Demis Hassabis

When Demis Hassabis co-founded DeepMind in 2010, he had a 20-year plan culminating in the creation of artificial general intelligence (AGI), an AI system that can do practically any cognitive task a human can. 14 years on, Hassabis thinks things are on track.

 “I believe [AI] is going to be the most beneficial technology ever created, but only if we apply it in the right way and build it in the right way,” he said at a talk hosted by the payments platform Stripe earlier this year.

Known as Google DeepMind since its 2014 acquisition, Hassabis’ lab has had many successes, including pioneering “deep reinforcement learning,” designing systems that beat the best human players at the board game “Go,” and solving the protein folding problem to help scientists design new drugs and understand diseases. 

In the last year, the lab has released their own LLM, Gemini; updated their protein-folding model to predict the structures and interactions of biomolecules like DNA and RNA; and developed models that achieve the same level as a silver medalist in the International Mathematical Olympiad; among other accomplishments. 

Their next big release, “Project Astra”, is billed as a “universal AI agent” that processes text, audio, and video in real time, naturally responding to almost any query. Its demo video shows a woman speaking to it about a problem on a whiteboard in front of her, before asking the system if it knew where she left her glasses—it did.

Hassabis explains that as DeepMind’s research becomes “more mature,” it will be able to develop “AI-first products” that could be integrated across Google's offerings such as YouTube, Chrome, and Android. Advanced systems could quickly be brought to everyday users. 

In a conversation with TIME, Hassabis explained that while Astra is an exciting step along the way to AGI, further breakthroughs will likely be necessary to create a system capable of, for example, “coming up with a new theory of physics.” 

Reflecting on how future systems might change human relationships, Hassabis says, “I think there'll be things happening that we can probably barely imagine today. And hopefully mostly that will be good—but we need to keep an eye on that and discuss, as a society, how we want that to go. It really matters how we deploy these systems and what we use them for. So we're just going to try and be as good a role model as we can be for that.”

Buy your copy of the TIME100 AI issue here