Why Is A 60-Year-Old AI Argument Suddenly Everywhere?

June 8, 2026

Why Is A 60-Year-Old AI Argument Suddenly Everywhere?

By Gary S. Stager, PhD & Sylvia Martinez

In 1965, Hubert Dreyfus, a philosophy professor at MIT, wrote a paper arguing that artificial intelligence was heading toward a dead end. He claimed the entire enterprise was built on a false assumption about how human minds work. Dreyfus asserted that computers, no matter how powerful, would never be able to do the things that made human intelligence genuinely human. The paper circulated as a RAND Corporation memo, was quoted in the New Yorker, and caused a sensation in computer science departments across the country.

According to legend, researchers at MIT’s AI lab began refusing to eat lunch with him. He became, in the words of people who were there at the time, professionally radioactive.

Sixty years later, the Dreyfus bullying story has been resurrected with something close to glee, It is circulating on social media, passed around in faculty meetings, cited in ed-tech skeptic newsletters, picked up by journalists, philosophers, and artists as much as by educators. For anyone uneasy about the current AI moment, Dreyfus offers something valuable: a contrarian view, MIT affiliation, and a reminder that someone else was skeptical about AI during an earlier wave of confidence, accompanied by a tale of professional martyrdom. The moral of the story may be reduced to “He saw it sixty years ago, spoke out, and was punished for his prophecy.”

The Dreyfus story is true but more complicated than the current clickbait version. The background and energy behind its revival is instructive for educators trying to think clearly about AI in the classroom today.

What Dreyfus Actually Argued

Dreyfus’s 1965 paper, Alchemy and Artificial Intelligence, surveyed the four main areas of AI research at the time: game playing, problem solving, language translation, and pattern recognition. In each area, he identified the same pattern — dramatic early success with simple problems, followed by mounting difficulty as tasks became more complex. His diagnosis of why this kept happening was philosophical.

The whole field of AI, he argued, was built on a false assumption: that human intelligence works by processing discrete, explicit rules and that computers, which do exactly that, should eventually be able to replicate it. Dreyfus said this was wrong. Human intelligence doesn’t actually work this way.

Instead, he pointed to capacities he believed were uniquely human and fundamentally unprogrammable — the peripheral awareness that guides attention without making everything explicit, the ability to instantly sense what matters in a situation, and the capacity to act without resolving every ambiguity in advance. He argued that human beings are not information processors moving through the world. We are embedded in it and oriented by it, in ways that cannot be captured by any formal system.

Why the Story Is Resonating Right Now

The Dreyfus story fits this cultural moment. A lone thinker challenges the established consensus of a nascent field, is socially scorned for his position, but is eventually vindicated by history. The lunch table snubs at MIT resonate with any educator uneasily watching AI enthusiasm blow through their schools with the force of settled fact and inevitability.

Discovering Dreyfus feels like finding confirmation that hesitation is justified. That skepticism about AI has a philosophical lineage and “they,” the very people who were supposed to know best, suppressed Dreyfus’s whistleblowing warnings from us for decades. History is repeating itself. Right?

The rediscovery of Dreyfus has the tone of a conspiracy theory — the truth was there all along. In a cultural environment saturated with AI enthusiasm, venture capital, breathless press releases, ed-tech vendors promising the moon, citing a philosopher who predicted the current difficulties in 1965 and paid a professional price for it feels like a serious and well-earned vindication.

Seymour Papert’s Response

In 1968, Seymour Papert, a mathematician at MIT’s Project MAC and one of the leading AI researchers of his generation, wrote a detailed rebuttal to Dreyfus titled The Artificial Intelligence of Hubert L. Dreyfus: A Budget of Fallacies. Papert would later become one of the most important education thinkers of the twentieth century — the creator of the Logo programming language, the father of constructionist learning theory — a man who spent his career arguing that children learn best by making things and that computers could be powerful vehicles for exactly that kind of learning. Papert possessed both the technical knowledge to engage Dreyfus on AI and the educational philosophy to understand what was at stake. He was also unafraid to speak clearly and courageously against injustice. Papert found the arguments of Dreyfus and the next two generations of similar critics not only wrong, but bad for humanity. No one asked Papert to write such a scathing “memo,” dissecting the views of a fellow academic. The reputational risk to Papert may have been more severe than finding it difficult to secure lunch companions.

