futile to fight it and I don’t want to let the night fade away into dullness, so I’m thinking about telling her I have to go.

Plus Olivia’s been texting me — Olivia from The Camp — sending now four messages to me through the night, asking me if I’m okay. Am I okay. Am I okay. I respond to her once back at the club saying “I’m fine, in SF” and then there’s a bunch of “huh?!!” and “watcha’ doin’?”’ like she’s somehow both worried about me and jealous I’m having a good time. Anyway I can’t, I don’t know, report back all night so I’ve ignored the other ones.

Steve’s incommunicado. Go figure. I text him to see if he’s still around and there’s no response. Veep is somehow defying physics and is back on the floor after a respite over on the lounge chairs. I wave and then come over to her. She’s proud of me I think, a good employee, and I decide I’d better go back to Palo Alto and how charming she is and how much I appreciate her.

“Come visit me in The City!” she’s saying. She’s all excited, like we’ll make plans to rendezvous at a rager again, picking it right up like this—at 6 am, another all night rager.

“I will, I will. I’d love to,” I say.

We hug. She leans in and kisses me on the cheek and then because she lingers and smells like perfume and sex and sweat I move my scruffy cheek across hers and we kiss on the lips. I’m hugging her and she’s firm and young and she smells good. She’s warm.

She leans in, yelling over the bump of the bass, “What’s your name?”

“Jake,” I yell back. As the song wraps up, she leans in close, her breath warm against my ear. “Mia!” she says, and then gets so close to my ear that she can whisper, “Come find me tomorrow.”

I nod, heart pounding. “Mia, how do I find you?” She grabs my hand and leads me off the floor to one of the makeshift bars, where she smiles at the bartender and asks for a pen. Fortuitously, he has one. While I look on in amusement—who doesn’t have a phone?—Mia writes her number down on the back of a receipt she plucks out of her pocket. When she pulls the receipt out of her tight front pocket, her slim phone pokes up. She has a phone. The Veep does what she does, I conclude.

God I’m lonely. I have to go.

To Be Continued…

Erik J. Larson

The Nut House

Up to the inimitable Nut House on California “Cal” Ave, the dive bar discovered by Facebook geeks making the small hike down from their world headquarters over on Page Mill Road. The Nut House is this dark, old style bar with big gaudy neon lights on the walls and a bunch of cranky thirty year old posters and plaques like the “How about a big cup of shut the fuck up?” and “If you’re drinking to forget please pay first”.

Subscribe now

It’s called the Nut House or officially Antonio’s Nut House because they serve peanuts in the shell out of this goofy looking plastic gorilla dispenser in the back, and you eat the peanuts and chuck the shells on the floor. So there are heaps of shells here and there between the bar stools and all over the place.

He knows all of the bartenders at the Nut House and most of the locals. It’s hard to get kicked out of the Nut House but easy to get crabbed at by the female bartenders, which Jake thinks is kind of encouraged by the owners — the bad attitude that is — as the old half-impotent drunks that pepper the front stools and the cocky Facebook kids on Fridays and Saturdays all feel like they’re getting a vintage local bar experience.

They’re getting some personality, not just a drink. And the bartenders will take care of you too, once they know you. They’re not all bitchy, either. This one working today isn’t.

“Hey!” she says, though strangely Jake can’t remember her name anymore. “It’s been a while.”

“Yeah, no kidding. But I’m here now,” Jake smiles. “How about a PBR?”

Sitting at the wrap-around bar, Jake spotted Ivan, a Ph.D. candidate from a southern university—Kentucky, Jake thought. He met Ivan last summer when the kid gave a talk at Stanford on complexity theory. Jake had studied it, but Ivan really knew it. Ivan was talking to a guy in a Facebook hoodie about consistency checking in first-order logic. Jake knew the answer: it was semi-decidable. The kid didn’t.

Facebook Kid: “You know, consistency checking in first-order logic is just NP-hard, right? You can still run some algorithm to figure it out. It’s tough but doable.”

Ivan: “No, that’s wrong. Consistency checking isn’t just NP-hard. It’s undecidable. There’s no algorithm that can determine whether a set of first-order sentences is consistent in all cases.”

Facebook Kid: “But NP-hard problems are difficult too, right? You can solve them with enough computational power, or there are heuristics to avoid the worst case.”

Ivan: “Right, but the whole problem with a consistency check is that the worst case is the case you’re considering. If you’re in a first-order system and have a universal quantifier, you have the possibility of infinite members of a set. What you’re talking about with NP-Hard is propositional calculus. In FOL, it isn’t just hard—it’s impossible to decide in general. That’s what undecidability means. You can’t guarantee an algorithm to check consistency for any arbitrary set of first-order formulas.”

Facebook Kid: “So you’re telling me it’s not just ‘hard’ computationally—it’s outright impossible?”

Ivan: “Exactly. It’s not in NP or any other solvable complexity class. There’s no process that guarantees an answer for consistency checking in first-order logic. It’s a theoretical limit.”

All the while, Facebook Kid grew more insistent. But Ivan’s mastery was getting to Facebook Kid. Jake stood up and walked over just out of their sight, watching with satisfaction. Jake had met Ivan after his talk and joined him for beers at The Nut House. That’s when Jake learned Ivan was not just wicked smart—Ivan was a serious drinker. Facebook Kid was toast.

Finally, Facebook Kid threw his hands up. “I’m from Facebook, dude, I know what I’m talking about!”

Jake couldn’t hold back, laughing loud enough to turn heads. Ivan turned, grinning when he saw him. “I’m from Facebook, man! It’s computable!” Ivan mocked, wrapping an arm around Jake’s shoulders.

They doubled over laughing. The Facebook Kid stood there, arms crossed, glaring, completely lost on why it was so funny.

“Good to see you, Ivan,” Jake said, catching his breath.

“Always an adventure,” Ivan replied, nodding at Facebook Kid, who looked ready to blow. “These guys think they’re God’s gift to computers.”

Jake shook his head, still grinning. “They make it fun, though.” Jake looked around, the eclectic, crazy vibe of the Nut House. Peanut shells, old friends, and debates that went nowhere but gave you the night away from the interminable troubles waiting just outside the door. It was good to lose himself in it, even for the night.

Kevin’s in the Nut House. He’s a good looking guy, maybe forty five, lives with his mother. He swears he made a million in landscaping and lost it all. Kevin’s a California guy that works outside and has this long curled brown hair and looks ten years younger. He has a big infectious smile and even when he’s soused and high he usually can manage a grin. He grins like a shit eating devil.

Jake once saw Kevin work a young college-aged Asian girl… cute as a bug, half the night Kevin and the girl were close-talking and flirting. Kevin was so high before she showed up Jake thought Kevin might glow. Jake smiled. Kevin believes he made a million in landscaping. The guy lives with his mom, but he has charm.

“What’s up Kevin?”

“Nada. Where’ve you been?”

“Around.”

Now a guy walks up between them and leans over to complain about his drink order. The bartender smiles and says she’ll be over. The guy stands there blocking them for a second, oblivious. The guy is poking at his cell phone. Finally Jake turkey necks around the guy and winks at Kevin, saying loudly “Oh, there you are...” Asshole Guy doesn’t even look at Jake, but Asshole Guy walks around to the other side of Kevin and goes back to his phone. The waitress hands Asshole Guy two drinks and Asshole Guy leaves and Kevin says the guy’s a douchebag and that the bartender has been rolling her eyes when Asshole Guy orders. He finally remembers her name, it’s Laura. She’s generally nice. Her boyfriend is a gun enthusiast and he thinks she moved out here from Washington or Oregon.

He looks back at the douchebag guy. Asshole Guy is sitting there poking a finger at someone else, all intense. Jake is straining to hear the conversation but all Jake can make out is numbers: “that’s thirty percent over two years”... “It’s up twenty percent last quarter”....

The only good looking women in the place are over with his group; they’re in from the office for an early happy hour, or maybe celebrating something.

Jake turns to Kevin.

“He’s not a douchebag, technically,” Jake offers. “You see, with derogatory slang everyone takes an insult like it’s interchangeable. But actually a douchebag means something specific, and it doesn’t apply to this guy. He’s an asshole. Plain and simple.”

Kevin pulls on his drink. Grins his devil grin. “Oh here we go.”

Yes, here they go. Jake has nothing better to do at the moment.

Jake is drinking to forget.

“Douchebag is, like, Bill Lumberg from Office Space. He’s not like that. You could also call that type of guy a tool. The tool is Guy Pearce’s character from L.A. Confidential. Remember that movie?” Jake asks.

Kevin grins.

“He could be a tool, yeah,” Jake concludes.

“Where’ve you been?” Kevin asks again. Pulls on his drink. Kevin can’t remember what he drinks. Usually a vodka cranberry, Jake thinks.

“Look,” Jake says, ignoring Kevin’s question. “I wrote a blog post about this once.”

Jake digs out his Blackberry and searches around until he finds the post about name calling slang.

“Jerks. Low class, mean spirited, don’t give a damn types.

Chet from Weird Science. Chet was a jerk.”

“Assholes. An asshole is a jerk that has found a little success. Sometimes a lot. Like Colonel Nathan R. Jessup from A Few Good Men. He’s an asshole. This numbers-spouting, finger pointing, piss-off-the-bartender with his attitude guy is an asshole, minimum.”

Kevin shakes his head. Grins.

“He could be a prick, too,” Jake adds. “A prick is just an asshole with an innate sense of entitlement. Like, the rich kid from Some Kind of Wonderful, remember that movie? Hardy Jenns. Hardy Jenns was a prick.”

Kevin is devil grinning, stirring his drink. “You’re a douchebag,” he says affectionately. Then he adds: “Where’ve you been?”

Goddamn Kevin. He made a million in landscaping, and lost it all.

Kevin goes outside to smoke some weed and then later he leaves to go eat and Jake is sitting there drinking and eating peanuts and talking to Laura.

Later Steve comes in but Jake doesn’t notice him at first because Jake is in some weird discussion about Oliver Wendell Holmes Jr. and The Common Law with a law student at Berkeley and her friend. Later Jake’s giving an impassioned defense of cultural elitism, basically an exposition of T.S. Eliot’s ideas, working off of his hardly-read Notes Toward a Definition of Culture.

This is an impossible task generally in Palo Alto or anywhere else as the concept that healthy cultures are hierarchical (and thus support and nurture an elite class) gets zero sympathy in modern academia, with its fearful shut-you-up political correctness masquerading as the new freedom—people think political correctness is somehow good and progressive when it’s really corrosive and retrograde.

So Jake is doing his damnable best to give a passable exposition of Eliot (who gets a modicum of respect because people have generally a respect for his poetry—Hollow Men and Wasteland and all), and Steve walks in. Steve is all hunched forward, gesticulating, and Steve walks in.

Steve is vintage Palo Alto: divorced, good looking, thirty three, BMW. He works for some wine shop and loves to sniff out raucous good times in The City. He doesn’t know Steve very well—Jake has talked to him once or twice before—but Steve is snooping in on Jake’s defense of Elliot’s ideas which is really a suggestion that they’re screwing up culture with a bunch of muddled Marxist libero-egalitarian, and, look everyone is treated equally under the law, that’s LAW.

What Elliot is saying is that culture is not law, in totality. And the parts of culture that folks care about most typically fall outside the scope of the law. This gets back to Jake’s discussion about Holmes and the common law and generally American pragmatism, but Jake’s not going into all of that. Anyway Steve comes over and he sort of stands there with a big mischievous smile with his blond hair and his nicely dressed look and he says:

“I’m going to The City. DJing a party. It’ll be a rager,” Steve says.

“Wanna go?”

“Now?” Jake asks. He’s still got his hand out in this professorial manner and Jake was making his kill shot point. The guy listening is crinkle-browing Jake now but his girlfriend is smiling and her eyes have gotten a bit larger since he started the project.

“Right now?” Jake asks again.

“Yeah, leaving now” Steve says, all decisively like, you know, the ship is leaving.

“Sure why not?” Jake says, leaning back and laughing. Jake looks at the guy and his girlfriend and the guy shrugs and smiles.

“Go to it” the guy says.

He’s a good guy. The girl blurts out a “yeah, well to be continued!” nicety, a warm statement. Like a—and maybe Jake’s imagining this—an opening to find him again somewhere, a venture outside the scope of this Elliot lecture he’s been delivering. Not sure. Jake smiles and downs the rest of his beer.

“Yeah, to be continued,” Jake says.

“Have a good night,” Jake says.

He’s satisfied that Jake made all his points without enduring any predictable responses or muddle-headed objections, or anything at all except his crinkled brow and her big eyes. He already knows his questions as he’s done this a thousand times. He’s taught philosophy and described it in bars et cetera. The students are always legal-heavy and he always has to pull this apart from culture so the point can come out correctly.

Now, about this rager, in The City, with good-looking Steve. To this we turn next.

Erik J. Larson

Rob Klein

Inside the corporate walls of The Cheesecake Factory, Jake ordered a martini and cheese fries he’d never touch. The scent of over-seasoned, over-buttered dishes clung to the air. He scanned the bar—it was like stepping into a twisted Stepford CEO mixer. Everyone had the same designer sneakers, minimalist glasses, and that smug air of I-don’t-need-you; I-need-your-capital.

“California Girls,” by Katy Perry was playing in the background. Little trick, Jake thought, feeling like he should apologize to someone for his tastelessness. Perry was talented, and she had a good story. She’d gotten her start singing in the church choir in Texas, he knew. Who could fault her, she was talented, and you had to be hard working to make it in that business. Jake knew a gal he’d worked with years ago who went to the same church and knew Katy Perry. She said no one would have thought she’d be a superstar. Just a nice pretty girl with a decent voice. We’ll lick your popsicle, Jake chuckled. Isn’t that the world. Careers were made and ruined by lots of factors, but Fortuna was one of them, no doubt.

Jake ordered a double Scotch and sat, mulling over his conversation with Mickey at The Tap Room. There was a gap between them, one they both understood but never acknowledged. Mickey had studied at the London School of Economics—he wasn’t a lightweight—but he wasn’t technical like Jake. And while it was never spoken outright, that distinction mattered in this town. Mickey’s game was copying design patterns and features that were trending—badges, gamifying online experiences, pushing multimedia content. His approach was a who’s who of what everyone else was doing—or trying to do. Jake had once, delicately, pointed this out. In response, Mickey explained his “secret sauce,” but it was too abstract, too speculative, to spark any real clarity. Even now, with a double Scotch in hand at the Cheesecake Factory, Jake couldn’t quite make sense of what Mickey was aiming for.

The irony, though, was that in Palo Alto, Mickey might just be the ultima hombre—the last man standing when it came to investment. That’s how the Valley worked. Mickey wasn’t a Ph.D. geek like Jake, but he had a high social IQ, and he was relentless in pursuing opportunities. He could outlast anyone, schmoozing with investors Jake found amusing but ultimately didn’t trust. That was the game. And if it was chewing Jake up, Mickey seemed to glide right through it, unaffected

Jake had risen through the ranks in research and development, using his Ph.D. dissertation as a foundation to build Nexus, a semantic search platform designed to understand meaning rather than just keywords. Nexus wasn’t about simply finding the right word but grasping the context—like distinguishing between a financial institution and the side of a river, based on how the term “bank” was used. To achieve this, Jake had incorporated machine learning algorithms like Hidden Markov Models and Maximum Entropy classifiers, which allowed the system to infer meaning. But it wasn’t until he explored Markov Logic Networks, championed by a young professor at the University of Washington, Pedro Domingos, that everything clicked. MLNs gave Nexus the ability to combine probabilities with logical inference, creating a hybrid system capable of handling ambiguity with precision.

