Google DeepMind CEO Demis Hassabis might have won a Nobel Prize for his work on AlphaFold 2, an AI model that can predict protein structures, but the solution to the problem he really wants to solve still evades him.
The problem in question is infinitely easier to grasp and more relatable than Hassabis' work in the field of chemistry. "The thing I really want that we're working on is next-generation email," he said, speaking at SXSW London on Monday. "I would love to get rid of my email."
Based on the crowd reaction, it was a popular sentiment in the room, where earlier that day, former British Prime Minister Tony Blair admitted to sending only one email during the entire 10-year period he was in office.
There is some irony to Hassabis' quest. The prize-winning scientist is responsible for developing some of the most complex and sophisticated AI models the world has ever seen, all in aid of working toward cures for diseases that are beyond anything we have access to today. His mission to render email (presumably Gmail?) -- an annoyance of our own human invention -- obsolete feels like small fry in comparison.
But it also exposes the duality of Hassabis' responsibilities at Google. He is, and always has been, deeply committed to pursuing AI for the benefit of humankind. "My personal passion is applying [AI] to the frontiers of science and medicine," he said. At the same time he is beholden to the corporate interests of Google, which acquired DeepMind in 2014.
Hassabis always imagined the development of AI to be more of a "scientific-led endeavor," spearheaded by a computer science equivalent to CERN, the famed particle physics lab in Switzerland. But the technology went a different way, becoming commercially viable much quicker than he anticipated. From there, he said, "the capitalist engine has done what it does best."
Hassabis almost speaks as though he is separate from the "capitalist engine," but he is deeply embedded within it. DeepMind being owned by Google means that as well as pursuing his passion project of curing disease with AI -- arguably the most noble use of AI -- he must split his attention to ensure Google's AI products, from Gemini to Veo and everything else the company announced at I/O last month are up to scratch in a competitive market.
In pursuit of AGI
The competition is "ferocious," and it's a hefty work schedule for one man, who says he sleeps very little and doesn't expect to until "we get to AGI," or artificial general intelligence. Along with developing DeepMind's core AI models and translating them into science, he continues to pursue the development of AGI, or AI that fully matches (or exceeds) human intellectual capabilities. "My feeling is that we're about five to 10 years away," he said.
His vision for AGI is that it will unlock a world in which "we can cure many, many diseases -- or maybe even all diseases," and "unlimited renewable energy." In some ways, the Google products are stopping-off points on the way.
One reason DeepMind has built Veo 3, its latest video generation software, said Hassabis, is that AGI needs to have a physical understanding of the world around it. The world models built for Veo 3 are key to this understanding. In turn, these world models will be essential for a breakthrough in robotics, which Hassabis believes is due in the "next few years."
While it's sometimes unclear where DeepMind's worthy mission ends and Google's commercial priorities kick in, Hassabis is clearly finding ways to make it work for him and his long-term pursuit of an AGI breakthrough.
In spite of the seismic shift he predicts this will cause, even he is skeptical of the hype around AI in the short term. "I mean, it couldn't be any more hyped," he said. "Therefore, it is a little bit overhyped."