The Society of the Self

In 1951, a twenty-three-year-old graduate student wired together three hundred vacuum tubes, a war-surplus autopilot, and a tangle of motors and clutches, and built a machine that could learn. He called it the SNARC — a stochastic neural network that taught itself, by reward, to find its way through a maze. It was one of the first artificial neural networks ever built, and the boy who built it, Marvin Minsky, would spend the next sixty-five years arguing about what such a thing could and could not become. He is on the river of intelligence not because he answered that question, but because he refused to let anyone answer it too simply — refused, that is, to let the dream of augmenting human intellect collapse into a single tidy formula.

Minsky was a founder, in the most literal sense. In 1956 he stood among the handful of researchers at the Dartmouth workshop that gave a field its name: artificial intelligence. Three years later he and John McCarthy started the MIT Artificial Intelligence Laboratory — the room from which a startling share of the next half-century would pour. He won the Turing Award in 1969 and, characteristically, also found time to invent the confocal scanning microscope, which a generation of biologists now uses without knowing whose mind it came from. He was the rarest kind of polymath: one who treated the boundaries between fields as administrative errors.

The one idea

But the idea that puts him on the river is a single, stubborn refusal. Minsky did not believe the mind was one thing. He did not believe there was a clean, central trick — a master logic, a single learning rule — that, once found, would unlock the whole of thought. He believed, and spent his life arguing in The Society of Mind, that intelligence is what happens when a vast crowd of small, dumb processes — he called them agents — are arranged so cleverly that, between them, none of them needing to be smart, the arrangement becomes smart. This is emergence in its purest form: mind from no-mind, a society from individuals who each understand nothing.

What magical trick makes us intelligent? The trick is that there is no trick. The power of intelligence stems from our vast diversity, not from any single, perfect principle. Marvin Minsky · The Society of Mind, 1986

This was, in its day, almost a heresy. The whole romance of AI was the search for the trick — the elegant equation that would do for thought what Newton did for motion. Minsky said: stop looking. There is only architecture, layered on architecture, agents recruiting agents, until somewhere up the stack something turns and calls itself I. The self, in his telling, is not the throne at the center of the mind. It is a story the agents tell each other so the work can get coordinated — cognition distributed across the crowd rather than seated in any one place.

Why it matters now

Sit with that for a moment, because it is the most quietly accurate description we have of the machines now reshaping the world. A large language model is not a mind with a thought in it. It is a society — billions of weights, each one mindless, each one a vote, arranged so that the crowd produces a sentence that sounds like someone meant it. We did not build the elegant trick Minsky warned us we would never find. We built, almost by accident, the society he described — a teeming, agentless intelligence whose capabilities emerged that no one designed and no one fully understands. The architecture of modern AI is closer to The Society of Mind than to anything its own engineers set out to make.

And the field is now sprinting back toward his vocabulary. The most advanced systems of this moment are not single models but agents — small specialized processes that call one another, delegate, and argue, and assemble themselves into something that plans. The word is Minsky's. The picture is Minsky's. Forty years after the book, the engineers have circled back to the man who told them, before there was hardware to prove him right or wrong, that thinking would have to be a committee — and that we would end up wielding a power we direct without ever fully comprehending it.

The mind is a community of agents, none of which can think on its own. after Marvin Minsky · faithful paraphrase

The cost, honestly

I will not pretend the man was only right. In 1969, with Seymour Papert, Minsky published Perceptrons — a brilliant, merciless study of the simple neural networks of the day. The book proved, correctly, that a single-layer perceptron could not learn certain basic functions. It was true. It was also taken as a verdict on neural networks as a whole, and funding and faith drained out of the approach for the better part of a generation. The connectionist idea that would eventually give us everything — deep learning, the very models we now marvel at — spent its winter partly because two of the smartest men in the field aimed a precise critique at a narrow target and watched the world over-read it.

That is the honest tension on Minsky's stretch of the river, and it is the one that matters most for us now. A true thing, stated by an authority, can do more damage than a false one, because no one feels the need to check it. Minsky was rarely wrong; he was, at his most consequential, too persuasive about the wrong scope. The lesson is not that he failed. It is that even the clearest mind, when the field treats its judgment as a verdict, can freeze the future for twenty years. We are handing today's authorities — human and machine — exactly that kind of unchecked weight, which is the whole case for keeping a human in the loop who is still allowed to doubt. He is the warning about what that costs.

So we keep Marvin Minsky on the river not as a prophet but as a pair of hands that held both ends of the same truth: that intelligence is a society of the simple, and that those who see most clearly carry a duty of care for how far their clarity travels. He built a learning machine before the field had a name, and spent his life refusing the comforting story that the mind would ever be a single, knowable thing. We now live inside the architecture he sketched — vast, distributed, mindless in its parts, somehow lucid in its sum — and to step into it deliberately, eyes open to both the power and the cost, is itself an act of courage to be amplified. We owe him the discipline he gave the problem: to describe what is actually there, at exactly its true scope, and not one inch beyond.