A little while ago, we told you about a computer named AlphaGo beating a three-time European champion at the game Go. Now, that same computer has faced Lee Se-dol, a man some regard as the best Go player in history, and trounced him. In a five-game tournament, AlphaGo beat Lee 4-1.
For people used to home computers routinely beating them at Chess, Tic-Tac-Toe, Othello and other board games, this might not seem like a big deal. However, it's actually a huge step forward for computers. In the future it might be regarded as the birth of computers that think.
One of the reasons a computer playing Go is such a big deal is because Go is the possibly the most complex board game ever invented. It features a huge 19 by 19 grid and the ability to place pieces anywhere.
Unlike chess, where a computer can calculate moves and counter-moves dozens or hundreds of steps into the future, Go has too far too many possible moves for a typical computer to process. Human players rely a lot on experience-honed intuition to make the best moves, which is something computers can't do. Until now.
After the match, Lee said regarding AlphaGo, "It made me question human creativity. When I saw AlphaGo's moves, I wondered whether the Go moves I have known were the right ones." He went on to say that AlphaGo has made him realize there's still a lot to learn.
The way AlphaGo got its skill is fascinating because it wasn't traditional programming. First, DeepMind, the team behind AlphaGo, exposed a neural network, which is a computer that mimics a human brain, to a vast library of Go matches so it could get an idea of which moves worked and which didn't, along with the various board arrangements preceding those moves.
Then two slightly different versions of the neural network played millions of games against each other to refine the information and come up with new strategies. Finally, the results from those millions of games were fed into another neural network, which went on to beat some of the best human players around.
We should point out that AlphaGo's ability to play millions of games in a very short amount of time definitely gives it an edge in experience over humans. In an entire lifetime, a human can only play a few hundred thousand games, and that's if they do nothing else at all with their time.
That experience is one reason that at several points during the tournament AlphaGo displayed human levels of intuition and trickery. In fact, one moment that has everyone talking is Move 37 in game 3.
For Move 37, AlphaGo played a move that no professional human player would have tried, and no human would have programmed in. However, AlphaGo calculated it had a good chance of succeeding, and, because it was so unorthodox, it threw Lee off for the rest of the match.
This ability to take such a massive amount of data and distill it down like this is a huge step forward for artificial intelligence. Computer scientists weren't expecting to be at this point for at least 10 more years. And, with DeepMind being a subsidiary of Google, you can naturally expect this technology to show up in Google's services sooner rather than later.
For those lamenting the demise of the human race, however, there is some good news. AlphaGo isn't infallible. In game 4, Lee managed to pull off a brilliant move that AlphaGo didn't even notice for another 8 turns. It was downhill for the computer after that.
Lee says he isn't quite ready to admit AlphaGo is a better player than humans. He feels that AlphaGo has an advantage in that it doesn't feel pressure or get distracted, which means it can focus purely on the game.
Do you think this is the beginning of the end for the human race, or just an interesting computer achievement? Let us know your thoughts in the comments.