Artificial intelligence: A rival to man?

What if the universe is just a giant game of Go?

When the Deep Blue computer beat chess master Gary Kasparov in 1997 in a giant advance for artificial intelligence (AI) , it did so with what computer geeks call "brute force". Using huge computer power it crunched through every possible move, looking further ahead than could any human, let alone Kasparov.

Now, Nature magazine reports, DeepMind, a Google subsidiary, has taken another mind-boggling leap that testifies to the almost infinite potential of new "self-teaching" techniques in AI, with a system called AlphaGo. The latter has beaten a master of the oriental board game Go that, like chess, is a test of pure skill, but with a degree of complexity that far, far outstrips it. There are more possible positions on a Go board than atoms in the universe, or 10170 permutations.

A pastime beloved by emperors and generals, intellectuals and child prodigies, Go is played on board with a grid of 19 lines by 19 lines that is gradually filled up with black and white buttons, each placed at the intersection of two lines. The aim is to surround one’s opponent’s pieces and to capture them, while avoiding a similar fate. Games run on average to 200 individual moves, four times as many as chess. While the first moves by black and white in chess may produce up to 400 possible board positions, the same in Go, is an eye-watering 129,960 positions.

“Deep learning” techniques, instead of exploring infinite future possible moves, are based on artificial neural networks that mimic the human brain. AlphaGo taught itself and developed its ability to deal with complexity by repeatedly playing other, lesser computers and then a version of itself millions of times over.

READ MORE

Playing Go is not an end itself. Such techniques can be applied to the development of robotics and research on complex patterns like weather. And it’s still some way from human intelligence. “It’s not really human-level understanding,” Ryan Calo of the University of Washington argues. But if AlphaGo can understand Go, then maybe it can understand a whole lot more, he wonders. “What if the universe is just a giant game of Go?”