Sebastian Malaby

And just to set the context here, in 2010, when he created DeepMind, AI could not recognize the photograph of a cat.

Like AI could do nothing.

It was deep, deep AI winter.

And it was only in 2012 that anything began to work.

And so part of that making it work was Demis and his team at DeepMind

building an agent that could play all kinds of different simple video games.

There's this suite called the Atari suite, you know, the games like Pong, Breakout, Sequest and so forth.

And they built an agent that without being told the rules of these games could figure out how to play it by itself and then figure out strategies for winning and get really strong at them.

And so this was an early proof that AI was going to work.

So his gaming background really ended up being key in the development of AI.

Absolutely.

So gaming was important partly because the games developers back in the day realized that if they had AI, they could make the games more compelling and interesting and fun.

Because you could have characters in the game that learnt new behaviours based on the feedback they got from the player.

So that kind of set the seed of the aspiration to build AI.

But then also, games turned out to be a fantastic environment in which to train an AI agent because you could...

play the game over and over and over, and the machine could learn by playing many times.

And so after that Atari system I told you about, the next thing was to build an AI system that could win the game of Go, this ancient Chinese game.

And in 2016, they went to Korea, played this big show match against a Korean champion, and the AI system defeated them.

Absolutely.

You know, ever since 1997, when the deep blue system that IBM had defeated Garry Kasparov in chess, it's been game over for humans, I'm afraid.