In this 16 minutes talk, co-founder and CEO of Deepmind (now Google Deepmind) Demis Hassabis presents his ideas and goals on artificial intelligence and learning. He goes on to say that utimately, we need to solve AI to solve the grand questions of humankind - the grandest one being the theory of everything, a perfect physical model of the universe accounting for litteraly everything. It might indeed take an AI to solve the inconsistencies physicists encounter at sub-nano level at very high energies, since those problems require to both use quantum mechanics and general relativity - which collide in the extreme cases.

As a Go player, I am tempted to make a prediction on the evolution of computer Go. I have been warned from making such risky predictions, but I feel that even if I turn out to be wrong, I’ll learn from them since I’ll have written them down - this Blog will serve as a track-record for my predictions, and I’m looking forward to seeing how badly I’m wrong in a few years!

I predict that computers will be able to play Go at the professional level within the next 5 years; more precisely, I expect a computer to be able to win at least a game out of four against a professional player with a rank of at least 1p on a 19x19 board with the computer taking black with komi $\geq 0.5$. This has to happen by 13/09/2020 – Simon Janin

I believe it’s within reach in 3 years or less, but 5 years seems like a fair bet.