Portrait of Geoffrey Hinton by Stable Diffusion
The rise of the LLMs
This is the way things stood before some recent events that led Geoffrey Hinton to make a disturbing statement. Geoffrey Hinton is the mastermind behind Large Language Models such as ChatGPT, Claude, Gemini or LLaMA3. In the press he goes often under the nickname of the “godfather of LLMs” and not without good reason. As proof, that in the two-volumed book that launched the connectionist approach of which LLMs are the current flagship, published in 1986 under the title “Parallel Distributed Processing” 1, out of the 26 essays in the two volumes, six are co-authored by Geoffrey Hinton.
Here is the unsettling remark Hinton had made: “I suddenly realised we might have something rather different, that was already better”, hinting at the fact that what had happened had taken place inadvertently.
Hinton was responding to the following question of the host:
“As you thought about how computers learn. Did it go the way you thought it would when you started in this field [artificial intelligence]?”. And here is what he said: “It did, until very recently, in fact. […] And all that was going on very well. And then very suddenly, I realised recently that maybe the digital intelligences we were building on computers were actually learning better than the brain. And that sort of changed my mind. After about 50 years of thinking we will make better digital intelligences by making them more like brains. I suddenly realised we might have something rather different that was already better” 2.
In Hinton’s mind, the task he had devoted his life to had gone astray: he was aiming for a type of artificial intelligence that would match that of Homo sapiens and, before he realised what had happened, he had created an intelligence superior to the original. He immediately resigned from the position that he held of chief scientist at Google Brain, Google’s AI division.
What further de-centring of the image of “us” did this event induce, of an artificial intelligence of a quality apparently higher than the natural one that was meant to be emulated?
=====================
1 Rumelhart, David E. & James L. McLelland and the PDP Research Group, Parallel Distributed Processing, 2 vol., Cambridge (Mass.) : MIT Press 1986
2 “Godfather of AI” Geoffrey Hinton Warns of the “Existential Threat, May 9th 2023
Portrait of Geoffrey Hinton by Stable Diffusion