The Man Heading Toward Infinity

Demis Hassabis founded DeepMind and won the 2024 Nobel Prize in Chemistry for AlphaFold, but the book portrays him not as an ostentatious genius but as a quiet, composed scientist. The author sees this image as contrasting with figures like Elon Musk or Sam Altman, making him appear a more trustworthy leader on the path to AGI.

But there's a separate core question: should AI research itself keep being pushed forward? The book's starting point is a confession Geoffrey Hinton made to Nick Bostrom—that the "temptation of discovery" is too strong to make stopping research easy—and this confession drives the whole discussion.

Hassabis's life story makes that question even more complicated. He was born in London in 1976, was a chess prodigy and game developer, and went on to a Cambridge computer science double-first, a UCL PhD, postdoctoral research at Harvard/UCL, and then co-founded DeepMind with Shane Legg and Mustafa Suleyman in 2010. After that, he had to run research and business simultaneously—raising investment to keep DeepMind afloat, the acquisition by Google, and the strategy of staying in London while dealing with the California headquarters.

The author places importance on the passage describing Hassabis as, above all, a scientist. He views AI as an object of almost religious curiosity, saying things like "I want to understand nature" and "computers are just sand and copper—how can they beat a Go champion?" He's also indifferent to accumulating personal wealth: he has only one old house, no luxuries like yachts or vacation homes, and his collection is described as amounting to little more than a first edition of Shannon's paper.

At the same time, the book carefully traces how Google/DeepMind, despite being pioneers of the transformer, lost the initiative after ChatGPT. Even Hassabis underestimated ChatGPT's impact, but once he accepted that the competition had begun, he felt there was no turning back—and the result was an even faster sprint under technologically deterministic pressure.

However, because the book focuses so heavily on the person, it leaves the machines and infrastructure that made the models possible largely blank. Moore's Law is mentioned only early on, Nvidia appears only once in the main text, and GPU doesn't even appear in the index. The book places far more weight on the drama of the people operating atop the AI boom than on its economic and industrial foundations.

In conclusion, the book makes Hassabis look like the most trustworthy face of the AGI era, but leaves open the point that this trust doesn't necessarily mean legitimacy. AGI may be "bigger than any company or individual, belonging to all of humanity," but who will control it and under what institutions remains an open question—and Hassabis himself says, "I'm just doing the best I can."