Hinton leaves Google to warn about AI
The man who trained the field's networks for forty years resigns so that he can say, freely, that he now thinks they may become smarter than us and that he regrets part of his work.
what had to happen · 51 events back to 1943
Every event this one built on, transitively, in order. Direct influences are marked.
00 · One neuron · 1
I · Foundations · 11
- 1948A mathematical theory of communication
- 1949Cells that fire together wire together
- 1950Programming a computer for playing chess
- 1950Computing machinery and intelligence
- 1958The perceptron learns
- 1959Samuel's checkers program coins 'machine learning'
- 1960ADALINE and the least-mean-squares rule
- 1965Moore's law
- 1966ELIZA
- 1969Perceptrons
- 1970Reverse-mode automatic differentiation
W1 · The first winter · 2
II · Connection · 3
W2 · The second winter · 6
III · Statistics and data · 9
IV · Deep learning · 8
V · Transformers · 8
- 2017Attention is all you need
- 2017Deep reinforcement learning from human preferences
- 2018GPT: generative pre-training
- 2019GPT-2 and the model too dangerous to release
- 2019The Turing Award goes to deep learningdirect
- 2020Scaling laws for neural language models
- 2020GPT-3
- 2020Learning to summarise from human feedback
VI · Everyone · 3
- 2022InstructGPT
- 2022ChatGPT
- 2023GPT-4direct
Geoffrey Hinton had joined Google in 2013 when it bought the company he had formed with his two AlexNet students, and had spent ten years there, working latterly on capsule networks and on the distillation method that makes large models small. On 1 May 2023, aged 75, he told the New York Times he had resigned so that he could speak about the risks without it reflecting on the company. He had changed his mind, he said, about how far off the danger was. The models had turned out to learn more than he expected from less, and the digital form of learning, in which copies share what they learn instantly, might simply be better than the biological one.
He listed the risks in order: misinformation, jobs, autonomous weapons, and, further out, systems that pursued goals of their own. Asked about regret, he said he consoled himself with the thought that if he had not done it, someone else would have.
The resignation made the safety debate impossible to dismiss as outsiders' alarm. The scientist most responsible for deep learning was now among its most prominent worriers, and eighteen months later the Nobel committee gave him the physics prize and he used the acceptance speech to say the same thing.
what it led to · 0 events downstream
A leaf, for now. Nothing in the archive has built on it yet.