Google Thinks Its AI Software Can Beat The Best

Computer’s have come a long way since Pong, and an even longer way since the early chess programs with their lo-res graphics and monochromatic color schemes that made game play difficult. But the term “playing against the computer” hasn’t had this much importance since chessmaster Kasperov lost to IBM’s version of outsmarting the smartest.

Google’s DeepMind Lab has been hard at work on all kinds of AI innovation, and the goal is far more involved than just beating a game. But one of the truest tests of how capable the mechanical intelligence of a computer is depends on its ability not just to retrieve information–as in the case of the game software Watson that readily beat multiple Jeopardy! champions with its ability to seek and recall factual nuggets of information–but in its ability to reason based on countless parameters in front of it.

That’s why the ancient and well-loved Asian game Go is the perfect test of reasoning ability. Like chess on steroids, Go requires its players to manipulate their pieces on a 19-grid board and make constant determinations about what moves are significant and which moves are just “throwaways.” And thanks to Google’s AI program AlphaGo, the real test–beating the world champion–is only weeks away.

In a live-streamed event broadcast from Seoul, AlphaGo will take on Lee Sedol in a five-game match over the course of a week next month. Both sides, Sedol and the head of Google’s AI division Demis Hassabis, are alarmingly confident that their abilities will prove to be superior, but one of them is about to come away disappointed. Rather than feel the sting of loss, though, this match should either show us just how powerful DeepMind’s tech is, or will demonstrate that there are still advancements to be made.