The key idea was to associate a probabilistic weight to logical statements that are often expressed as absolutes but, in the real world, might have exceptions—like ‘If something is a bird, then it can fly.’ This statement is mostly true, but there are exceptions, such as penguins or ostriches. The Markov Logic Network (MLN) formalism that Jake and his team had extended captured these outliers, allowing Nexus to handle uncertainty and ambiguity, making it a powerful tool for understanding meaning rather than just matching keywords.

This breakthrough transformed Nexus from a smart search engine into something that felt almost intuitive. After years of testing algorithms in their parallel pipeline architecture, Jake felt a surge of pride each time the system made connections in subtly different contexts. Nexus was learning to think in shades of meaning, just as Jake had envisioned in his early research days.

Jake couldn’t help but smile, reflecting on his dream with Nexus. He was fond of telling investors and the occasional media interview that Nexus was humanizing machines.

Now, he was teetering on the edge of losing it all.

Seemingly on cue, Rob Klein sauntered into the bar. “Fuck me,” Jake cursed as he spotted Klein’s shark grin plastered on his face, his bespoke Charvet shirt announcing his high status. Klein had been chosen by Verity’s investors to sit on Verity’s board. Jake didn’t trust him nor, for that matter, like him. As far as Jake was concerned, Klein was everything wrong with the Valley. He was everything wrong with the world. Jake avoided him and when he heard rumors Klein was gunning for him, he resolved never to be in the same room if it could be avoided. When they did meet, they did the usual bullshit dance—Rob with his cheery veiled threats, pretending to do Jake a favor by explaining “concerns.” Jake with his nods of understanding and speeches about progress. But Jake knew Klein saw Jake as a threat and wanted him out. and Jake had given him plenty of reasons lately. Jake wouldn’t be the first to be kicked out of his own company at any rate. In this town, you were either at the top or a footnote.

Jake forced a grin, meeting Klein’s gaze. “Rob. Didn’t expect you here.”

Klein slid onto a stool, exuding his usual slick confidence and smelling of scented soap. “You’re a tough guy to find. Figured you’d pop up sooner or later. Tough times, huh?”

Jake kept his face blank. “Nothing I can’t handle.”

“Good to hear.” Rob leaned in, voice dropping to a low, conspiratorial tone. “Because, let’s face it, you’re in deep. Investors are spooked. We should talk it through.”

Jake felt anger rising, sharp and quick like a blade. “Not right now, Rob. But soon.”

Rob shrugged, unfazed. “You might benefit. Desperate times and all that. Just give it some thought. You know where to find me. We can noodle on it.”

Jake forced another smile, but as soon as Rob turned his back, it vanished. He watched him slink off, that smug grin lingering. The noise of the restaurant faded as Jake stared at his drink. This was classic Klein—a vulture circling, waiting for his moment. Jake knew the board was restless, and Klein was playing both sides—whispering in Jake’s ear while plotting his demise behind his back. It’s Shakespearean, he thought, but it’s not like shallow pricks like Klein would ever know.

Jake ran a hand through his hair, exhaling slowly, trying to tamp down the anxiety gnawing at him. How had it come to this? His mind drifted back to the early days of Verity again, when DARPA contracts had gotten him started. They were prestigious, but the goal was private money—venture capital.

Then IBM came knocking. Jake had used IBM’s interest to go to the VCs with leverage. He could still see the day the IBM execs visited—blue shirts, corporate as hell, walking into Verity’s laid-back space. His team, all shorts and sandals, focused on building, not dressing up. Jake made the pitch, showing how Verity’s platform could revolutionize their operations. IBM was impressed.

But then, Jake’s life started falling apart—the move to San Francisco, his mess with Iva, the bar crawls. He began skipping the office, leaving the IBM deal to his top programmer, Stephen, and his executive assistant, Silva. They were good, but they weren’t him. They couldn’t close. They kept calling Jake, needing signatures. Half the time, his phone was off. The other half, he sent vague replies—“I’ll come by soon,” “We’re on track.”

That pushed IBM’s patience too far. Their lead on the project—some corporate tool with a moustache—lost confidence. The tone of his emails shifted, becoming clipped. Then it was over. IBM pulled out, and the board learned about Jake’s absenteeism.

That’s when Klein moved in. Rob Klein—always ready, always waiting. The board was already frustrated, and Klein knew how to play the game. He rallied them, pushing them to turn on Jake, swooping in like a shark sniffing blood. Klein had a simple plan, and in truth it was a good one given the circumstances: feed Jake to them, and they’ll return to the table to invest. So, that’s what was happening. Jake knew in rehab that he would get out too late. The irony of him getting “better” was that he was losing Verity—his source of income and his life’s work—while doing yoga and sniffing lavender and talking about his past.

Jake’s grip tightened on the glass. The worst part wasn’t losing Verity—it was knowing he’d handed it to Klein. The IBM deal had been his to win, and he’d blown it. Now Klein was circling, waiting for the final push to knock him out. Jake had heard from his lawyer that Sequoia Capital had pulled their 10-million-dollar term sheet with Verity over the IBM disaster. Now the word was that it was back on the table. Jake wasn’t stupid—it was back on the table because Sequoia had assurances that Jake wouldn’t be CEO anymore.

Jake took a deep breath trying to wish his anger to subside. He’d been caught off guard. But at least, here and now, he hadn’t let Klein see him rattled. That was something.

Jake pushed away from the bar. Klein could play his game, but Jake wasn’t biting anymore.

He downed the rest of his drink, waived over the bartender to pay his tab, and left the Cheesecake Factory behind.

Outside, Jake wandered down the streets of Palo Alto, intent on walking off the stink of Rob Klein and finding another spot. Then he saw it—a neon sign above a jazz club, The Blue Note.

He hesitated, then ducked inside, the warm air inviting him. The sound of Coltrane burst through the room. He ordered a bourbon and sat at a small table near the bar. The music be-bopped, raw and unpredictable. This wasn’t artificial intelligence—this wasn’t neat or calculated. This was messy. Real. Music, he thought, was handled by the right hemisphere of the brain. Language by the left. He’d studied neuroscience and read The Master and the Emissary by Iain McGilchrist. Jake knew he was a right-brained person in a left-brain world, which he called “Machineland.” The left brain liked control and precision. The right brain saw the whole rather than the parts and appreciated context over logic. The left brain was ruining the enchantment of the world.

He sat in the jazz club, the band playing “Blue Train,” and his mind turned to science again, to neurons firing like notes on the saxophone.

A tap on his shoulder startled him. He blinked and turned.

Iva.

Her wide eyes caught the dim light, making them seem even more perceptive. Her familiar scent somehow reached him. “Jake?” Her voice was too quiet, barely audible over the saxophone. “What are you doing here? Are you okay?”

His mind scrambled. “Iva… didn’t think I’d run into you here. I thought you were at Lake Tahoe,” he blurted.

“I was this morning. It’s… Jake, let’s not…”

Her gaze lingered on him, searching. She slid onto the stool across the table, close but not too close. “There’s a jazz night. Live performance. I came with friends. I didn’t think you’d be here.” Jake liked jazz but Iva’s point was valid. The chance of bumping into him should have been very low.

Jake blinked, remembering too late. Jazz. Iva had always been into it, and of course she’d be here tonight. He should’ve known.

Still, he found it curious she’d been at Lake Tahoe that morning. If it wasn’t true, why say it? Iva could be stubborn, but she wasn’t a liar. He decided not to press it. She would say what she wanted.

He cleared his throat, the bourbon in his hand feeling heavier. “Yeah, I’m okay. Obviously, rehab didn’t stick, but it’s not your fault.”

“Jake, if you drink because we’re in a fight, how are you ever going to get sober? You’re not ready yet,” she said in that way that always pissed him off.

“Sometimes you just have to follow the music,” she added cryptically, like she was importing a pearl of wisdom. He felt like he’d crossed a point of no return with her. They wouldn’t make it together. And it wasn’t all bad.

They sat there, gazing at each other across the table, the music filling the silence, Coltrane’s wails curling through the air. Jake wasn’t ready for this conversation. He wasn’t ready to see her. The bourbon was already thick in his system, and he knew she could smell it. All that cash spent on rehab, and here I am. A dose of self-loathing settled over him like sweat.

The Girl, bless her heart, beat him to it. Iva shifted, her eyes sweeping over him, unreadable. “Jake,” she said quietly, her voice almost drowned out by the music. “You’re not ready for this. Call me when you’re back on the ground, okay?”

“You mentioned that, yeah,” he replied with sarcasm. “And yes, I will.”

He dropped his eyes to the bourbon in his hand. He wanted to say something more, but the words stuck. Isn’t it typical, he thought. His beautiful girlfriend, always playing Nostradamus. He hated thinking badly about her. Nothing is working, he thought. An old drunk at The Nut House flashed in his mind with his wickedly apt phrase: it is what it is.

He winced. She always had a way of making him feel like a guilty puppy that peed on the floor. She was always full of disappointment.

It wasn’t that she didn’t have reasons to feel that way—he’d given her plenty. But it was hard to separate her expectations from his guilt. She’d twist things, wrapping his failures around her disappointments. For almost a year, he’d danced to her tune, too trapped in his emotional limitations to realize the difference.

“Iva,” he started, but she was already standing. She blew him a kiss, turned, and walked back across the floor. Jake stayed to have more drinks, unsure of what to do next. A few minutes later, she left with her friends—one Jake recognized, the other a man who he didn’t. None of them including Iva looked in his direction. The door swung shut behind her, leaving him alone with the music. At least he didn’t need an excuse to leave now.

He rose and stood for a moment, catching the last notes of “My Favorite Things.” At least it wasn’t “A Love Supreme,” he thought.

He didn’t want to think about Iva, but she always wormed her way into his thoughts—left brain or right brain, she was there. They’d met on Mamba, a Russian dating site, and from the moment he saw her profile picture, he was hooked. Her long brown hair swept back in the ocean wind, her eyes holding that quiet, piercing look that felt like it could strip him bare. She had photos of her on a yacht. He’d wondered if she was rich, but realized she could get an invitation to someone else’s yacht. It didn’t matter with someone who looked like that.

She was turning twenty-seven soon, twenty six when they met last autumn. She wasn’t just stunning but mentally sharp, too. She was an amateur photographer, loved to read fiction, and was obsessed with rare teas. She knew more about tea than anyone he’d ever met. She liked to travel. A Ukrainian, she spent much of her childhood in the Czech Republic. Her English was perfect, with just enough of that accent that drove him wild. Every time she spoke, he felt a contradictory mix of comfort and excitement.

Through Mamba he’d asked her out, and she’d accepted. They met for coffee in San Francisco. She suggested they split a bottle of wine, but Jake wasn’t drinking at the time. When he declined, she’d tilted her head, curious. “Why?” she asked.

“I’ve had a problem with it,” he admitted.

“Just tell me you’re not in AA,” she said with a sly smile. He lied, saying he wasn’t, even though he loathed the meetings but respected the occasional gem from someone sharing their experiences. Driving home from the cafe, he couldn’t stop thinking about her.

They started dating, and on his 36th birthday, they met at Fisherman’s Wharf. Later, they ended up at a Turkish place she’d picked out, drinking strong tea and munching kebabs. She wouldn’t sleep with him, but when they kissed, he was over the moon. Best birthday he’d ever had. The taste of that tea still lingered in his memory, rich and spiced. It reminded him of her.

He wondered how it had unraveled so quickly. It was like he was staring at a complex equation, able to glimpse the answer, but it remained maddeningly out of reach. That was Iva—a beautiful, intricate mystery.

Time to go.

As he was leaving The Blue Note, a member of the band, the trumpet player—a white guy with a trimmed beard and a Fedora hat—called out his name. “Jake.” He froze. Turned to face the band. “Jake,” the man said again, “Thanks for stopping in. It gets better, you know?” Jake stood by the door, eyes wide, realizing he had no explanation for this. He glared at the trumpet player, but the guy was already focused on the singer introducing the next song. The only explanation that came to mind was Iva setting it up, but that seemed as likely as pigs flying. It unnerved him.

Outside, the cool evening air hit his face, sobering him a bit. He decided he’d forget about The Girl. He was good at partying. By all accounts, he was the life of the party. He was good on a bender, and that’s what he planned next.

Erik J. Larson

Iva in San Francisco

Jake and Iva had been crammed into that tiny Russian Hill apartment—a place just big enough for two. The walls were thin, the paint peeling, and the faint smell of takeout hung in the air. But in that cramped space, it had felt like home. They’d talked about moving, maybe finding a spot in the Mission District where they could spread out and breathe easier. He’d promised her a fresh start. But somewhere along the way, things went down hill. He’d started to withdraw, ghosting his own life.

For a while, things with Iva had been good. Not perfect, but better than anything he’d known. She believed in him, in them, in his ambition to build something in the tech start-up world. They’d fall asleep together on her king-sized bed, two figures curled up, lost in space. In the mornings, she’d make borscht and pierogi in the tiny kitchen, humming songs in Ukrainian, while Jake sat at the desk with his laptop, checking the news and email. He’d glance up occasionally, thinking this, this was what life was supposed to be. But over time, the cracks showed. The drinking crept back in, and before he knew it, he was checking out of their plans. The fresh start they’d talked about slipped further away with each day.

Jake went out while Iva was at work, sometimes hitting the Starbucks down the street and working for a while. But he always ended up at the bar. He was supposed to be apartment hunting in the Mission District, looking for a place for them both, but he never could get the gumption to do it. He didn’t know why.

He’d come home at all hours, offering little explanation. Iva put up with it until it started disrupting her early work schedule—bed by eleven, up by six. Then it turned to arguments, no longer just cold looks and mumbled complaints. Jake found himself sleeping on the couch half the week. He knew it was his fault. But bogged down by his divorce and growing distance from his company, he couldn’t figure it out. It made no sense. Anyone could see he drank too much, but Jake knew that was just the symptom. The problem was, he didn’t know what was driving it. And he didn’t buy into the “recovery” pitch. He was executing loops, it seemed: loops in code, loops in machine learning training and testing, and now he saw that the most successful web 2.0 companies were trapping human being in behavioral loops. Drinking, getting inebriated, waking up with a hangover, and doing it all over again was just another loop. It wasn’t a journey; it was nothing to be proud of. But then, what here in the Valley could one be proud of? Pride seemed a bit too quaint after millions of dollars in investment and promises to change the world.

Iva was beautiful, nine years younger, hard-headed but soft-spoken. Charming. Complicated. Jake loved her, but not enough to change. Not enough to stop all the loops.

