Demis Hassabis on AGI, the Turing Test, and Games as the Crucible of Intelligence

Demis Hassabis, CEO and co-founder of DeepMind, discusses his lifelong journey into artificial intelligence, beginning with childhood experiences in chess and programming. He highlights DeepMind's groundbreaking AI systems like AlphaGo Zero and AlphaFold 2, which have surpassed human capabilities in complex tasks. A central theme is the pursuit of Artificial General Intelligence (AGI) and the challenges of defining and measuring it. Hassabis critically examines the Turing Test, viewing it more as a philosophical thought experiment than a rigorous formal benchmark, advocating instead for testing AI across a vast range of cognitive tasks and modalities to assess true generalizability. A key distinction is drawn between specialized AI, exemplified by Deep Blue's chess victory, and general intelligence, which DeepMind aims to achieve through systems like AlphaGo and Gato. Deep Blue, while powerful, was hand-coded and lacked generality, unable to perform even simpler tasks. In contrast, AlphaGo learned from scratch through self-play, demonstrating true learning and extrapolation. Hassabis also categorizes AI creativity into interpolation (averaging existing data), extrapolation (generating novel solutions within a domain, like AlphaGo's Go moves), and true invention (creating entirely new domains or rule sets, like inventing chess itself), noting that current AI excels at the first two but struggles with the latter due to the difficulty of specifying high-level abstract concepts. From a practical standpoint, Hassabis emphasizes the strategic importance of games as an efficient and measurable testing ground for AI algorithms. Games provide clear rules, quantifiable rewards, and established human benchmarks, allowing for rapid iteration and progress, as seen in DeepMind's early successes. He suggests that future AI could serve as powerful tools for game designers, capable of auto-balancing complex game rules overnight, a task that currently requires thousands of human hours. This approach leverages AI's ability to run millions of simulations in parallel, optimizing game mechanics for compelling human experiences. The broader implications of AGI are explored, touching on the philosophical impact of AI surpassing human intellect. Hassabis reflects on the humbling realization that tasks once deemed impossible for machines, like mastering Go, are now achievable, challenging human perceptions of intelligence and capability. He envisions a future where AI could design entire worlds or games that are optimally compelling, akin to "planet earth," raising questions about the nature of reality and simulation. The discussion underscores the profound societal and philosophical shifts that AGI could bring, from redefining creativity to altering our understanding of what it means to be intelligent.

I don't think he meant it to be a rigorous formal test I think it was more like a thought experiment almost a bit of philosophy he was writing

I think certainly we as humans use language as our main generalization communication tool so I think we end up thinking in language and expressing our solutions in language

prediction seems to be fundamental to intelligence and what you're predicting doesn't so much matter

I always had this feeling and I've always loved this about computers that you can set them off doing something some task for you you can go to sleep come back the next day and it's solved

games have been significant in my life for three reasons so first of all to to I was playing them and training myself on games when I was a kid then I went through a phase of designing games and writing AI4 games... and then the third use... is in deep mind from the beginning which is using games as a testing ground for proving out AI algorithms

it's humbling to realize we're not as smart as we thought it's humbling to realize that the things we think are impossible now perhaps will be done in the future

kasparov I think rightly called it the drosophila of of intelligence right so it's sort of I I love that phrase and and I think he's right because chess has been hand in hand with AI from the beginning of the the whole field

deep blue brilliant as it was at chess it had been hand coded for chess and actually had distilled the knowledge of chess grand masters into into a cool program but it couldn't do anything else like it couldn't even play a strictly simpler game like tic-tac-toe

I think the kind of lowest level of creativity is like an interpolation so an averaging of all the examples you see

alphago played you know millions of games of go against itself and then it came up with brilliant new ideas like move 37 in game two bringing a motif strategies and go that that no humans had ever thought of

could you invent go right could you invent chess and not just come up with a brilliant chess move or brilliant go move but can you can you actually invent chess or something as good as chess or go

come up with a game that only takes five minutes to learn... but many lifetimes to master