At the beginning of the month, the Swedish Academy of Sciences awarded the Nobel Prize in Physics to John Hopfield and Geoffrey Hinton ‘for fundamental discoveries and inventions enabling machine learning with artificial neural networks.’ The second of them, the British Hinton, has been considered for years as ‘the godfather of AI’ for his role in laying the foundations for its development and for having had a more visible career than that of his Nobel colleague.
This 76-year-old researcher developed in 2012, together with Ilya Sutskever and Alex Krishevskya neural network capable of analyzing thousands of photographs and learning by itself to identify common elements in them: animals, people, objects, etc. This technology is at the base of the operation of chatbots such as those later created OpenAI and Googleand for her he received 2018along with his colleagues Sutskever, co-founder of OpenAI who recently left, and Krishevsky, the Turing award, considered the Nobel Prize in computing.
Hinton began working on AI for Google in 2013, company that he left a decade later to have the necessary freedom to warn of the dangers of artificial intelligence without contradicting his work in the company. He has done it on several occasions since then, predicting a bleak future and he has done it again in a new interview, already as a Nobel Prize winner, granted to Bloomberg, in which he talks about the ‘existential threat’ posed by the uncontrolled development of AI.
‘Yes, it really is an existential threat. Some people say this is just science fiction and until recently I thought it was something very far away. I always thought it would be a very long-term threat, but I thought it would be a hundred years before we had really smart things, maybe 50 years. We had a lot of time to think about it. Now I think it’s quite likely that At some point in the next 20 years, these things will become smarter than us and we really have to worry about what will happen then,’ Hinton tells the outlet.
It was this insight that led him to leave his job with AI at Google. Until then, he had been comfortable with the company’s approach of not releasing anything until it was secure enough, but OpenAI’s success with ChatGPT changed that. Like Microsoft, like Meta, like Open AI, Google began to rush the development of new artificial intelligence products.
The new Nobel Prize in Physics is not only doomsaying, but proposes alternatives. He has long advocated for global regulation of artificial intelligence and in Bloomberg he has pointed out that the United States Government should put ‘the best young researchers to work on this problem with the necessary resources.’ Governments do not have these, large companies have them. The Government should insist that large companies spend much more on security and security research. Right now they employ a low percentage and almost all of their resources go into creating bigger and better models‘.
When asked what amount they should dedicate, Hinton explains that it is not a number but a percentage of your computing resources. ‘I think it would be perfectly reasonable to say a third. That’s my starting point, but I could settle for a fourth,’ says Hinton. That is to say, that one in every three GPUs that companies like Google, Microsoft or Meta use to compute their language models is used to investigate their security.