“Where are we with apartment hunting?” she’d ask, and he’d give her some half-truth about researching AI and putting out fires at his company in Palo Alto. In reality, he spent hours in cafés, reading whatever caught his interest—Moby Dick, The Brothers Karamazov, some book on chaos theory—before winding up in bars while the sun was still up. He hadn’t been to the office more than six times in three months, and the silence between them hung heavy. Jake was ashamed of his failure to launch, but he had a fallback—his condo in Palo Alto, or a hotel if needed. He could break it off and return to his old life. Still, he wanted to make it work and urged her to be patient. But when her dirty looks or cold silence hit him, he’d end up back on the couch, staring at the shadows until sleep took him. By morning, he’d pretend to be asleep. He adored her, yet somehow, he found himself doing nothing for her.

The nights blurred together, one bar into the next, impromptu chat with strangers and rounds of drinks carrying him further from her. Finally, he packed up his Porsche and moved out while she was at work. He felt bad, but he wanted out. Part of the issue was that he was living in her apartment. That was their agreement, and it was supposed to be temporary, but after a month they weren’t any closer to moving.

He liked having his condo in Palo Alto, though he hadn’t renewed the lease, and for some reason, he didn’t like the idea of apartment hunting on his own. That was something they were supposed to do together. He didn’t like her dictating how things should go. All of this was true, but he still couldn’t figure out why he’d fallen apart. But he had. He wasn’t good for her, and if he was honest—it was harder to admit this—she wasn’t good for him.

After he’d moved out, she bombarded him with texts until he told her where he was, and she showed up later, knocking on the door of the Best Western he’d made home. She stood in the doorway and cried, and that did it.

“Jake…” She looked at him like he was ill. That was the beginning of her insistence that he go to rehab. He needed help, she said, and he almost shot back that she sounded like a brochure. But he wasn’t good at being rude to her. He was good at running away.

No matter how hard he tried, Iva always found her way back into his thoughts. Her green eyes lingered there, sometimes in his dreams. She was beautiful, the kind of beauty that you couldn’t extinguish—no amount of insults or hard words would erase what she and he knew, that she was a gift and a beautiful woman. She was so beautiful that he stopped using her name. She became “The Girl.”

Palo Alto

The taxi arrived in Palo Alto from Scotts Valley, and Jake stepped out with a polite goodbye to the driver, squinting into the sun, the familiar bustle around him. The air carried the scent of bay laurel and lavender, a gift from the “not in my backyard” locals who lived steps from the sleek glass buildings where the future was being built. Mythic Palo Alto. The heart of Silicon Valley, full of Stanford brats, hopeful transplants, and wannabe tech billionaires. They swarmed the sidewalks with cell phones clutched like lifelines, and eyes alight with dollar signs. Everyone here was talking about “changing the world,” flipping channels to the next venture.

Palo Alto; mythic Palo Alto. It was a little town south of San Francisco (aka “The City”), west of San Francisco Bay, located on what was locally known as “the peninsula.” Everywhere else it was in Silicon Valley. Palo Alto was sixty thousand people. It was full of Stanford brats with a sense of entitlement, recent transplants hoping to make it rich from other places like Austin or Seattle or London, and generally hordes of rich young people and wanna-be rich young people all incessantly, pompously talking about the next new thing. Women walked with noses skyward in expensive skirts, sporting faux accents, driving European convertibles, making love to snarky Polo shirt-clad husbands yacking into cell phones six days a week at venture firms or corner offices in software companies. Jake thought it was beautiful. It was beautiful.

Xerox PARC (Palo Alto Research Center) had put Palo Alto on the map (though Stanford had arguably already done this) in 1970, inventing technologies like laser printing, Ethernet, Graphical User Interfaces (GUIs), on and on. In the technology boom of the late 1990s a host of Internet companies had sprung up from the volumes of venture funding pouring in from nearby Menlo Park, and in the wake of that debacle new Internet companies had sprung up again, only to die and make room for more, ad infinitum, Lion King circle-of-life style. The latest example was that shitball company Facebook, founded in the shadows, with rumors of stolen intellectual property, a Harvard U startup by that Ivy League brat what’s-his-name, with his groundbreaking technology vision of masturbating to nineteen year old Harvard chicks online. Fucking Facebook. They were in Palo Alto.

So it was small, Palo Alto, and almost year-round sunny. This attracted a certain breed, a certain person-type. If Jake spent any time there he would get it, he would figure it out (unless he was one of them, of course). He got fire hosed in Palo Alto with a weird, almost religious zeal for all things novel, gadget-y, and monetary (Californians generally had this, to some degree). Products.

But the steroidal materialism driving the place—driving the product zeal—was carefully disguised and concealed by a full throated Silicon Valley double speak, a ubiquitous lingua franca appearing everywhere online and in print, in big lettered glossy ads, and blasted out of the pie-holes of tech-company “leadership”—founders of the perpetual startups, CEOs, CTOs, Chief Marketing Officers. Hell, security guards. They were all bringing about change, namely by parroting out a 24/7 stream of proclamations about change: “we’re changing the world!” and “changing the way you do business!” and “changing the way we communicate!” The Facebook guy, Zuckerberg—that was his name—was always going on in some quasi-articulate, shallowly philosophical fashion about changing the nature of privacy and creating a new, more shared world and all of this rubbish. No wonder he was in Palo Alto. Maybe this shit didn’t fly at Harvard, but everyone was knee deep in it in Palo Alto.

Jake continued. He didn’t blame or dislike capitalism, not at all. But in Palo Alto you couldn’t be overtly capitalist, like say a Donald Trump could, or even a restaurant owner, or the CEO of I don’t know Black and Decker or what have you. This was what made it annoying. You had to have a kind of religiosity to what you were doing, believing in the future, in making the information available to everyone, in bringing families together, saving lives, saving the little kids with the flies in the eyes down in Africa. Embracing change. (Here was some change: get off the Internet for a while.) The exhortations of “change” he was pelted with daily were, mind you, from startups actually mimicking other startups in basically a lemming-type, a sheepish kind of behavior where only rarely was there a new, really novel, really interesting idea, like maybe the mouse back in the 1970s, or more recently say Google. Not fucking Facebook.

So in the midst of the VCs funding what looked like everything else, and the software guys adding gadgets on top of gadgets like everyone else, everyone was making money hand over fist by market-speaking him onto the Internet where he got twenty pop-up distractions and music and IM and email and stupid YouTube clips while he was trying to read something about physics or some treatise by Kant. This was the change he was getting. In the meantime he never heard a peep about the stock options or the fat salaries or the Lamborghini he saw on University Ave with the twenty-something kid emerging from the custom leather seats, wearing sandals.

This load of utopian-preachy make-the-world-better B.S. no one really swallowed tout court, even in Palo Alto, but you couldn’t just run around worrying about when your stock options would let you retire, so you could get a gold couch and walk around like Bob Guccione or that guy who started Playboy, you know with slippers and a robe and the fake tits following you around everywhere. This wouldn’t do in Palo Alto. You had to say something inspiring, even if you were making a goddamn video game designed to play on a mobile device enabling thirteen year olds to shoot pedestrians with AK-47s, splattered blood and brains everywhere, almost like it was real. Inspiring. World changing.

Anyway this was Palo Alto. It had this magical, fey quality to it, which was worth witnessing firsthand, even with the B.S.

Palo Alto was also a decent place to go on a prolonged pointless drunk, to bender around. To be clear, it wasn’t as good as The City (San Francisco), or down on Sixth Street in Austin, or Lakeview/Wicker Park/Wrigleyville in Chicago, or Duval Street in Key West, or the inimitable Rue du Bourbon in the French Quarter in New Orleans, or... he made the point. But still it had its share of nightlife for a small place. The Rose and Crown. NOLA. Scotty’s. The Old Pro. (The bender-ready Nut House, but that was another issue as it wasn’t downtown.)

The main issue was that his bendering happened under an iron curtain threat of getting gossiped around in little snobby whispers and guffaws by locals and bartenders, which would make it more difficult to show up all innocent-looking the next weekend, or on that same Wednesday. This was the small-town aspect to Palo Alto, the nowhere-to-hide part of it which, when combined with the mega-dose of heightened professional expectations—changing the world and good-guying it and all that—ended up making things tricky. Jake knew all of this, of course, but had ceased giving a damn. He knew all this because Palo Alto was where he had lived and paid the exorbitant rent, until a few months ago. He knew all this because Palo Alto was, in fact, where he had located his own software startup.

Jake smiled and almost laughed out loud thinking about Palo Alto, as he moved through it in the typical assortment of geeks and nerds, tourists, academics and models. It was, he thought, olfactory heaven, too—endlessly new whiffs of colognes and perfumes and the various smells of the cafes, bars, and shops. European convertibles lined the streets, driven by Ivy League-educated CEOs or twenty-something millionaires. Two twenty somethings walked by outfitted with Facebook paraphernalia—shirts, caps, and hell, backpacks. Goddamn Facebook, he thought. He hated Facebook. And yet, here he was, walking the same sidewalks, scanning for his vintage Porsche parked a few blocks down. He exhaled and hoped he’d be in his car soon, to reclaim part of his fractured life. He missed his car, if that were possible.

Jake wouldn’t say a lot about his Porsche except the damn thing had been sitting for a month while he had been, well, also sitting for a month, in The Camp. It was chock full of his things from driving down from his folks’ place in Spokane to The Girl’s place in East Bay earlier that summer. He hadn’t been at The Girl’s long enough to unpack much, but that was another story.

So, he got to the Porsche at the address given to him by the guy who parked it and paid the taxi driver, and the first rule of finding your car was that your car was not at the address written down but close by. This made sense, because it was open parking on the street so the guy found a spot, like, nearby. Next rule was a corollary, which was that unless the guy was on acid, if it had been closer to another address, you would have that address on the paper, so it was very near what you were reading. Look around. Up the street, uncomfortably crammed against a bunch of residential bushes, he found it.

A big conspicuous sign perched on the front windshield saying “not abandoned — if need to move car please call xxx-xxxx.” His buddy had hooked all this up. The door was inaccessible on the driver’s side and opening the passenger door resulted in an eructation of vitamin bottles—some lids not closed tightly—so he was standing in venerable residential Palo Alto with his dusty month-long sitting Porsche, beer buzz wearing off, staring at innumerable vitamins and fish oil capsules and fuck, he didn’t know. All this street cleanup. He was feeling drinks, that was all; not keen on any of this. The goddamn Porsche was chock full of everything he owned and he could barely climb over it all to get into the driver’s seat.

Top of Form

Bottom of Form

He yanked open the passenger door, and vitamins, loose change, and bric-a-brac spilled onto the sidewalk. The seat was crammed with his belongings—stuff he’d thrown in after leaving Iva. The musty smell hit him; stale air and old upholstery. He crouched to gather the mess, shaking off the thought that his life had been like this for months: scattered and unclaimed. He slid into the driver’s seat, and putting his hands on the leather wheel he felt better. His dad was a retired airline pilot and had been an amateur but quite enthusiastic drag car racer. Later he stopped racing and would renovate old muscle cars. Jake viscerally loved cars and especially loved old antique cars restored to their former glory. Computers were obsolete within months. His Porsche was over four decades old and appreciated now more than ever.

Jake turned the key in the ignition. The throaty growl of the Porsche cut through the stillness. Merging into traffic and hanging a left, he took in the mix of business and pleasure down University Avenue: glass buildings, Thai restaurants, sports bars with Stanford students and software developers and investors with shades and expensive watches and perma-smiles.

He pulled onto Emerson and drove past The Tap Room, a bar-restaurant a short walk from his old office. Back in the day, he’d swing by after work—sometimes early—to meet Mickey, his unofficial advisor and mostly, his drinking buddy. They practically lived there, hashing out startup headaches over greasy burgers and cold beer.

By now Jake’s Porsche and Jake’s mind had melded, and he felt more than fine. He hadn’t driven in a month, and the feel of the wheel in his hands reinvigorated him. He didn’t have anywhere to go, though, and he had no plan. This was a brake on his open road feeling and he scanned for parking spots still downtown, and parked down the block from The Tap Room. He remembered he’d parked in the exact spot before. Flashes of a bleary night crept back to him—driving from the East Bay, getting stuck on train tracks in Pleasanton, getting to Palo Alto late and busting in here for serious drinking. He’d almost gotten cut off—an impossible outcome with his drinking skill.

His Porsche was a machine, he thought, with a smirk. Yet behind the wheel, he felt free. Digital tech promised that same feeling, but it was a trick. Digital tech was more of a dystopia, a “Machineland”—a looping mechanism tightening its grip. Technology only moved in one direction, he thought. Who was it—Heidegger? That ex-lover of Hannah Arendt warned of the world becoming a “standing reserve,” a resource to be tapped and put to some utilitarian end. A waterfall was beautiful until it was made to turn turbines. This was the pattern now, he thought, shrinking any notion of spirit or freedom down to a dot. Palo Alto is a standing reserve.

Who was it? Jake thought, his favorite quizzical prelude to recollecting something. It was Mumford, Lewis Mumford, who had written about “megamachines”—huge social machines powered not by engines but by people moving in sync, pressed into service by a few autocrats at the top. The Egyptian pyramids were megamachines. They were built by thousands of Slaves and workers, under the control of a handful of architects, themselves reporting progress to a smaller handful of elites. Now the World Wide Web was the new megamachine, enlisting millions to feed algorithms and pump up stock prices for an ever-shrinking modern elite. Progress? Or were we just running in circles? He’d been in a Faraday chamber for news while in The Camp the last month, but he’d been wondering about the Machineland around him and had grown increasingly wary of the web 2.0 designs, which championed “democratic” decision making with “Like” buttons and other one-click ways to allow a user base to signal intent. Before, media companies hired professional writers and journalists and editors to filter garbage from the front page. With Facebook, Twitter, and other ventures like Digg, users voted and the vote count was the new filter. The most votes or “likes” won and promoted the news or other item to the top to be viewed by tens of thousands, then millions, now billions of eyeballs.

That was the architecture. The question was how it made money.

The monetizing-by-attention loop wasn’t a crooked plot by some Babadook evil genius. Larry Page and Sergey Brin, now legends in the Valley who launched Google search on Stanford servers, then cashed a hundred-thousand-dollar check from a lucky seed investor to start it in a garage, had stumbled into the business model now strangling the world. Jake would have done the same thing. Anyone would.

By the early 2000s, Google was growing exponentially and wasn’t making any money. Investors in the Series A “big round” initial financing wanted to get paid back, not cheer on a couple of Stanford dropouts thrilling nascent web users by making the web searchable with simple, intuitive keywords.

But how to pay back the investors? There wasn’t a product. The service was free, as was the ethos of the early web.

So? Advertise.

The only remaining problem was getting millions of users to click on the ads. That was easy enough. Make them relevant to the keyword search. Now someone searching for “great Caribbean getaways” would get ads for suntan lotion, not umbrellas.

And the money poured in.

There was just one problem. Since the web was now doing double duty as a library of information and a voracious advertising engine, how do you hide the strategy?

Simple.

