Demis Hassabis

Hassabis and John Jumper co-led DeepMind's AlphaFold. It won the 2018 CASP13 protein-structure competition and, as AlphaFold2 in 2020, determined about two thirds of the CASP14 targets to laboratory-comparable accuracy, according to the independent organisers. DeepMind then released the code and predicted structures for nearly all of the roughly 200 million known proteins, and in 2024 the two shared half of the Nobel Prize in Chemistry 'for protein structure prediction'.

He co-founded DeepMind in 2010 with Shane Legg and Mustafa Suleyman, and Google bought it in 2014. Its AlphaGo beat Lee Sedol four games to one in March 2016, a milestone forecasters had not expected for about a decade. DeepMind and Google Brain were merged into Google DeepMind in 2023 under his leadership, and the lab released the Gemini models; he ran it as chief executive until August 2026, when he became its chair and Alphabet's chief scientist.

Before DeepMind he was a chess master at 13, the lead programmer of the game Theme Park at 17 and the founder of the games studio Elixir (1998 to 2005), then took a PhD in cognitive neuroscience at UCL (2009) and used what he learned about the brain to design better neural networks. A 2007 UCL study of which he was first author, showing that patients with hippocampal amnesia cannot imagine new experiences, was reported by Nature News and, the Academy of Achievement says, listed among Science's top ten breakthroughs of the year.

In 2013 a team of Google executives flew to London, and Hassabis impressed them with a prototype that had learned to play Breakout by reinforcement learning, a technique Google was not then using. Google bought DeepMind the following year for about $500 million, and, Time reports, he turned down a larger offer from Facebook.

This is consistent with an ability to win over a buyer by showing a working demonstration of his lab's approach, though the account rests on his own recollections. Persuasiveness

In April 2005 Elixir Studios, the games company he had founded in 1998, ceased operations and sold its assets after its main project was cancelled as too risky; Game Developer reported that neither of its two released games had been a major financial success. Hassabis said that going down the licensing route was something the studio could never compromise on.

This is consistent with a preference for original, high-risk work over safer licensed projects, at the cost of the company. Risk Tolerance

After AlphaGo's 2016 win, which the Nobel Academy says Hassabis saw as a means and not the goal, DeepMind put a small team on protein structure prediction and entered the 2018 CASP13 competition. AlphaFold won with about 60 percent accuracy, short of the 90 percent needed, and the team stalled; John Jumper's ideas and a transformer-based design led to AlphaFold2, which in 2020 CASP's organisers said had largely solved the problem for single proteins.

This is consistent with a willingness to turn a lab built around games toward a harder scientific problem and to share leadership of the rebuilt system with a newer colleague. Opportunity Sensing