The Teacher, Part 4: when the teacher began to warn
Part 4 of 5 · The Teacher
There is a version of the great-scientist story where the hero builds the marvel, takes the prize, and rides off. Geoffrey Hinton did not take that version. In the spring of 2023, at seventy-five, he left his job at Google — where he had worked for more than a decade — specifically so that he could speak freely about the dangers of the thing he had spent his life building. That is the turn this part is about, and we think it is the most important thing he ever did.
Why he left, in his own words
The departure was reported first by The New York Times on 1 May 2023. Hinton told the paper that part of him regretted his life’s work, adding, with characteristic dryness, “I console myself with the normal excuse: If I hadn’t done it, somebody else would have.”
He also described how badly his own timeline had collapsed. For most of his career he assumed machines overtaking human intelligence was generations away. “I thought it was 30 to 50 years or even longer away,” he told the Times. “Obviously, I no longer think that.” A man who had spent fifty years being more optimistic about these machines than almost anyone had suddenly become more worried than almost anyone — not out of ignorance, but because he, more than most, could see what was coming.
It is worth being precise about one thing, because he was precise about it. Hinton did not leave to attack his former employer. After the Times story ran, he clarified publicly that he had “left so that I could talk about the dangers of AI without considering how this impacts Google,” and that “Google has acted very responsibly.” He wanted the freedom to warn, not a grievance.
What he is actually worried about
Speaking to MIT Technology Review the next day, he laid out the shape of his fear in plain language. These systems, he argued, are “a completely different form of intelligence. A new and better form of intelligence” — not a toy imitation of human thought but something genuinely new, and in some ways already ahead. He pointed out that a large model “knows hundreds of times more than any one person does. So maybe it’s actually got a much better learning algorithm than us.”
And he was honest, almost disarmingly so, about his own state of mind. “I have suddenly switched my views on whether these things are going to be more intelligent than us,” he said. “I’m mildly depressed. Which is why I’m scared.”
His worries are not science fiction and he is careful not to let them read that way. He has warned about misuse by bad actors; about a coming flood of fabricated images and text that could leave ordinary people unable to tell what is true; about a competitive race between companies and nations that pushes capability faster than anyone’s ability to keep it safe. In the same MIT interview he used his famous example of how different this intelligence is from the symbol-shuffling old guard imagined: “Crows can solve puzzles, and they don’t have language. They’re not doing it by storing strings of symbols and manipulating them. They’re doing it by changing the strengths of connections between neurons in their brain.” The point — that real intelligence is learned connections, not written rules — is the same one he spent fifty years proving. He simply follows it all the way to its uncomfortable end.
Why this matters to us specifically
We build tools, and in the grand opening this series sits inside, we are handing some of them to strangers and telling them to have fun. That is a generous posture, and it is also a dangerous one if the maker is not honest about the edges. Hinton is the model of the honest maker. He did not hide behind the prize. He did not let his authorship make him the technology’s defender. He used the credibility he had earned building the thing to tell the truth about the thing, and he paid for that freedom with his job on the way out.
We take two commitments from his example, and we would rather write them down publicly than keep them private. First: the people we hand tools to deserve the whole truth about them, including the parts that should frighten them — anything less is not a gift, it is a trap. Second: the fact that something can be scaled is never, by itself, a reason to scale it; capability and judgement are different things, and the maker owes the second one. The cognition behind our tools is provided by Obsidian Delta, and the discipline of saying plainly what a tool can and cannot safely do is not an afterthought to that — it is the condition of handing anything to anyone at all.
A teacher who will not tell you the dangerous truth about the subject is not protecting you. Hinton told it, loudly, at cost. That is the part of his record we most want to be worthy of.
This is Part 4 of five. Next, the last one: what a teacher is actually for — and why the whole of our mission is an answer to that question.
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, Google, The New York Times, or MIT Technology Review. Every quote here is attributed to its public source as reported; nothing is put in anyone’s mouth.