Make the ads small and unobtrusive… “smart.” Imbue them with contextual connections to what users search, make them seem part of the technological magic of the new web. The ads weren’t annoying, like on TV. They were, somewhat perversely, part of the marvel itself. Another score for the golden-boy entrepreneurs. There was nothing they—and the web they were creating—couldn’t do.

What these guys had figured out—what everyone was copying—was simple enough once you stripped the slogans off it. You weren’t the customer. You were the inventory. Every click, every pause, every scroll—time-on-site, they called it—was logged, measured, and fed back into the system. The product wasn’t the page. It was your attention, bundled and sold upstream to advertisers.

So the game wasn’t to inform you. Or even entertain you, not really.

The game was to keep you there.

Longer sessions meant more ad impressions. More impressions meant higher revenue per user. Higher revenue meant a better story for the VCs, a higher valuation, another round. And once you took venture money, there was no backing out of that logic. Growth wasn’t optional. You had burn, you had expectations, you had a clock. If engagement flattened, you were dead. So you didn’t ask whether keeping people hooked was good for them. You asked how to move the metric. Daily active users. Session length. Retention curves. Whatever moved the graph. The system didn’t reward restraint. It punished it.

The early visionaries spent their days and nights changing the world, by upping retention.

Endless scroll so there’s no natural stopping point. Notifications to pull you back in. “Like” buttons so you could register a micro-hit of approval without thinking. Feeds ranked not by truth or importance but by engagement—whatever kept you clicking, commenting, and reacting.

It was feedback control, really. A loop. User behavior went in, metrics came out, the system adjusted, and then it went again. The tighter the loop, the better the business. And the worse it got for the person inside it.

Users got dopamine hits from clicking like buttons and subverting old editorial authority with sites like Kevin Rose’s Digg and blogs and all the rest of the new user generated content web sites. Everything was a loop. Same architecture, he thought. Change the input—booze or clicks—and the loop still closed.

Customers click buttons for dopamine hits; the founders spent long hours finagling more obsessive use of their web sites. Behavioral loops sold you advertisements. Dopamine loops kept you clicking (you’re changing the world!).

The machine that came out of the Valley wasn’t democratic. It was, in fact, the opposite—a walled empire for billionaires. Instead of freedom, people were tethered to algorithms, building new pyramids for the gods of consumerism and the one percent.

Jake cut the engine and leaned back, taking a breath as he conjured the familiar wood and brass front door of The Tap Room. A pair of polished tech bros walked by, glancing at the dusty Porsche, then at his rumpled shirt. They said nothing, but Jake could feel their judgment. He didn’t care. Not anymore. He cracked his knuckles and stepped out of the car, letting the pull of the bar guide him toward the taps where a beer was waiting.

Jake sighed, exhaling. Time for happier thoughts as he was just now entering The Tap Room.

The Tap Room

Inside The Tap Room there was a long wooden bar, an admirable line of beer taps, and rows of liquor on the back wall. The head bartender was a generally jovial, sometimes moody Bostonian named Dean. Jake knew him only by his first name. He and Mickey would rack up big tabs in The Tap Room, so Dean would generally give them the four-star treatment.

“Jake! Hey, how are ya’?”

“Fine, Dean, fine. I’ve been in Seattle,” Jake said, lying, because how was he supposed to tell people who knew him as the founder of a tech startup that he had been hanging out with twenty-year-old heroin addicts in a rehab for the month of February? Screw that.

“You look great!” Dean bellowed. “What have you been up to?”

“I quit drinking for a while,” Jake said, which was the very epitome of an honest statement. He was all about honesty, whenever it wasn’t too inconvenient. If he took a hard look at honesty sometime, he saw how it was a Platonic Ideal and there was no real honest man from morning until night. That was why this one felt so good. He got another honest statement in for the day with no real hassle or worry.

“Wow, good,” Dean said, pouring him a beer.

Jake asked him what was new and Dean said this and that and “ya know,” and Jake didn’t really care because he had a cold beer from on tap in front of him and it was a sunny day outside and he was feeling fine.

Fast forward a few beers and he was reciting this Edgar Allen Poe ditty in his head:

Quaintest thoughts, queerest fancies, come to life, then fade away. What care I how time advances? I am drinking ale today.

This was reportedly inscribed on the wall of a pub somewhere in New England, by Poe. It seemed like a Poe-ism.

There was a younger sorority-looking girl down a few bar stools sitting with a bored-looking young guy, eye on her body not her mind no doubt. She was maybe twenty-two or three and she must have wandered off of 101 from somewhere else; San Jose State, maybe. She wasn’t Stanford material, no. There were no mentions of Constitutional law or business or politics or whatever subject—he was at Stanford, that was the point—in that self-assured loud confident speak that announced to everyone you were somebody and you were going somewhere.

The Stanford girls had that energy in the bars around here, and you respected it. Of course like any college kid they were still waking up half the weekends with Jell-O shot hangovers and pained recollections of taking their shirts off and there was that disgusting used condom experience, all red and still kind of wet, bunched up in the covers from what’s-his-name who didn’t bother to dispose of it in the bathroom after passing out last night.

This girl, yeah, she was San Jose State, maybe.

On her barstool she was leaning forward toward the guy, a tanned puffed-up looking fellow, and she was doing all the talking, which by degrees Jake discovered was an attempt to describe something to him. What was it? He leaned an ear in her direction.

“It was very cool,” she was saying to the guy, with this kind of head-jerky, avian manner, and vowels—drawn-out vowels. The guy was nodding and saying nothing.

“I mean later it was like, not so cool, but at first we were [inaudible] and...”

The guy nodded. Said nothing.

Jake realized at this point that she was working with “cool” and “not cool” pretty much exclusively; she was a meat-and-potatoes girl and used what she knew best. Like finger painting with red and blue only; if she needed a purple she mixed and matched. Like: something sort of cool was a few “cools” and a couple of “not so cools,” or if it was mainly cool it was mostly “cools” but a few “not cools” or maybe even one.

He figured she had all this figured out, like a calculus, you know, and who was he to judge? Dean had the music playing low in the background and it was that “what I am is what I am are you what you are (or what?)” jangle with Edie Briquette or Brickel, yeah, and Jake was listening to this and listening in on the girl so it all became, you know, what she was, was what she was, and was he what he was, or what? He raised this large Hefeweisen with a lemon and it all drained down with the music and the beautiful girl with her cool calculus in the afternoon sunlight, which was streaming in from the windows facing Emerson.

The beer was gone. “Dean!” One more, he said.

“Yeah,” Dean said, kind of smirky, like a “here we go again” combined with a “why not?” Dean made his tips. He was from Boston, for Christ’s sake. He didn’t drink much anymore, Jake didn’t think, but Christ, he was Boston.

Okay, on this girl again. She was San Jose State, Jake figured. Physically, she was blond and she had on an orange tank top with bronzed, thin arms, and these well-developed breasts. You could see the little straps of her bra set off from the straps of her tank top. They—her breasts—they were sort of...they were really very, very nice and unfettered, like landmarks in protected national forests. Really pristine.

More Poe was in order. Fill with mingled cream and amber, he would drain that glass again. Jake wished he knew some Rameau or Baudelaire. The girl—maybe he could impress her—anyway the girl could say “that was like, really cool” if he could reproduce some Baudelaire, say, and maybe the guy she was with was actually only a friend, or fuck, he didn’t know, a robot. He was the laconic type. Was he flexing? Not sure. These were atypical peeps in Palo Alto. Dean largely ignored them. He was chatting with an older guy with a newspaper down at the other end. But Jake liked the girl and thought she was pretty and she had a cool calculus and all he had was a bunch of goddamn recovery words like “relapse” and so on. He left it at that.

Out of the blue, Mickey appeared. “Jake!” he called out, big shit-eating grin. A few regulars looked up, beers in hand. “Thought you’d disappeared for good.”

Mickey slid into the stool next to him, still smiling like a maniac.

“Yeah, just took a little sabbatical,” Jake said, smiling back. “Good to see you, Mickey.”

Mickey chuckled. “This place will do that to you. Everyone’s got a dream, but half of us don’t know how to survive it. What’s the latest?”

Jake knew he meant work. His company, Knexient. His software platform was turning heads despite Jake’s Houdini act, inviting big money, Series A money, from the VC sharks.

Jake shrugged, his eyes on the worn leather strap of his watch. “Yeah, barely. Trying to scale some of the faceted search, but that’s compute intensive, and we’re cash lean for servers and infrastructure. Don’t know yet. Investors want results yesterday.”

Mickey nodded, his face clouding. “We’ve got some traction with badges. The idea is to gamify the user experience. We’re not the first with the idea, though, so we’re looking for the unique value proposition,” he admitted.

Jake wasn’t interested in being a killjoy, but he couldn’t stop himself from saying, “No one who looks ever finds that.”

Mickey clouded for a second, then recovered his confidence. “Well, we’re on track anyway. Should be okay with the latest build, and we’ll see what the beta testers say.”

Mickey was chastened, though. Jake knew he had broken an unwritten rule: those who dared to dream and build a startup should always be treated as on their way to success. Jake shrugged. He liked Mickey, but the relationship was transactional. Still, Mickey stuck around, and they chatted in a way Jake enjoyed.

After Mickey left, Jake lingered in the bar for another drink, his fifth. A bit drunk, his thoughts drifted to Iva, to the life he’d imagined with her, the company he was barely holding onto. He thought of the late nights, the pitches, the promises. It all felt so fragile now, like a house of cards, a foundation built of sand. At least, he thought, he wasn’t sending petulant texts to The Girl. It was a clean break. So far.

Jake paid his bill and waved to Dean on his way out of The Tap Room. Outside the entrance, it occurred to him that he had no plan: nowhere to go, nowhere to sleep—his condo technically not his anymore with the lease expired, and all his stuff still inside. No plan for dealing with the mess he’d created with his company while working through some co-dependent, toxic nightmare with The Girl in The City. No idea how he’d stop drinking now that hope seemed lost. This was Benderland. Be careful what you wish for.

Jake walked up Emerson to University and headed for a place where he was sure no one would recognize him. Of all the corporate non-Palo Alto eateries, he was headed for The Cheesecake Factory.

Erik J. Larson

Scotts Valley - A Quick Escape.

This was all Girl-related; in fact, Jake would go as far as to blame it on The Girl. He was sick of all of this rehab business and he had this abominable emotional torment going. Everyone around the place—staff, that is—seemed artificially slow and almost deliberately annoying, like they were trying to eke out some last scraps of discipline and self-control and good decisions from him. After the stink he’d made the night before about keeping his cell phone it was obvious enough he was in some pickle with someone from the outside. Well, so what? He was getting out this morning regardless. Sure, he’d smoke another three or four cigarettes and bump fists and hug and force a few unctuous smiles, but he was getting out. Today. This morning. Oh, it should be said: Jake was in rehab.

Details on the rehab are another story. Let’s umbrella-describe it as alcohol abuse; the technical term was alcohol dependence, meaning a physical and psychological reliance on booze. True enough. But anyway, The Girl had screwed up his hopes at a fresh start, a different go at things, as she had rather callously announced via short text to his BlackBerry that she wouldn’t be picking him up from rehab, this followed by some salt-in-the-wound commentary about her belief that he could “take care of himself.” This sort of thing.

What amazed him was the utter imperviousness of The Girl to basic relationship concepts. Everyone could take care of themselves, but if her logic enjoyed much of an audience the whole point of dating would be lost. Androgynous robots walking around, taking care of themselves. A system with no shared state. That was the world The Girl seemed to envision. Imagine if she ran things. The Girl. Queen Girl.

Truth was—and he knew it—she was scared he’d drink again, or maybe she was scared he wouldn’t. She had a keen sense to her, The Girl did, as he was fixing to ease seamlessly from rehab discharge into a steady stream of drinking, a continuation of things past, as Proust might almost have said.

Anyway, Jake was still at The Camp. Details don’t really matter, but since he was still standing there, smoking perilously close to the line that marked the smoking zone, the title of the place is worth a moment’s thought: “The Camp.” Alarmingly ambiguous. Shades of vacation fun coupled with something edgier, boot camp and so on. It conjured camping, too, which was generally positive, so the poor bastards checking themselves in for thirty days of their lives were rewarded immediately with confusion about whether they were being disciplined or handed marshmallows. The truth was somewhere in between, as they say, though Jake deplored that phrase. In between where, exactly?

At The Camp they mixed a healthy dose of daily structure with a kind of Mother Teresa patience for endless talking. Counselors, yoga, meditation, films about addiction, AA meetings. For dessert you stood around the smoke pit—the smoker’s area, which was effectively the entire inpatient population—with surfer-looking guys and pill-popping divorcees and tatted-out beautiful girls, a few of them whorishly beautiful, trading stories and laughing and joking until a staff member broke it up and pushed everyone into another meeting or back up to the cabins for bed.

Teenagers and burnouts together, shuffled around the rustic buildings, cigarette smoke rising in slow plumes into the clean air of the Santa Cruz Mountains. Swearing and laughter echoing off the rock escarpments above. This was The Camp.

Jake smiled. He was out today, discharged as they say, and it felt like a kind of graduation after the forced intimacy of the last month. Odd that he’d only now, on the day of his departure, noticed how rustic the place was, set back off a country road and a short hop from the little conglomeration of stores and shops called downtown.

And here he was, smoking on the edge of a painted line, thirty days removed from everything, waiting to be released back into it.

Anyway. This was all in his rearview, so to speak. He had finished his thirty days.

Today.

Thirty days was about two hundred meetings, some yoga, a few dozen acupuncture treatments, two visits from The Girl (not conjugal—would disrupt recovery), and one torturous string of text characters on his BlackBerry the previous evening. The goddamn Girl was an amalgamation of something like hope and purity and Nurse Ratched and a kind of Slavic hardness that emerged exactly when Jake’s brand of American neediness flared up. Supermodel-looking, Ukrainian-born, Czech Republic–living, immigrating-to-the-Promised-Land Girl. Anyway, what bounced around in The Girl’s head was of little interest to him at the moment; he was getting out of there and he was going on a bender. The rest would take care of itself. Just like she said.

Ty was a staff member at The Camp who had picked Jake up from his entombment in a Palo Alto hotel thirty days earlier: barefoot, chain-smoking, and day-drinking vodka in poisonous quantities. That was Girl-inspired as well. Now, finally, with Jake’s belongings stacked neatly outside the nursing station in the Northern California sunlight, his head pulsing with impatience, Ty appeared in his “The Camp”–emblazoned polo, driving the minivan, ready to evacuate him out of there to Scotts Valley, the jumping-off point, the yawn of a town a couple of miles up the road.

“You ready?” Ty said, all toothy, the remnants of tattoos from his own drinking days still visible on his arms. He stood there smiling as if this were a big moment, and honestly Jake agreed. It was.

“Yup. Sure am.”

This was all pregnant with meaning and faintly ridiculous, as he was effectively t-minus something to bender, which might have been funny if he weren’t in such a pinched state over The Girl. She sat squarely on his frontal lobes and limbic system, pressing into the folds, dragging her bare foot across the contours of his brain. In that state Jake stepped into the minivan with Ty—The Recovererrrr!, as he imagined it announced, like a main event—and they drove out and away from The Camp, finally, toward Scotts Valley. Free, sort of. Not really. Gone, at any rate.

