Google's AlphaGo Beats 'Go' World Champion In Historical Moment For AI
Google’s DeepMind-based AlphaGo AI managed to beat Lee Sedol, the world’s premiere Go player and 18-time world championship, in what represents a historical moment in the progress of artificial intelligence. Granted, the victory came in just the first match of five; the remainder will be played out over the next few days, through until March 15.
Lee Sedol seemed to have the upper hand for most of the game, until the last 20 minutes when AlphaGo gained a bigger lead on him. This may have happened because it’s harder for the AI to figure out an overwhelmingly winning strategy in such a complex game as Go. It can’t calculate too many movements ahead of time due to the enormous processing power that would require.
However, towards the end of the game, fewer moves are possible, which could mean it became much easier for AlphaGo to determine the best move at any given moment. Therefore, any mistake from Lee Sedol, no matter how small, could then be turned in a bigger and bigger advantage for AlphaGo as the game approached its conclusion.
Still, AlphaGo managed to hold its own with the world’s best Go Player for more than three hours. Perhaps if Sedol had taken a larger lead early on, he may been able to fend off the AI bot.
Even if Sedol wins all of the other four matches, Google still achieved a historical breakthrough in artificial intelligence with last night's win. It would then likely be only a matter of months or a year at most before AlphaGo and its DeepMind core, as well as the hardware processing power behind it, would improve enough to become unbeatable by any human player.
Since Google announced that its DeepMind AI can play and finish games from the 1970’s only about a year ago, the AI has already gone to play games such as Go and the 1990’s Doom video game. The improvement rate has been quite astonishing. The DeepMind AI is likely not too far behind being able to play any modern game in the same way a human would - by learning everything about the game from scratch.
The next match between AlphaGo and Lee Sedol will happen today at the same time, 11pm ET. It will be interesting to see whether Lee Sedol will use a different, perhaps more aggressive, strategy early on against AlphaGo, now that he learned how the AI plays, in order to get a larger advantage early on.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Lucian Armasu is a Contributing Writer for Tom's Hardware. You can follow him at @lucian_armasu.
Lucian Armasu is a Contributing Writer for Tom's Hardware US. He covers software news and the issues surrounding privacy and security.
-
thor220 The name DeepMind is ironic in that AI programs like this are the exact opposite. They calculate every possibility and simply select the one with the highest chance of winning.Reply -
heliomphalodon Sorry, @thor220, you're wrong. The possibilities of go are far too large for a brute-force approach. The whole point of DeepMind is that it DOES NOT rely solely on deep-ply search. Educate yourself, please.Reply -
ajpaolello Didn't Deep Blue win the first match against Kasparov and Kasparov ended up winning in the end?Reply -
therealduckofdeath The first match Kasparov did vs Deep Blue turned out to be Kasparov vs Deep Blue and a bunch of chess players, as the engineers updated (helped) the AI during the match.Reply -
esrever Watching this live was a roller coaster of emotions. On the one hand I wanted the human to win but on the other, this is a monumental breakthrough in Computer learning and AI development. It looked really good for Lee Sedol until 2/3 of the way through the game when he made 1 critical missplay and it all went down hill from there. I really hope the AI doesn't 5-0 him now due to emotional stress or something.Reply -
paladinnz Watching the whole game and listening to the commentary one of the things that stood out is that AlphaGO doesn't get tired where as Lee Sedol has to concentrate for >3 hours solid.Reply -
alextheblue ReplyIs their D-Wave part of this?
Not unless it's present as extremely expensive decor, which is about all it's good for currently. -
hoofhearted ReplyIs their D-Wave part of this?
Not unless it's present as extremely expensive decor, which is about all it's good for currently.
Just curious, because this article (Dec 2015):
http://www.popularmechanics.com/technology/gadgets/a18475/google-nasa-d-wave-quantum-computer/
seems to indicate that they made a breakthrough -
hoofhearted I am certainly no epert (not even a novice), but this article seems to describe the problem presented:Reply
http://www.wired.com/2015/09/googles-quantum-computer-just-got-a-big-upgrade-1000-qubits/
(At the bottom after the heading "reversing the Trend")