Papert began his paper by identifying what he saw as the central danger of the Dreyfus phenomenon — not just the bad scholarship, but the reason people were willing to overlook it. “So many people praise his papers because they like his conclusions,” he wrote, “and show no concern for the quality of his arguments. The conclusions are banal: the taxi driver and my maiden Aunt Agatha believe them as firmly and express them as well.” This observation cuts directly to why Dreyfus is popular again right now. Liking where an argument leads is not the same as the argument being sound, and in moments of rapid technological change, the two are especially easy to confuse.

Papert’s technical objections were specific and often devastating. The gloves really came off. He showed that Dreyfus had repeatedly misunderstood or misrepresented the AI programs he was criticizing. Dreyfus’s description of what chess programs actually did was simply wrong, and his account of machine translation research ignored work being done within walking distance of his office. Papert was particularly pointed about Dreyfus’s tendency to evaluate entire research programs by their earliest and most primitive results. “The amateur scientist starts from the premise that the purpose of Artificial Intelligence is to make intelligent artifacts,” Papert wrote, “and then judges each piece of work by whether it has, in itself, actually produced intelligence — as if one were to judge workers in aerodynamics by their ability to fly.”

What makes Papert’s response so useful now is that he identified specific logical errors that are just as common in today’s AI debates as they were in 1968.

The Unthinkable Fallacy

Dreyfus repeatedly looked at problems AI had not yet solved and concluded not merely that we didn’t know how to solve them, but that they could not in principle be solved. Papert named this the Unthinkable Fallacy — “declaring a formal procedure to be impossible for no better reason than that one cannot think how it could be carried out.” The gap between “this is very hard” and “this is impossible” is enormous, and Dreyfus crossed it repeatedly. The same move appears constantly in current AI debates, among both enthusiasts who assume current capabilities will scale indefinitely and skeptics who treat current failures as permanent limits.

Papert made an observation about AI criticism that applies with equal force today: “Almost everyone, whether he has seen a program or not, has strong convictions about what computers will never do. Ironically, the arguments used to show that they cannot emulate human intelligence follow, with machine-like regularity, a small number of standard patterns.” And: “Intellectuals have the advantage of a technical vocabulary for the discussion of human thought and of machines. But, instead of helping them see the real problems posed by the project of creating intelligent automata, their greater sophistication often leads them all the more securely into the same misconceptions.”

What Papert was pointing out is that the conviction that something is impossible tends to feel like genuine understanding. One can find great comfort if blessed with a limited imagination. The person who cannot imagine how a computer could ever truly recognize a face, or grasp the meaning of a sentence, is not displaying ignorance, they are experiencing certainty. The person who has more conceptual resources for describing human cognition is exactly the person who is more confident they can explain why machines will never replicate it. Sophistication and confidence do not protect against this fallacy; they make it more likely.

What History Did and Didn’t Vindicate

The current Dreyfus revival requires some careful management of inconvenient history. The things Dreyfus said could not be done, have been done. Chess computers defeated the world champion. Machine translation became genuinely useful. Pattern recognition surpassed human performance across a remarkable range of tasks.

Papert distinguished carefully between what he called Turing Barriers: absolute limits on what any finite machine can do, and Theory Barriers: limits set by what we don’t understand now rather than by any fundamental impossibility. “Theory Barriers are set by our current state of knowledge,” he wrote, “the limitation is in us, not the machine.”

A Turing Barrier is a genuine, permanent ceiling. There are things no computer can ever do: certain mathematical problems are provably unsolvable by any algorithm, no matter how powerful the machine. These limits are real and mathematically precise. A Theory Barrier is something different entirely. It is the edge of what we currently know how to do, which can look, from the inside, exactly like a permanent wall, but is actually a frontier that moves as knowledge advances.

In 1965, getting a computer to recognize a human face looked like it might be a Turing Barrier — something so dependent on intuition, context, and embodied human perception that no formal procedure could ever capture it. It turned out to be a Theory Barrier. Once researchers developed better mathematical tools for representing images and trained systems on large amounts of data, face recognition became not just possible but routine. The same story played out with chess, translation, medical image diagnosis, and a long list of tasks that experts at various points declared to be permanently beyond the reach of machines.