Scotts Valley, a sleepy little town a handful of miles from Santa Cruz, nestled in the Santa Cruz Mountains and sporting a collection of boring restaurants, fast food, a grocery store, hardware stores—a small town, everyday ennui. Scotts Valley didn’t feel like Silicon Valley. It sat just over “the Hill,”—as the locals called the up-and-down drive on California State Route 17—but a world away from San Jose, Cupertino, Mountain View, Palo Alto, Sand Hill Road and the rest of the Valley. Far away from the endless announcements of funding rounds and product launches. If Silicon Valley was the valley of opportunity, Scotts Valley was the valley of blue skies, chirping birds in lush forestry, and an almost aggressive peace and quiet.

No wonder they put The Camp here. The boozers and pill poppers and meth heads and dope fiends could finally relax, stand in the smoke pit watching the curls from their hundredth cigarette of the day, shuffle between buffet-style meals with half-caffeinated coffee—someone clearly mindful of the speed freaks—and talk themselves into something like stability. The Santa Cruz Mountains rose clean and indifferent above it all.

Scotts Valley. What else? There was a Peet’s Coffee, turns out, and just then in the parking area in front of it was a minivan from The Camp driven by Ty, containing a recently discharged inpatient, who, if viewed through the window in the late morning California sun, seemed perhaps on the precipice of a new type of life. The guy—Jake—was stepping out now and Barnum and Bailey–like smiling and volleying back responses to Ty’s enthusiastic, slightly condescending patter about aftercare and recovery.

Now, this effort Jake was making was half-valiant but also half-fearful, as it was so bizarre and treacherous to be planning benders just now. With secrets like this inside him, he had carefully adorned a smiling exterior—a full-of-hope head nodding and all that. The perfect simulation of a discharging inpatient for his man Ty, as Ty deftly cheer-led him out of the van in front of the Peet’s Coffee. Along with Jake’s exit came his huge duffel bag and a little piss-ant “The Camp” backpack, like the ones you give a nine-year-old on the first day of school, full of the desiderata of rehab: notes and books and handouts and a half dozen Bic pens he had collected here and there. Blame all this shit on The Girl for now.

Ty drove off and in a comic moment Jake imagined him still prattling about aftercare, his encouragements gradually weakening until they were drowned out entirely by engines and road noise, a distant staccato hammering on a roof somewhere, neighbors watering yards and hollering across the street; the panoply of small-town sounds. As he turned onto the road he waved.

He was a good guy. Jake might have connected a bit more with him in their final good-bye, but The Girl, still rubbing her bare foot voluptuously on his limbic system, was now eating something in his brain; she looked up with a mouthful of spongiform and guffawed almost un-Girl-like, huge giggles erupting up and down her beautiful body.

So here was Jake, in this type of pickle, which was only thinly metaphorical as he was feeling an almost medical need to get a drink on the heels, so to speak, of all this torment. Was there a bar in Scotts Valley? IS THERE A BAR IN SCOTTS VALLEY?

So, was there a bar in Scotts Valley? He doubted it.

Currently outside the Peet’s Coffee, center of town, eleven in the morning, duffel bag sitting with the little backpack perched on top. Sweat beaded on his scalp. Inside, a couple of queries to the baristas about “getting a beer” yielded mention of a liquor store open in the constellation of stores behind him. Maybe it was a curious exchange with the little barista girls, but he could give a damn. It was California and they could probably tell him where to get some meth to enjoy with his coffee. Christ, no one was really from there, right? They had probably moved out from Santa Cruz or from over the hill in the Valley—got a boyfriend making an extra dollar an hour down the road and traffic so much better and, you know, mom and dad ten minutes away. Life was good. Can’t wait to move to L.A. someday, but you know? That “you know” just trailed off, he supposed, because, you know.

Where was he? Oh right! Jake’s homing in on the liquor store like a salmon laboring up a river, singular in purpose. Genetic movements. Instinctual. He carried the duffel bag for a fraction of the few hundred yards, but that became ridiculous in the rising sun, so he hauled it all back to the front of the Peet’s and left it there. No one would steal it, he figured. If they did, they would have procured a pile of smoky T-shirts, shorts, dirty socks, reams of rehab literature, and a Big Book, complete with names and numbers penned in from his fellow inmates—a thousand ways to buy good drugs. They all fell off, especially the junkies.

And the boozers. A large, strong bottle of ale. A bottle opener.

Was Ty back to The Camp yet? Ty was gone. The beer was there. The Girl shifted nervously atop his troubled brain. She was quiet now and had stopped poking at him with her bare foot. Her little deer eyes darkened and she squinted as a slight flush of something passed across her face. She could just goddamn sit there. The German-accented man behind the counter in the liquor store smiled wryly as he handed over the receipt. So this promised—this happening, this event—promised with near mathematical certitude to be the beginning of something. It was called a bender—maybe a royal bender—and he would unpack the meaning of this in due time. Mostly, he would just drink.

Little details could bedevil a bender right out of the gates. Like: the ale bottle poking conspicuously out of the brown bag as he exited the liquor mart, and he had no place to drink it either. His truck was over the Hill in Palo Alto. He didn’t know anyone in Scotts Valley. The sun was high in the sky now as noon approached and the little town was a menace of quietude and peace and locals shuffling about, attuned to notice people like him with a big brown bag and a sweaty, nervous glance.

Where was he? In the relative epicenter of a horseshoe cluster of shops to his right and left. A front sidewalk snaked along the lines of shops, and lots of asphalt and nothing much in the interior, which was where the liquor store was, so he made his way toward the semicircle of shops with his big beer, onto the shade of the front sidewalk, and down along the storefronts until finally there was a bench set back on a side alley, safe from the immediate visuals of passersby. This damn, sunny, somnambulant yawn of a town.

Ethyl alcohol. This was Jake’s subject. Ethyl alcohol, or ethanol, and hereafter alcohol, was often and notoriously misunderstood. So he was there to clear matters up: alcohol was shit, literally. It was the excrement of a fungus known as yeast. Yeast was a ravenous little bastard that ate anything sweet, shitting out carbon dioxide and alcohol. This process was known as fermentation.

Now, during fermentation the bastard yeast would keep feeding on the sugar until it ended up dying of acute alcohol intoxication. True. Yeasts were boozers. Like frat kids who die of alcohol poisoning. There were no moderate strains of yeast. When they died out, what remained was roughly 13 or 14 percent pure alcohol. These were beers and wines.

Distillation picked up where fermentation left off. Distillation was developed by the Arabs around AD 800, and it enabled pushing the alcohol content far beyond yeast fermentation. These were distilled spirits. Booze. People already knew that booze was “proofed,” about half the actual percentage: 80 proof was about 40 percent alcohol. Proofing was a practice used in Europe hundreds of years ago to “prove” the alcohol content of distillations. It was still used today.

What else? Alcohol was a kind of shape-shifter, hard to classify. It was the only drug—the only drug—that could also be classified as a food. An ounce of pure alcohol contained 170 calories, about what was in an average baked potato or a glass of milk. Alcohol was energy to the body. It was also a sedative and a stimulant and a poison. Drinkers understood all of this.

Jake’s first drink out of rehab was a little foreign, almost unwanted. Cough-syrup-like, it drained down into his stomach, warm and ale and bitter. The alcohol in the fermentation entered his bloodstream through the lining of his stomach; once in the blood it was sent, like everything in the blood, to the liver, where the liver recognized it as a toxin and immediately began breaking it down. In the meantime, it found its way to his brain, where it would make itself known in a sequence of neurochemical changes. He paused, almost reflexively, to walk through the brain changes.

The mesolimbic system was where the action was, with booze and with psychoactive drugs generally. There was a sequence of neural connections in this system known informally as the “pleasure loop,” because they reinforced desirable activities like eating and sex by releasing the feel-good neurotransmitter dopamine. Alcohol caused this release of dopamine, for free. Just drink and it was as if you had done something pleasurable. The same mechanism held for other drugs—this was the key to understanding physical addiction. Booze also spiked other neurotransmitters, serotonin and norepinephrine among them. It was a shotgun-type drug.

Now, with all these neurotransmitters shooting across his synapses, his conscious experience was fixing to take a major upswing. This was colloquially known as feeling a buzz. For present purposes, the buzz could be defined as the sort of feeling that eased the discomfort of drinking after thirty days of expensive rehab, on a bench tucked away in the quaint town of Scotts Valley, middle of nowhere. The alcohol buzz made all of this seem okay.

The Girl, too. She was still up in his head somewhere, but shortly she would seem more abstract, like flushing a critter out of an unwanted corner of a room; she would have to take cover as his brain rapidly signaled that things were much better than they had seemed a few minutes earlier. Screw The Girl.

But he was getting ahead of himself. The ale was going down like cough syrup, like he had noted, and he was a little chagrined that it was almost upsetting his stomach. Keep drinking. This was the answer. Drink.

This was the answer. Drink.

So he was riding a wave, neurotransmitters shooting around and binding to sites, and suddenly he realized there must be a burger joint that served beer and he could use some food. Absalom, Absalom! The reference surfaced without much justification, and he noted to himself that many writers had been boozers and that the great Southern gentleman, author of that novel and of The Sound and the Fury, might well have begun drinking after a thirty-day stay at a place like The Camp. Faulkner, of course. And Faulkner might also have sought out a good burger and additional beverages. Come to think of it, Jake had never read Absalom! and wasn’t entirely sure what it was about.

In Scotts Valley, now feeling at home, he was rapidly metastasizing into a kind of weapon, a creature marvelously effective at securing short-term objectives at the expense of long-term goals. You found yourself in the moment and completely clear about the next step—in a tactical sense. Up the road to a diner he went. The hostess told him there was no beer. She mentioned some other place, a BBQ joint, and pointed. He exited and started up the road, then realized he was unsure which way she had pointed, so he was back in the diner and she gave him a clear, unambiguous direction this time, still smiling. Back down the road. The place wasn’t open yet, so he stood outside masquerading as someone in dire need of a BBQ burger, which, in fairness, he was. Inside they had domestic beers and burgers, and so now he was finally relaxed. He had arrived.

Texting was a kind of bugbear when drinking. Ostensibly a social good—how better to coordinate with friends when out—but it often became a nightmarish tangle of insensible, insensitive messages fired off once a few drinks transformed the usual measured outlook. The phone beckoned. Why always texting? He was drawn to it with a kind of mathematical certainty once his BAC crept up. It was treacherous, of course, because it ceded to her his dignified silence, a silence that would otherwise have been justified for a venomous Girl.

Someone might ask what The Girl had done, why he was texting her. Nothing was an answer. She hadn’t picked him up. That scratched the surface of his grievances. Good enough for now. He would explain in due time. So he was texting her in the BBQ joint in Scotts Valley in this natural way, ignoring the counterfactual—that had he not been drinking that morning, he almost certainly would not be communicating with her. Not so soon, anyway.

But it was all innocent in inception. Trojan horse–like.

“Sorry I overreacted. I’m out and doing well. Going to Palo Alto to get truck. Take care.”

She texted back immediately. She was somewhere looking at her phone. Lying in bed, texting him, instead of driving out to Scotts Valley to pick him up. She was responding. That made it worse.

Oh, that’s great you’re out. You see, you can take a taxi and you don’t need me for support.

Not her exact words, but the gist. Calm, measured, Russian. He knew she was protecting herself, her own vulnerability, but that remained an intellectual cogitation and set against his own torments it produced no real understanding. None. Why was he texting her back? Christ. He was too blunt-force traumatized by her sudden withdrawal to maintain any self-respect, but he was damn sure not going to get into a back-and-forth exchange. The burger arrived, mercifully. Another Coors Light, too.

Jake sat back in his seat and began, finally, to relax. He looked out at the parking lot and beyond that to the road, and to the green forests sprawling up beyond. Spending a month ensconced here was a departure, to be sure. This idyllic, sleep town, he thought.

There’s nothing here, he thought. Wait. No, that’s not true.

Netflix had started here.

Netflix chose Scotts Valley as its first headquarters in 1997. The company started as a DVD rental-by-mail service when streaming was still a far-off concept. Co-founders Reed Hastings and Marc Randolph set up their modest operation in an office park here, perfect for a startup looking to disrupt the home entertainment market without burning through cash on a more expensive location. Scotts Valley provided the essential proximity to tech talent and investor networks in nearby San Jose and Palo Alto, but maintained the quiet, slightly removed ambiance that made it easier for smaller tech firms to grow.

At the time, Scotts Valley was already a place where smaller tech firms thrived, with tech operations dotting the town alongside recreational areas and residential neighborhoods. Surrounded by lush forests and winding mountain roads, the town felt removed from the high-stakes world of Silicon Valley, with a mix of small-town charm and rustic beauty that stood in contrast to the sleek offices and mirrored windows of Palo Alto or Cupertino. This sense of tranquility made Scotts Valley appealing to founders who were after a laid-back but still strategically located environment. The “real” Valley, after all, was just over the Hill.

Netflix operated out of Scotts Valley for over a decade, gradually shifting from a DVD rental business to its current model as a streaming giant. By 2007, when Netflix launched its streaming service, the company had outgrown Scotts Valley and moved its headquarters to Los Gatos, another tech-friendly town further up the mountainside, closer to the main Silicon Valley circuit. The move marked Netflix’s rise as a major tech player, and it left behind a sense of nostalgia in Scotts Valley—this small town that had unknowingly hosted the early days of a future global entertainment disruptor.

After Netflix’s departure, Scotts Valley shifted back to a quieter pace, though the tech influence never entirely left. The town retained a handful of small tech companies, but it was mostly known for its residential pockets and its proximity to the mountains. Surrounded by redwoods and hiking trails, it had become a place people drifted to when they wanted to get out of the denser parts of Silicon Valley.

Jake was, he smiled to himself, “out of the denser parts of Silicon Valley” just then. “What awaits?” he asked. Thai food. Wine. Jake’s impromptu reconnoiter of Scotts Valley in the moments after his glorious discharge from rehab had revealed several drinking options, all of which had seemed hopelessly obscured from view when Jake stepped out of the minivan, too nervous and wound up to formulate the easygoing plan he had going on now.

Jake would add to his buzz on the heels of damnable rehab, before he caught a taxi over the Hill to Palo Alto to get his cool-ass vintage Porsche. Sure, sticking around and drinking in Scotts Valley might have seemed odd given Jake’s lack of enthusiasm for the place, but any drinker would recognize the logic of not wanting a just-started drinking experience to be put on pause for forty-five minutes in a taxi ride: buzz waning, restless, fidgety, are-we-there-yet type of feelings in the back of a cab. Wrong-o. This was all filed in the avoid-if-possible drawer, and so he was in Scotts Valley and this was why. After the next glass of Chardonnay—he loved the dry whites—he would hoof it over to the Peet’s and get a number to call a taxi.

Jake eventually left the Thai restaurant and now the baristas back at Peet’s Coffee seemed wonderfully pneumatic and transformed into smiling, cute little balls of customer service; soon enough a taxi pulled up outside and with waves and smiles and strange goodbyes—the sum total of their connection a medium dark roast and a huge duffel with a Fisher-Price backpack left on the sidewalk—he was finally, finally leaving rehab.

