W.Back in 2011, Marc Andreessen, a venture capitalist with aspirations of being a public intellectual, published an essay titled “Why Software is Eating the World,” predicting that computer code would take over large swaths of the economy. Thirteen years later, the software seems to be making its way into academia as well. This is, in any case, a possible conclusion that can be drawn from the fact that the computer scientist Geoffrey Hinton shares the year 2024. Nobel Prize in Physics with John Hopfield, and that the computer scientist Demis Hassabis shares half of the Nobel Prize in Chemistry with one of his DeepMind colleagues, John Jumper.
The prize for Hassabis and Jumper was, in some ways, predictable, because they built a machine: AlfaFold2 – allowing researchers to solve one of the most difficult problems in biochemistry: predicting the structure of proteins, the building blocks of biological life. Their machine has been able to predict the structure of virtually all of the 200m proteins that researchers have identified. So it’s a big problem for chemistry.
But Hinton is not a physicist. In fact, he was once presented at an academic conference as someone who had “failed in physics, abandoned psychology, and then joined a field without any standards: artificial intelligence.” And he spent a year after graduating working as a carpenter. However, he is the guy who found the method (“back propagation”) that allows neural networks to learn, which was one of the two keys that unlocked machine learning and triggered the current manic rise of AI. (The other was the invention of transformer model by Google researchers in 2017).
Where is the physics in all this? That comes from Hopfield, with whom Hinton shares the award. “Hopfield networks and their later development, Boltzmann machines, were based on physics,” Hinton explained to the man from the New York Times. “Hopfield networks used an energy function and the Boltzmann machine used ideas from statistical physics. So that stage in the development of neural networks depended – to a large extent – on ideas from physics.”
Then it’s okay. But the media often describes Hinton as “the godfather of AI,” which has a vaguely sinister tone. In person he is the complete opposite: tall, kind, courteous, cerebral and gifted with an ironic and sometimes caustic wit. When Cade Metz asked him his reaction when he received the news of the award, he responded that he was “shocked, amazed, and flabbergasted,” which I guess is what most people say. But in 2018 he shared the Turing Prize, the equivalent of the Nobel Prize in computer science, with Yoshua Bengio and Yann LeCun, for their work on deep learning. That’s why he was always in the top league. The thing is that there is no Nobel Prize in computer science. Given the way software is eating the world, maybe that should change.
There is an old joke that the key to becoming a Nobel Prize winner is to “live longer” than your rivals. Hinton, now 77, clearly took note. But actually the most admirable thing about him is the tenacious persistence with which he continued to believe in the potential of neural networks as the key to artificial intelligence long after the discipline had discredited the idea. Given the way academia works, especially in a rapidly developing discipline like computer science, that required exceptional determination and self-confidence. Perhaps what sustained him in his darkest moments was the thought that his great-great-grandfather was George Boole, the 19th-century mathematician who invented the logic behind science. all of these digital things.
We also think about the impact that the award has on people. When news of Hinton’s prize came I thought of Seamus Heaney, who received the literature prize in 1995. He described the experience as “like being hit by a mostly “benign avalanche.” Note that “mostly”: one of the consequences of a Nobel is that the winners immediately become public property, which everyone and their dog wants a piece of. “All I do these days is ‘show up’,” Heaney wrote resignedly to a friend in June 1996. “I am a function of schedules, not an agent of my own being. And it will continue like this for weeks and months… Whatever the Stockholm effect ultimately is, its immediate result is the desire to quit work and start again. in person (in my own person).”
So… Memo to Geoff: congratulations. And keep control of your calendar.
what i have been reading
talk this way
Is a chat with a bot a conversation? it’s a fabulous New Yorker rehearsal by historian Jill Lepore about her interactions with GPT-4o, Advanced Voice mode.
Interesting times…
October 2, 2024. This particular problem from Heather Cox Richardson’s indispensable Substack blog is a gem.
A real page turner
Elite college students who cannot read books are a intriguing report in it Atlantic by Rose Horowitch.