Dreyfus kept mistaking Theory Barriers for Turing Barriers — seeing the boundaries of current knowledge and calling them the boundaries of what was possible. This is an easy mistake to make, because from where you are standing, you genuinely cannot see past the current limit. But the cost of this mistake is that it encouraged people to stop working on problems that were in fact solvable, and it gave intellectual respectability to a kind of helplessness about what machines could ever do.

Someone who looks at the current failures of large language models, their confabulations, their inability to reliably reason through novel problems, their sometimes-baffling errors, and concludes that these failures reveal a permanent ceiling is making the same mistake Dreyfus made. The history of the field gives us strong reasons to refrain from drawing final conclusions, and to distinguish carefully between “we don’t yet know how to solve this” and “this cannot be solved.”

What Dreyfus got genuinely and durably right was a philosophical point about the nature of embodied understanding. Human cognition is not best modeled as explicit rule-following. There is something important in tacit knowledge, peripheral awareness, and contextual engagement that is very hard to formalize. His critique that AI researchers were too quick to assume their early successes in clean, well-defined domains would transfer smoothly to the messy open-endedness of everyday life remains relevant.

However, there is a difference between the argument of Dreyfus and the story of Dreyfus. The argument deserves serious engagement. The heroic narrative, featuring the misunderstood prophet, the vindicated heretic, and the secret the establishment suppressed, is providing emotional validation for positions that many people have already arrived at for other reasons. That is precisely the pattern Papert warned about in 1968; people praising conclusions while remaining indifferent to the quality of the arguments supporting them.

If Dreyfus’ critique of AI had any validity, it was in relation to the symbolic approach to AI dominant more than a half century ago. Relying on his arguments to reject generative AI, or whatever may succeed it, is disingenuous.

What Educators Actually Need

Despite his limits as a prophet, Dreyfus made a distinction that educators will recognize immediately: there is a difference between knowing something and knowing how to do something, and the second kind of knowledge doesn’t reduce to the first. Expertise isn’t a large set of rules you’ve internalized. It’s something built through experience, often impossible to fully articulate, and inseparable from the struggle of acquiring it. That distinction matters in the classroom today. When a student uses AI to produce an essay, the question isn’t only whether the output is good. It’s whether the iterative, frustrating work of trying to put thought into words is where a certain kind of learning actually occurs and whether handing that work to a machine bypasses the process by which the knowledge gets constructed at all.

Dreyfus provides a philosophical vocabulary for asking such questions. Papert gives us the logical discipline to ask them carefully. “Our culture is indeed in a desperately critical condition if its values must be defended by allowing muddled thinking to depose academic integrity,” Papert wrote in his response to Dreyfus. That warning is as important today as in 1968, but even more urgent since AI is now in everyone’s hands, not just an academic pursuit.

The sad Dreyfus lunch table story resonates because today’s “AI experts” dismiss the genuine anxiety of people worried about an uncertain future. But Dreyfus was also, on many specific points, demonstrably wrong. The seemingly intractable barriers he identified have been continuously overcome. The vindication narrative may be emotionally satisfying, but its virality does not make it either true or useful.

The optimism of AI pioneers and visionaries like Seymour Papert is hard to find amidst the noise generated by generative AI. Those seeking a humane, creative, empowering future of learning with artificial intelligence should seek comfort in work and powerful ideas of Dr. Papert. The roots of the MIT AI tree are deep in Jean Piaget and progressive education traditions. What I find most startling about this particular moment is the alacrity with which novices and nincompoops switched their LinkedIn bios overnight to anoint themselves as “expert in AI and education.”

Artificial intelligence is a legitimate scientific domain. Would these folks announce themselves as neurosurgeons, astrophysicists, or geneticists? I would love to ask Hubert Dreyfus for a hypothesis about how and why AI is such fertile ground for charlatans and hucksterism. The fact that school leaders take such instant experts seriously exposes a willing gullibility that is ultimately corrosive to the enterprise educational enterprise.

The most useful thing an educator can bring to this moment is not a predetermined verdict on AI, neither hype or hysteria, but the ability to distinguish between a strong argument and a satisfying one, and the patience to keep asking hard questions even when the cultural pressure is to choose a side. Since Piaget teaches us that “knowledge is a consequence of experience,” it couldn’t hurt for teachers to begin messing about with AI to understand it better themselves.