To the taxi driver: “Palo Alto.” “Sure.”

“What brings you to Scotts Valley?” the driver asked. He was a bigger guy, rotund, maybe fifty, with a leather vest, Harley fashion, and a graying goatee.

“Rehab.”

The driver laughed. “Right on.”

The taxi driver, a wiry old guy with sun-bleached skin, gave Jake a once-over, his gaze lingering on the duffel Jake slung into the backseat. “You traipsing across the country or something?” the driver joked, eyeing the bag. “Looks like you’ve got baggage for the long haul.” Funny, Jake thought. Baggage. More than he could carry, both in his hands and in his head.

Jake gave a quick laugh for an answer, then lapsed into silence, staring out the window as they drove, running his fingers through his hair occasionally and exhaling deeply. The driver seemed to take his silence as an invitation to drop the back-and-forth. “Okey doke,” he said, switching the station and leaving the volume low, falling silent too.

The town crept by, a little tech graveyard, Jake thought. The fresh summer air slipped through the cracked window of the taxi. The interior smelled of worn vinyl and faint cigarette smoke. Jake stared out at the quiet streets again. Take a good look, he thought, you’re about to return to the belly of the beast.

It was summer, 2011. The world felt on the brink of something—revolution or collapse—no one could tell. You could feel it even out here on the road, coming over the Hill, back toward the action, the endless churn. In Egypt, crowds had risen to tear down a government, and in Silicon Valley, companies were rising just as fast. It was a new gilded age. Every week, there was another IPO, another billionaire minted overnight. Venture capitalists were sinking millions into any idea with a whiff of disruption, hungry for the next big thing. Social media was a way of life, a new kind of gospel, connecting everyone and amplifying the pulse of change.

Twitter was a megaphone for movements; Facebook was the new water cooler; and Google+, newly launched, was supposed to overthrow them all. Jake didn’t see the point—everyone was already living in Facebook’s walled garden—but that didn’t matter. The relentless drive for innovation kept pushing boundaries faster than anyone could process. Every day, some app, some algorithm, or some widget was proclaimed the new holy grail. Jake could feel it in his bones: no one was sleeping. No one dared to miss the next big thing.

In spite of himself, Jake found himself ruminating about The Camp. In the penumbra of Silicon Valley it had, he thought, an unusually large percentage of good looking and talented customers. There was a girl in the teenage buildings set apart from the adults by a few hundred feet but locked down for, presumably, improper adult and “non-adult” relationships. She came up to sing in the “talent show” The Camp put on, with microphones, guitars and a stage. A counselor had told Jake she was seventeen and worked with the biggest names in pop. Jake had asked because when she took the stage to sing Rhianna’s part in Eminem’s “I Love the Way You Lie” he was struck dumb by how powerful she was, singing it better he wondered than Rihanna, if that were possible. Then there was Olivia. And what’s her name. His rehab buddy would roll his eyes when Jake mentioned one of the women, for the obvious enough reasons.

Rehab collects people cast out from their lives, expelled temporarily from the glitter, the hopes and dreams and plans. Jake exhaled and sat back, grateful the driver wasn’t chatty. After thirty days of talk therapy, he could live with some silence. The ale he’d knocked back was working through him, and he let himself drift back into thought, and what a time it was to think and to live. Scotts Valley, Palo Alto, Mountain View, San Francisco. He was here, in the eye of the story. Living in the dream.

On Highway 17 they rolled along over the Hill toward the Valley—ahh, the local lingo. Scotts Valley and The Camp were somewhere behind him. Destination was in front of him.

Destination: Palo Alto, California.

Hi everyone,

I’ve avoided writing this post. I’m still fearful of reprisals. Let me jump in.

I studied math and philosophy as an undergraduate and enrolled in the Ph.D. program in Philosophy at The University of Texas at Austin many years ago. With my master’s degree in hand and a pregnant wife who was getting understandably alarmed at my lack of job prospects, I found an AI company in Austin that effectively threw me a lifeline: Cycorp, meaning “Encyclopedic knowledge” corporation, founded and led by a brilliant pioneering AI scientist, a rotund and jovial fellow, named Doug Lenat.

Lenat had (as I recall) been a professor of computer science at Stanford and other venerable institutions like Carnegie Mellon, and made a name in the field with a program for discovering new knowledge—think proofs of theorems and even scientific laws—named Eurisko. Eurisko was classed as a “discovery system,” and though it didn’t use machine learning as we understand it today, it did employ a bottom up or empirical approach to AI; the idea was from a set of rules and heuristics to perform intelligent search to discover new knowledge.

As the story goes, Lenat became frustrated with the (successful by the standards of the time) difficulty of encoding new “knowledge” or in other words rules and heuristics into Eurisko, and soon branded it as a “shallow” system in AI—a catch-all term of opprobrium, tracing back to the symbolic vs. connectionist approaches to AI since the inception of the field. “Shallow” meant connectionist. “Shallow” meant lacking knowledge of the world. “Shallow” meant not enough to get to “true AI.”

We see this same debate play out in discussions about the limitations of LLMs today. Researchers and semi-apostates like Yann LeCun bemoan, in effect, the shallowness of current neural network—in traditional lingo, connectionist—approaches lacking world models and relying only on data for inference.

“Shallow” meant founding Cycorp. Lenat’s disillusionment with symbolic machine learning systems like Eurisko led him to his astonishing hypothesis: the culprit in the failure of AI to date was the brittleness of “ground up” empirical systems; machine learning systems, mostly. By Lenat’s lights, the deep neural networks we use today are still brittle and shallow, because they don’t have knowledge, in the form of explicit facts and rules, about the world. This means they don’t have “common sense” (another dog whistle phrase for the symbolists). This means they’re hopeless candidates for artificial general intelligence, or AGI.

All well and good. Lenat wasn’t alone in turning away from machine learning approaches, and in the 1980s while this fight in AI gained intensity, it’s true that machine learning was severely limited, due in large part to a paucity of available training data, and powerful machines to crunch it. Trying something else made sense.

Lenat formed Cycorp in 1984 under the auspices of a large research institute, the Microelectronics and Computer Technology Corporation (MCC), and from the get-go the new knowledge-first venture attracted Department of Defense money, and in particular funding from DARPA. The DoD was eager to find an answer to Japan’s much ballyhooed “5th Generation” project, which adopted the “knowledge not data” premise, and promised a surfeit of advanced AI applications in the form of “expert systems,” AI programs with explicitly coded domain knowledge suitable for solving problems in medicine and other scientific fields with specialized knowledge that could be encoded programmatically.

I joined on January 3, 2000, at the beginning of a new millennium.

There was an immediate problem. I add already convinced myself in my graduate studies in logic, math, and AI (I was also taking classes in the computer science department) that knowledge-based AI was hopeless. We couldn’t write down everything that might conceivably play a role in some reasoning or another, and even if we could, the computer didn’t have any natural mechanism to make use of it. In my 2021 book, The Myth of Artificial Intelligence: Why Computers Can’t Think the Way We Do, I called these bugbears the “bottomless bucket” and “magical inference” problems. Bottomless, because no amount of hand-coded rules could capture dynamic real-world environments required for true AGI, and “magical,” because the only inference mechanism we had with a semantics—we know what will result from one proof or another; we know what’s valid or not—was some flavor of logic, either propositional or first-order, the latter including the possibility of infinite sets of elements using operators for existential and universal quantification.

But the world wasn’t a logic proof. I had barely finished my master’s thesis at UT Austin on the necessity of causal—nor formal or logical—knowledge in AI when I found myself in an office at Cycorp tasked with entering a bunch of logical statements into the Cyc knowledge base (KB). The statements were relatively simple and inane by design—a typical construction: “all living humans have heads”—as “Cyc” was supposed to be learning like a baby all the commonsense knowledge in the world.

So began all sorts of trouble, details of which I’ll spare the reader. The upshot? Cyc didn’t work, we all knew it, and it had become obvious by the time I joined that Lenat’s astonishing hypothesis was false. Just “adding” knowledge to a computer system wasn’t adding “knowledge” at all. It was adding logical statements that seemingly never filled in the gaps in what even a six year old knows, and as I’ve mentioned, even given adequate coverage of some domain—say, Steve Wozniak’s famous “coffee cup challenge,” where an AI enters a standard kitchen not seen before and makes a cup of coffee—the system had no way of getting to the right knowledge at the right time.

To me, this was just obvious. And yet in a hall of mirrors way, I was now spending my days in meetings with very bright philosophers and other shaggy headed visionaries and AI apostates explaining how we needed better “OE”—ontological engineering, my title was the absurdly pretentious “Ontological Engineer.” But we didn’t need “better OE.” We needed a better approach.

Add to this, by 2000 Google search was spreading through the halls of Cycorp and everywhere else, a kind of futuristic vision of AI that had nothing to do with hiring philosophers to code in first-order logic statements into a bloated and perpetually fragmented knowledge base (KB). Cyc had (foolishly, as it turns out) promised the moon in a competition for the best search, and was roundly defeated that year by a system using—wait for it?—shallow term-based machine learning methods. The writing was on the wall: Lenat’s astonishing hypothesis was false.

I was poisoned fruit at Cycorp, essentially, because I didn’t buy what Lenat and seemingly everyone else was selling, and the company was more like a cultish graduate program with strict rules about what could or could be said about AI. The top dogs there were typically philosophy Ph.Ds., not computer scientists, who would turn crafting first-order logical statements to feed Cyc into a game of intricate and complicated logic tricks. “Good OE” was, in other words, precisely the complicated logical origami that wouldn’t mean squat to the computer itself. It was all elegant B.S.

I teamed with a colleague turned friend Todd Hughes—later a DARPA program manager who helped fund my first company—and we worked on a “shallow” computer security application dubbed “Cyc Secure,” which, while Todd and I held the reins, didn’t even use the Cyc KB, except insofar as we had basic declarative statements about the components of a computer system, a network, and so on. We used a “shallow” planning algorithm to simulate or red team attacks on hypothetical networks. This, along with me bailing out on giving a damn about coding that living humans have heads and so on (I made terrible “OE”! For shame!), meant that I was transferred out of core engineering and in effect became a marketing guy for Cycorp’s just around the corner next-generation AI applications, that would surely change the world. Cyc Secure was one of them.

When the NASDAQ plummeted from its high of 5,000 in the Spring of 2001, and the DotCom bubble burst seemingly overnight, I was summarily fired by Doug, a hapless non-believer in a bad idea that was employing frumpy philosophers to wax eloquent about OE, but who had no real value outside of these niche approaches to AI. Many if not most of them couldn’t even program a computer. The folks who could program typically weren’t good at OE, either, and I surmise from my own experience that they likely didn’t give much of a damn either, a bad recipe for excelling at something declared by fiat to be of utmost importance to the company, itself declared to be of utmost importance to the field.

Doug passed away in 2023, and when I received the news I was sitting on a train going over the Alps toward Milan, where I was scheduled to give a talk about my book and the future of AI. The encomiums and panegyrics and eulogies poured in from the likes of the diaspora of former employees—"Cyclists”—and big players like Gary Marcus weighed in to inform newer generations about this powerhouse of AI, a man with a vision who rolled his sleeves up, took who knows how many millions of tax dollars to do “good OE,” without making really a dent in the core issues of AI. Don’t kick a man when he’s down, sure, and don’t talk poorly about the dead. Sure.

But Doug had fired me ignominiously, with a year old infant and a wife at home pregnant with our son. I never forgot that, even as I fully understood that not playing the game in a cultish clique shop like Cycorp was sure to spell employment trouble one way or the other. Still, my heart beat fast when I said publicly and in response to condescending reprisals from my former colleagues that Doug’s idea was bold but as it turns out entirely untenable, and anyway we were hardly friends as he’d fired me and left me hung out to dry (I ended up taking a job as a Data Analyst for Electronic Data Systems, programming finally, and making more money. But still.)

So here’s the problem in a nutshell. We need computers to have knowledge, to have access to some stable repository of knowledge, or in today’s parlance to possess some usable model of the world. In fact, many of the limitations of today’s language models or neural networks broadly, involve the lack of such robust modeling of the world. Scaling doesn’t work, it doesn’t get us to AGI that is, precisely because in some sense Doug was right all along: the systems relying only on data as input about the world are entirely too shallow, too brittle. But the answer was never to add hand-coded rules; a hopeless gambit from the get-go, and certainly after sixteen years of trying by the time I joined in 2000, the failure should have been stark and clear.

We lack robust models of physics and shapes and the world as it changes—and many of our benchmarks tests, like François Chollet’s General Intelligence Benchmark, or ARC-AGI—are intended to expose this lacuna in the system’s inferential capabilities. But when confronted with seemingly basic problems like how shapes are understood and manipulated, it doesn’t follow that we should try to spoon-feed the computer system with elegant logical statements from philosophers.

ARC-AGI tests for:

• Objecthood (this is a single thing: a square)

• Boundary vs interior

• Containment (inside vs outside)

• Completion / solidity (objects tend to be filled)

The Soul of the Machine, err, the Philosopher

I have to end with another I think well-deserved jab at what you might call “the soul of the philosopher” in AI. I’m punning off the classic book by Tracy Kidder, The Soul of the Machine (1981), the nonfiction narrative about a small team of engineers at Data General racing to build a new 32-bit minicomputer (the Eclipse MV/8000) under intense time pressure. (The ethos was, I think, also captured nicely by the AMC drama Halt and Catch Fire.)

Here, though, the Soul of the Philosopher captures the indefatigable conviction among “Ontological Engineers”—and we still have them. I know folks formerly with Cyc who have been employed by Amazon, by banks, and by all manner of organizations to organize and taxonomize information, an exercise closer to library science, but at any rate still useful and needed in many applications involving large repositories of information that reside in some database.

Amazon, for instance, maintains vast databases of products, all organized for fast retrieval and discovery. We could in principle simply use object oriented programming and a standard relational database to handle the knowledge repositories in knowledge-intensive applications or systems, but in some cases hierarchical or taxonomic organization can simplify the maintenance of such repositories, so that developers can easily access, add, delete, and update products and other items on the fly. No one really needs first-order logical elegance to do this, and by and large the idea that expressive knowledge representation languages are ipso facto superior to less expressive ones is another conceit of the Ontologist. But being able to represent knowledge is still a thing. Let me expound here.

When I was in graduate school I recall reading an early paper by Ron Brockman (formerly with DARPA) titled something like “the fundamental tradeoff between expressiveness and tractability in knowledge representation languages). Brockman was referring to the well-known problem of the inverse, you might say.

That is, a language capable of expressing diverse and fine-grained details about the world—not just that living humans have heads but ALL living humans have heads and if there exists a human without a head, then the human is dead, and not only these obvious statement but that SOME heads have hair, and hair is not in principle uncountable but in practice probably is, and that mouths typically have teeth but some mouths don’t, and so on—is typically what’s called intractable from the standpoint of computational complexity—a course I had suffered through in the CS department by the time Lenat offered me a job.

