The Teacher: the man who taught machines to learn

This is the start of a short series, and it is a tribute. We did not build anything with anyone’s blessing, and we claim none. What we have is a debt, and we would rather name it out loud than pretend our tools fell out of the sky. The free-learning mission this series sits beside — Meta DAX, learning handed to every human being, any subject, for nothing — runs on a kind of machine that did not exist when most of us were born. Someone had to teach machines to learn. For the better part of fifty years, more than anyone else, that someone was Geoffrey Hinton.

We are not affiliated with him, endorsed by him, or partnered with him in any way. Read the last line of every piece in this series. This is our reading of a public record, tied to what we are trying to give away.

What the record says

On 8 October 2024, the Royal Swedish Academy of Sciences awarded the Nobel Prize in Physics to John Hopfield and Geoffrey Hinton, in the Academy’s own words, “for foundational discoveries and inventions that enable machine learning with artificial neural networks.” It was an unusual prize — a physics medal for work most people file under computer science — and that oddity is part of the point. Hinton had reached into nineteenth-century statistical physics to build a machine that could learn, and the physicists decided, four decades later, that this counted as physics.

Five years earlier, in 2019, he had already received the other great prize in his field. The Association for Computing Machinery named Yoshua Bengio, Geoffrey Hinton, and Yann LeCun the recipients of the 2018 A.M. Turing Award — the “Nobel Prize of Computing” — “for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing.” Two of the highest honours a scientist can hold, for one idea, pursued with a stubbornness that for most of his career looked like a dead end.

The idea almost nobody believed

The idea is simple to say and was, for decades, deeply unfashionable. Instead of programming a computer with explicit rules for every task, you build a network of simple units loosely modelled on neurons, and you let it learn — adjusting the strengths of the connections between those units until the network does the thing you wanted. No rules written by hand. Learning by example, the way a child learns, the way, Hinton came to argue, a brain has no choice but to work.

For most of the 1970s, 80s, 90s, and into the 2000s, the mainstream of artificial intelligence thought this was a waste of time. The money and the prestige were on the other side — symbols, logic, rules. Hinton kept at the unfashionable thing anyway, and he taught others to. When the approach finally broke into the open in 2012, it did not trickle; it flooded. Computer vision, speech recognition, translation, the systems that now sit behind almost everything called “AI” — nearly all of it traces back through work he did or work he supervised.

He is also disarmingly honest about how credit works. Describing the 2012 breakthrough that changed everything — a project built by two of his graduate students — Hinton summarised the division of labour for the Computer History Museum this way: “Ilya thought we should do it, Alex made it work, and I got the Nobel Prize.” We will come back to those two students. One of them went on to help start one of the most consequential labs in the world.

Why we are writing this

Because our mission is learning, and he is the person who taught machines to learn. Meta DAX is an attempt to invert the oldest ratio in education — to put far more teaching within reach of far more people, free, forever, built on paid compute — and it is only possible because the underlying machines can now be taught rather than hand-programmed. That capability is his legacy more than any single other person’s. The cognition behind our own tools is provided by Obsidian Delta; the lineage of the science that makes any of it work runs straight through his lab.

There is a second reason, and it is the one that makes this series more than a thank-you note. Having spent his life building these machines, Hinton spent the last few years of it warning the world about them — loudly, specifically, and at real personal cost. A teacher who tells you the truth about the thing he taught you, including the dangerous parts, is the only kind worth having. We take that seriously, because we are handing powerful tools to strangers and telling them to have fun. We had better mean the safety part too.

So over the next four parts we are going to take the record seriously, one piece at a time: the two inventions that started it; the 2012 moment and the students who made it; the turn, when the teacher began to warn; and finally what a teacher is actually for — and why that question is the whole of what we are trying to do.

This is Part 1 of five. He taught machines to learn. We are trying to use that to teach people, for free. Those are not the same project, but one would be impossible without the other, and we think the debt is worth saying out loud.

---

This is a tribute and our own commentary. The DAX Foundation and Obsidian Delta are not affiliated with, endorsed by, or in partnership with Geoffrey Hinton, the University of Toronto, Google, the Royal Swedish Academy of Sciences, or the Association for Computing Machinery. Every quote here is attributed to its public source; nothing is put in anyone’s mouth.