Intractability is measured in complexity classes; a well-known one is P, for polynomial time, and NP, for non-polynomial time. A famous conundrum in computer science and mathematics is whether P = NP: is there a proof that reduces non-polynomial time problems to the class of polynomial ones (there probably is not). NP-Hard or NP-Complete problems must remain toy problems, in effect, because as the search space grows it explodes exponentially. This means that Moore’s Law won’t be enough, and you’ll find yourself waiting entirely too long for a response from a system burdened by intractability in this sense. Smart developers avoid this by designing-away such complexity in the system in the fist place.

“Ontological engineers” are seemingly oblivious to this problem, and prefer the power of the increased expressiveness of a language, to pesky considerations of how it might be useful or not. There are, as you might imagine, a host of tricks to cut off programs from running headlong down an intractable path. But the basic tradeoff remains.

One particularly egregious example of this isn’t really computational complexity at all. For representation languages using first-order operators, like existential and universal quantifiers permitting in the latter case infinite sets of elements, you introduce what’s known as undecidability. This is a charming term that says what it seems to say—problems that are intrinsically undecidable, because they deal with the specter of infinity. But first-order logic (FOL), technically a semi-decidable language (details don’t really matter) is a favorite among philosophers eager to seem smart to their increasingly skeptical computer science colleagues. Oh, what the philosopher’s mind (and soul) can do with such expressive languages! Just think of the (perfectly worthless) expressions we can make, showcasing our acumen for logical precision! But the computer doesn’t care.

Using the full expressiveness of FOL is beautiful in the eye of the beholder—but the beholder is not the computer system tasked with making use of it, it’s the clever-minded philosopher “engineering” them. In fact, the situation is basically another tradeoff, as with Brockman’s early and memorable one. As we increase the expressiveness of the knowledge representation language, we can say more (express more) things about the external world. We can be more and more precise.

This seems like progress to the philosopher, who can’t be bothered with practical questions about how the inference engine will ever make use of this in real-world queries. Mostly, good knowledge representation takes the opposite of the philosopher’s approach: it seeks to simplify the expressiveness of the language down to the bare minimum required to say what needs to be said about the data or items in question.

If I’m newly hired by Instacart, and tasked with organizing all these damn vegetables and sauces, I’m not going to wax eloquent about the shape of the bottle of ketchup, or the taste differences between a Red Delicious and Fuji apple. My job is to simplify; not impress friends and neighbors with my logic skills, untethered as they’ll inevitably become from the practical requirements of knowledge-based systems development.

So it is with the misbegotten Soul of the Philosopher, a conceit that I would think today should be on life support, given the stunning success of all those shallow and brittle methods Lenat was once fond of execrating.

I’ll finish with this, also quasi-personal anecdote. At my last real tech job, as a Research Scientist at Knowledge Based Systems Incorporate (KBSI), I worked mainly as a python developer for natural language processing applications internal in the company R&D. It was, largely, fun. I like NLP/Information Extraction, I like wrangling text, and I like programming using libraries like SpaCy, using an ever growing list of Hugging Face models.

But here again, I bumped into my nemesis (not personally, I liked the guy), a died-in-the-wool old timer, a relic of the Lenat era still grumbling that machine learning was shallow and brittle, and how such-and-such knowledge representation language—say, RDF or RDF(S), or OWL—isn’t expressive enough to talk about so-and-so. Aye, here it is again. The Soul of the Philosopher. The well-meaning but too smug and unself-conscious enough blind spot where perfectly intelligent folks go on and on about expressivity, as if that aspect was somehow standalone, apart from the practical requirements of the system we’re building.

It’s hard to argue for using something simple when your opponent is steeped in a self-righteous fire about all these dastardly shallow systems taking over the world and extinguishing their stupid hypothesis. You’re arguing with someone who doesn’t want be corrected, even if demonstrably wrong.

I’ll end with this: a while back I wrote on LinkedIn—to an ex-Cyclist by the way!—that the problem is that what humans find elegant and powerful is not typically what a computer does, and so what clicks in the philosopher’s brain doesn’t transfer in any effective way to the computer system he or she is supposedly improving. To paraphrase the late great AI researcher John Haugeland, the problem with computers is that they just don’t give a damn.

Erik J. Larson

Stephen Pinker’s recent book, When Everyone Knows That Everyone Knows… Common Knowledge and the Mysteries of Money, Power, and Everyday Life, explores a subtle but fundamental feature of human communication known to psychologists and other researchers as: common knowledge. Pinker has a nose for these kinds of taken for granted gems in research and repurposing them for popular consumption. “When Everyone Knows…” is no exception.

“Common knowledge” in this not quite colloquial sense is beguiling at first blush because we’re well acquainted with the phenomenon yet often are surprised anyway when confronted with it operating in our everyday lives. It flies under the radar, we might say, which is odd given its ubiquity.

Common knowledge, in a nutshell is this: Many things we say only make sense because we assume not merely that others know something, but that each person knows that the other knows it—and knows that the other knows they know it, and so on.

This recursive structure is what philosophers and cognitive scientists call common knowledge.

A simple coordination problem illustrates the idea. Suppose two friends, Bob and Mary, lose cell service in a large city and must decide where to meet. There are dozens of cafés and restaurants they might choose from. The challenge is not merely selecting a location; it is selecting one that each expects the other to select as well.

Now imagine Bob and Mary agree to meet at the clocktower in the central square.

Why does this solve the problem? Because the clocktower is not merely known. It is mutually known to be known. Bob and Mary can reason:

• I know the clocktower.

• She knows the clocktower.

• I know that she knows the clocktower.

• She knows that I know she knows the clocktower.

This recursive awareness is what turns ordinary knowledge into common knowledge, and common knowledge is what makes coordination possible. If bomber pilot 1 sends coordinates for the target to bomber pilot 2, bomber pilot 1 must ALSO know that 2 received it, and so on. There’s an epistemic trap set here; we can’t eliminate the possibility that someone didn’t received the instructions, or that someone who did receive the instructions couldn’t confirm.

It smacks a bit of Zeno’s Paradox. But unlike that zinger, there are practical consequences to common knowledge, and the phenomenon I think bears on the issue of LLMs and AGI.

Read more

Pre-order here.

From the publisher:

Augmented Human Intelligence: Empowering Minds in the Age of AI (MIT Press)

A clear-eyed look at what AI actually is—and why we must focus on AI systems that augment human cognition, rather than mimic it.

From a seasoned AI researcher and a VC investor and entrepreneur.

Augmented Human Intelligence cuts through all the hype and noise around AI by returning to a central, often overlooked question: What is intelligence actually for? Rather than aiming to build autonomous superintelligent agents, AI leaders Erik Larson and Chee-We Ng argue in this book that we should focus on systems that support and extend human judgment, understanding, and decision-making—capacities that the current AI largely ignores.

Most state-of-the-art models rely on statistical induction at scale. They generate text, recognize patterns, and predict outcomes with remarkable speed—but without any grasp of meaning, consequence, or purpose. The authors of this book contend this is not merely a technical limitation, but a conceptual one: intelligence is not pattern-matching alone. It emerges from interaction with the world, guided by goals, feedback, and relevance.

This book aims to clarify the confusion around what AI is and isn’t, across everything from self-driving cars to robotics to conversational systems. In doing so, the authors hope to spur a shift in perspective—toward AI systems that augment human cognitive strengths rather than mimic human beings.

What makes this book unique is its insistence that judgment and understanding aren’t optional extras—they’re central to the very idea of intelligence. By reframing what we mean by thinking, the authors invite readers to imagine a different future for AI: one grounded not in simulation or spectacle, but in support for real human insight.

Get your copy, and lets empower minds in the age of AI.

Erik J. Larson

A Note to Readers

I don’t normally post on consecutive nights, but occasionally something comes together that feels worth sharing immediately.

Some background: About two and a half years ago I was invited to interview for a program manager position at DARPA, where I would have been working on problems related to artificial intelligence. At the last minute I had to withdraw from that process because I accepted a generous offer from the Thiel Foundation to write a second book. The terms of that contract required that I focus on the manuscript rather than taking on a full-time role.

That project has now worked its way through the long publishing pipeline and is under contract with MIT Press, with publication scheduled for later this year.

As I’ve wrapped up the manuscript, I’ve found myself turning more and more toward a class of problems that sit at the intersection of hardware, software, and adaptive systems. The ideas that follow emerged from that exploration. They grow out of a line of thinking I’ve started calling “the new cybernetics.”

The basic premise is simple: many problems we currently try to solve with ever larger AI systems may be better addressed through architectures built around sensing, feedback, and adaptive response.

Below is a short essay outlining that perspective. At the end of the piece I’ve attached a short concept paper describing one concrete application of this approach: a distributed sensor architecture for terminal defense against autonomous drone threats. The proposal is unclassified and is something I expect to discuss with colleagues working in the defense research community.

If any readers here have experience in these areas—or simply find the ideas interesting—I would welcome your thoughts. And since the concept paper is public, feel free to share it.

Now, on to the larger idea.

Toward a New Cybernetics

Over the past few years, most of the discussion about artificial intelligence has been dominated by a single idea: bigger brains. The field today is stuck to what’s called the “scaling hypothesis,” an idea that amounts to larger models, larger datasets, and larger compute clusters. In prior writing of I’ve referred to this semi-pejoratively as “Big Data AI.”

The implicit assumption is that intelligence emerges primarily from, yes, scale. If the system sees enough data and performs enough computation, useful behavior will eventually appear. But there is another tradition in the history of technology that is older, quieter, and in some ways more powerful. It comes from the field once known as cybernetics.

Cybernetics Primer

Cybernetics, pioneered by Norbert Wiener and W. Ross Ashby, was a movement that began in the mid-20th century and in many ways presaged the field of AI. Yet, cybernetics—from the Greek word κυβερνήτης, meaning to govern or steer—was not primarily about building machines that think. It was about understanding systems that adapt. Radar control loops, autopilots, thermostats, and biological regulation are all original examples. The key insight was simple but profound: intelligence often emerges not from centralized reasoning but from feedback between sensing, action, and environment. Cybernetics, a term coined by the avuncular mathematics professor Norbert Wiener, matured in the kiln of the technological pressures of World War II.

Wiener worked for on anti-aircraft technologies that now required mechanical solutions as the velocity and altitude of aircraft and later misses exceeded human performance. The work, though never fully adopted in the war, led to a class of adaptive systems that responded to environment feedback in ways that made technology newly adaptable and, yes, intelligent.

These adaptive technologies were not “big brains” with what sociologist Andrew Pickering has called a “representational ontology”—calculating everything as symbols in large representational systems—but rather closer to what later AI apostates like MIT’s Rodney Brooks and now many others would call bottom-up AI or robotics. Cybernetics fizzled; its insights are still alive and perhaps more important now than ever.

Over time this tradition faded from view. Artificial intelligence took center stage, and the dominant research strategy became increasingly clear: build systems that model the world internally through vast amounts of data.

Today that strategy has produced impressive tools. But it has also revealed its limits. Systems built around massive centralized models often run into scaling problems. They require enormous compute resources, vast training datasets, and complex infrastructure. In many real-world domains—transportation, infrastructure, defense, robotics—the question is not how to build a bigger brain. The question is how to build systems that adapt effectively in the world.

This is where I think a new cybernetics gets interesting.

The basic idea is to return to the original insight of cybernetics, while incorporating modern sensing and networking technologies—and “Big Data AI.” Only in this case, AI is no longer a panacea, the hammer because everything suddenly looks like a nail, but rather a component in a larger adaptive technological systems.

Instead of assuming that intelligent behavior must come from ever-larger models, we ask a different question:

How can relatively simple systems, equipped with the right sensors and feedback loops, solve problems effectively without requiring massive centralized intelligence? Once you start thinking this way, an interesting pattern appears. Many practical problems are not, in fact, failures of intelligence. They are failures of architecture.

Consider the emerging challenge of autonomous drone attacks. A common instinct is to respond by building more sophisticated detection systems: better radar, larger AI models trained to identify drones at great distances, increasingly complex surveillance infrastructure.

But another approach is possible. Instead of trying to detect drones everywhere, you can focus on the final few kilometers around high-value targets. Within that smaller zone, relatively inexpensive sensors—acoustic arrays, passive RF monitoring, optical cameras—can work together. Individually these sensors are imperfect. But together they can form an ensemble system that detects threats reliably enough to trigger a response.

The intelligence in this system does not come from a single powerful model. It comes from the structure of the feedback loop.

The same pattern appears in other domains. Take the mundane problem of winter traction on roads (I’ve discussed this on this Substack before). Drivers face a persistent tradeoff: studded tires provide excellent grip on ice, but they damage pavement and wear rapidly on dry roads. The result is a seasonal compromise that satisfies no one.

A cybernetic approach asks a different question: what if traction could be adaptive? Imagine a tire that deploys studs only when the road surface demands it and retracts them otherwise. The solution is not a massive AI system that models road conditions globally. It is a small adaptive mechanism that senses local conditions and responds in real time. Again, the intelligence lies in the feedback between sensing and action.

A third example appears in modern logistics. Much of today’s transportation infrastructure still assumes centralized optimization: vast routing systems that attempt to compute the best possible distribution of vehicles and goods across large networks. But increasingly, we see that adaptive local systems often perform better. Ride-sharing fleets, autonomous delivery networks, and warehouse robots frequently rely on local sensing and distributed coordination, where each agent responds to nearby conditions rather than relying on a single global model of the entire system. Intelligence emerges from the interaction of many simple components rather than from a single centralized brain. I am tempted to say here: “single centralized brain” thinking is not in fact smart. Yet modern generative AI can plug-in to systems that have been designed to be “smart.”

Seen this way, many of the problems we attribute to insufficiently powerful AI are actually design problems. We are trying to solve them with larger brains when we should be building better feedback systems.

This shift in perspective has practical consequences. First, cybernetic systems often scale more gracefully than centralized AI systems. Because they operate locally, they do not require vast amounts of data or compute to function effectively. We do not need nuclear powerplants to deploy correctly designed smart systems.

Second, they tend to be more robust. Distributed feedback systems fail gradually rather than catastrophically. If one component fails, others can continue functioning.

Third, they integrate naturally with the physical world. Real environments are noisy, uncertain, and dynamic. Systems that rely on feedback rather than perfect prediction are often better suited to these conditions.

None of this means artificial intelligence disappears. On the contrary, modern machine learning becomes one tool among many. But it is no longer the center of gravity. It becomes part of a larger design philosophy focused on adaptive systems. It is my belief that we need to move toward this new paradigm to not only innovate better AI, but to fully use the AI we have toward the best outcomes.

My thinking here represents a return to the roots of cybernetics, but now with twenty-first-century technology. Cheap sensors, embedded processors, and distributed networks make it possible to build adaptive systems that would have been impractical decades ago. We are in a position now to make good on some of the brilliant but largely forgotten ideas of the 20th century.

Welcome to the new cybernetics.

The central question of this new cybernetics is not:

How do we build machines that think like humans?

It is:

How do we build systems that behave intelligently in the world?

The answers to that question may not come from ever-larger models. They may come from carefully designed interactions between sensing, feedback, and action—systems that are less like giant brains and more like living organisms that continuously adapt to their environment. In mission critical domains, such systems are manifestly superior to very large and energy sucking models that are, after all, not fault or error tolerant. It’s not so much that they are black box technologies (they are), but that they are monolithic black boxes vulnerable to error and failure.

If this line of thinking is correct, it opens a large design space. Problems in transportation, infrastructure, defense, robotics, and energy can benefit from architectures that combine simple sensing with adaptive mechanisms.

The future of intelligent systems does not belong solely to bigger models.

It belongs to better feedback.

See a DARPA-friendly proposal I’ve whipped up, attached. Comments welcome.

Erik J. Larson

Well, I can’t attach it. Big Brain AI won’t let me, apparently. No matter here it is in simple text format:

Terminal Defense Sensor Mesh (TDSM): A Low-Cost Multi-Modal Detection and Cueing Architecture for Counter-UAS in the Terminal Defense Zone

Concept Paper

1. Problem Statement

Recent conflicts have demonstrated a growing vulnerability in modern air defense systems: the emergence of low-cost autonomous attack drones capable of traveling long distances and delivering payloads against high-value targets. These systems often use inexpensive commercial components, including GPS navigation and small internal combustion or electric propulsion systems. As a result, they can be produced at extremely low cost compared to the defensive systems designed to intercept them.

This has created a fundamental cost-exchange asymmetry in modern warfare. A one-way attack drone costing a few thousand dollars may require the use of an interceptor missile costing hundreds of thousands—or even millions—of dollars. When drones are deployed in numbers, this asymmetry becomes strategically unsustainable.

Traditional air-defense architectures were designed to monitor and defend large volumes of airspace against aircraft and missiles. However, small drones present a different challenge: they are slow, low-altitude, and often difficult to detect using conventional radar systems, particularly in cluttered urban environments.

Yet the operational requirement for many defended assets is simpler than wide-area surveillance. For most high-value targets—including military installations, energy infrastructure, ports, government facilities, and dense urban districts—the critical requirement is preventing drones from entering the final few kilometers surrounding the target.

This observation suggests a new defensive architecture: rather than attempting to detect small drones across wide areas of airspace, defense systems can focus on terminal defense within a limited perimeter surrounding high-value assets.

The goal becomes straightforward:

Detect, classify, and cue defensive action against drones entering a

defined terminal defense zone, typically within the final 5–10

kilometers of a protected site.

1.1 Terminal Defense Geometry

Defining a terminal defense zone dramatically reduces the sensing problem.

If a defended volume is defined by a radius R and altitude band H, the search volume scales with:

πR²H

Reducing the detection radius from 20 km to 5 km reduces the monitored area by roughly 16×, significantly lowering sensor coverage requirements.

Importantly, this range also corresponds to the distances at which inexpensive sensing modalities begin to function effectively, including acoustic detection and electro-optical imaging.

The appropriate radius should be chosen based on time-to-impact requirements rather than an arbitrary distance.

For example, if an incoming drone travels at approximately 40–60 m/s (typical of small cruise-type drones):

A 5 km perimeter provides roughly 80–125 seconds of warning

A 10 km perimeter provides roughly 3–4 minutes of warning

This interval is sufficient to support the defensive sequence:

detection → cue generation → interceptor launch → terminal acquisition → intercept

2. Concept Overview

We propose the development of a Terminal Defense Sensor Mesh (TDSM): a distributed network of inexpensive, heterogeneous sensors deployed around high-value targets to detect and classify drones approaching the protected zone.

The key insight behind TDSM is that no single inexpensive sensor can reliably detect small drones in complex environments, but multiple complementary sensing modalities can produce robust detection when combined using probabilistic fusion techniques.

Tripwires and Gate Guards

The TDSM architecture follows a simple defensive logic: tripwires detect intrusions, and gate guards engage the threat.

The system operates analogously to a human watchtower guard. A guard may first hear an approaching object before visually confirming its presence. Similarly, the TDSM architecture uses omnidirectional sensors as tripwires that detect potential threats, which then cue directional sensors for confirmation.

The sensor mesh produces:

- a probabilistic estimate of drone presence

- a coarse directional cue

- an approximate time-to-target estimate

These cues enable defensive systems—such as interceptor drones, electronic countermeasures, or directed-energy systems—to engage incoming threats within the terminal defense zone.

By concentrating sensing and response within the final kilometers surrounding a target, TDSM significantly reduces the scale and cost of the detection problem.

3. Technical Approach

3.1 Multi-Modal Sensor Ensemble

Each node in the TDSM network integrates several low-cost sensing modalities that function as complementary “senses.”

Acoustic sensing

Small drones produce distinctive acoustic signatures that propagate omnidirectionally. Distributed microphone arrays can detect these signals within several kilometers depending on environmental conditions. Acoustic sensing functions as an early tripwire for potential threats.

Electro-optical / infrared imaging

Fixed EO/IR cameras provide visual confirmation once a directional cue is available. Computer vision algorithms can distinguish drone shapes and motion patterns from birds and other objects.

Passive RF sensing

Some drones emit telemetry, control signals, or video transmissions. Passive RF sensors can detect these emissions without emitting detectable signals.

Compact short-range radar (optional)

Low-power radar systems may be included in selected nodes to improve detection reliability in poor visibility conditions and to provide track continuity.

Individually, these sensors produce incomplete signals. Combined as an ensemble, they form a robust detection system capable of operating in cluttered environments.

3.2 Evidence Fusion and Decision Logic

Sensor outputs are integrated using probabilistic fusion methods designed to combine weak signals from heterogeneous sources.

Each sensor produces:

- a probability estimate that an observed signal corresponds to a

  drone

- directional or positional information when available

- a confidence score reflecting signal quality

These outputs are combined to produce a posterior probability that a drone is approaching the protected zone, along with a coarse directional cue.

The decision process follows a sequential logic:

1. Tripwire detection Omnidirectional sensors detect potential drone

  signatures.

2. Cueing Directional sensors are automatically tasked toward the

  indicated sector.

3. Confirmation Visual or radar confirmation increases threat

  probability.

4. Response trigger Once the threat probability crosses a defined

  threshold, defensive effectors are cued.

This architecture balances rapid response with low false-alarm rates.

4. Operational Concept

The TDSM network is deployed around high-value assets or urban districts as a distributed mesh of sensor nodes mounted on rooftops, towers, or existing infrastructure.

The defended area is defined as a terminal defense zone, typically with a radius of approximately 5–10 kilometers.

The purpose of the system is not to detect drones throughout regional airspace but to ensure that any drone approaching the protected zone is detected and classified with high probability before reaching the target.

Within this zone, defensive effectors such as interceptor drones can engage incoming threats. This architecture transforms the counter-UAS problem from wide-area surveillance to perimeter defense, significantly reducing complexity and cost.

Incoming Drone(s)

  ↓

  ┌────────────────────┐

  │ Sensor Mesh Ring │ ← acoustic / RF / EO nodes

  │ │

  │ interceptor │

  │ ↑ │

  │ defended │

  │ asset │

  └────────────────────┘

  5–10 km terminal defense zone

Program Hypothesis

This concept paper advances the hypothesis that high-probability interception of small autonomous drones can be achieved within a 5–10 km terminal defense zone using inexpensive multi-modal sensors, probabilistic evidence fusion, and low-cost interceptor drones—without relying on traditional fire-control radar systems.

5. Technical Objectives

The proposed research program will demonstrate:

1. Reliable multi-sensor detection of small drones in complex urban

  environments.

2. Probabilistic fusion of heterogeneous sensor data into actionable

  threat cues.

3. Low false-alarm rates despite environmental clutter such as birds

  and urban noise.

4. Rapid cue generation within seconds of perimeter crossing.

5. Scalable deployment cost per defended square kilometer.

6. Demonstration Plan

The program will culminate in a live demonstration of the TDSM

architecture deployed around a representative defended site.

Testing will evaluate:

- detection of multiple drone types approaching the defense perimeter

- discrimination between drones and non-threat objects

- cueing of defensive interceptors

- system performance under urban noise and environmental clutter

Key metrics will include probability of detection, false-alarm rate, cue

latency, and engagement success rate.

7. Expected Impact

If successful, the Terminal Defense Sensor Mesh will provide a scalable

and cost-effective approach to defending high-value assets from

autonomous drone threats.

By leveraging inexpensive sensors and probabilistic fusion rather than

expensive wide-area surveillance systems, the architecture could enable

district-scale protection of critical infrastructure and urban centers.

This approach addresses the growing strategic challenge posed by low-cost autonomous drones while preserving favorable cost economics for the defender. Given the rapid proliferation of inexpensive one-way attack drones—first observed at scale in the Ukraine war and now increasingly across the Middle East and other conflict zones—cost-effective terminal defense architectures will likely become a strategic priority in the coming years.

MODEL-CENTRIC INFERENCE

Greetings,

I’m finally into copy editing phase with Augmented Human Intelligence (AHI), and so can turn more to writing about all things AI. Start here:

Why AI Can Simulate Abduction Without Experiencing Model Failure

One of the most interesting things about human reasoning is that it occasionally forces us to abandon the very assumptions we started with. Most of the time we reason within a framework: given certain premises, what follows?

But we frequently observe something that doesn’t fit what we expect. Charles Sanders Peirce pointed out over a hundred years ago that much of the world we encounter even day to day doesn’t quite “fit” in one way or the other, and so we resort to a type of inference he called abduction. We abduce when a surprising fact forces us to consider that one of our assumptions may simply be wrong. I used this example in my book: If the streets are wet we might infer that it rained, but if the sky is cloudless and the ground is still soaked we should start looking for another explanation—a broken hydrant, perhaps. What matters in these moments is not just that we produce a new explanation. It is that the surprise forces a shift in the underlying model we were using to make sense of the situation. The world is constantly disappointing us, which is to say, our models based on expectations. No matter; we abduce what might be true given that something surprising has been observed.

Read more

Hi everyone,

I just had this thought watching a movie titled Dead of Winter, with Emma Thompson.

Have a look at my prior posts on this in 2024:

Okay, so, the entrepreneur. The poor sad entrepreneur. I couldn’t figure out, back then, how to get the spikes deployed in all foreseeable states on the road.

See here from my prior post (see above):

Adaptive Claw Mechanism for Enhanced Grip
The core feature of the tire is a retractable claw system that extends automatically when slippage is detected. The mechanism operates based on the disparity between the rotational speed of the wheel and the vehicle’s actual velocity. Mathematically, when the angular velocity of the tire’s rotation (ω_wheel) exceeds the linear velocity of the vehicle (v_vehicle), it indicates wheel spin, suggesting that the tires are encountering a low-traction surface, such as ice, snow, or mud. This disparity serves as the trigger for the claws to extend, which will (quite drastically) improve traction.

Provisional Locking and Threshold Force Mechanism
Once the claws are deployed, they are held in a provisional lock to engage the tire’s contact patch—the small, but critical area of the tire in direct contact with the road. As the tire rotates, the claws make contact with the ground, and the system—I don’t want to give too much away here, but there’s a force measurement— measures the applied force on the claws. When the force (F_claws) on the claws exceeds a predefined threshold (F_threshold), indicating a hard surface like asphalt or concrete, the claws automatically retract, and lock back into place. This action is governed by a simple decision rule based on a force threshold:

If the force on the claws exceeds the threshold, the claws retract to prevent road damage.

This ensures that the claws retract when they encounter hard surfaces to prevent damage to the road. The claws remain deployed when driving on soft, low-traction surfaces (e.g., snow or ice) to provide maximum grip.

Empirical Calibration for Optimal Performance—Enter AI
The threshold force (F_threshold) is calibrated using empirical, or that is, real-world testing. The threshold is set to ensure that the claws retract only under sufficient road hardness. Specifically, the calibration ensures that the force required to retract the claws is above the force encountered on soft surfaces, such as ice or snow, but below the level that would damage paved roads. I will use empirical testing with this, but—here’s the AI part—it’s obvious that simulating various conditions to set an optimal threshold will be helpful. As far as AI is concerned, I envision training a very simple “low tech” decision tree to handle the myriad possibilities. I’m not using AI to communicate back to a web giant in Silicon Valley—I’m using it to make sure Dick and Jane get home safely. The model is not centralized. What I care about is a precise determination of the threshold.

The exact value of this threshold is determined empirically by subjecting the tires to various surface types under controlled conditions to strike the ideal balance between road protection and traction. The calibration process aims for the force range where road damage is avoided without compromising performance on slippery surfaces. In 1950, we could have done this in theory. Today, I can use AI to do it expertly and precisely.

The two transitions states (yes, this is a state transition machine) are:

1. When the tire’s angular velocity (rotation) exceeds linear velocity (speed of vehicle), your tires are slipping. You need more traction.

2. When the tire’s angular velocity goes to zero, while linear velocity remains non-zero, your tires are locked and the vehicle is still moving. You need more traction.

The problem with this design from an entrepreneur’s standpoint is that firstly, there’s no easy way to get the signal to the tires. Secondly, the Department of Transportation would not be happy if the advanced snow tire detected slippage on a summer road in July, when someone may be “burning rubber” or in other words just accelerating on either a dry surface or a wet road surface. We don’t want to tear up tax payer roads with our innovation.

The answer to the communications problem I’m still working on. The answer to the second question here, how to detect an anomalous state and therefore the system will not deploy spikes into the road surface (if it’s not ice, but dry or wet and you’re “peeling out”) is solvable straightforwardly.

As an empirical matter, the foot pounds of energy to breakage on ice versus pavement will always be different. I don’t know what the number are, but ice will break under steel or titanium penetration long before a road surface will. This will be true even in the case of fresh asphalt (it may not be true of unpaved roads, but then the DoT will be less likely to care).

The edge cases involve tearing up the road, in either of the above central use cases.

The mechanism is therefore discreet, or thresholded that is, and the force on what’s called the traction patch (the part of the tire touching the road surface) under some determined amount will either permit or disallow deployment.

This is handled “cybernetically,” so to speak, at the point of contact and by feedback from the pressures the system is undergoing at the wheel.

How it ties into AI with further clarification of state machine:

The state machine is slightly richer than two states

I identify four:

1. Normal traction

2. Slip detected (candidate state)

3. Surface confirmation phase (micro-probe test)

4. Claw deployed

And the crucial addition:

5. Rapid retract override (hard surface detected)

That confirmation phase prevents road damage and eliminates burnout false positives.

The AI component

This is not a cloud AI problem.
It is a tiny adaptive classifier running on local signals.

We could use:

• decision tree

• logistic classifier

• small reinforcement threshold tuner

The learning problem is simply tuning stiffness / penetration thresholds across:

• temperature

• tire wear

• surface mixtures

• vehicle weight

This is exactly the sort of bounded, local adaptation where simple ML shines.

See? AI can be useful, and it doesn’t have to be Big Data/Cloud AI.

Have a wonderful weekend! I love this stuff.

I wish it came to me more often, but it comes when it comes.

Erik J. Larson

P.S. I also have a workable idea, I believe, for a better algorithm for surfacing content on Substack. Working on that.