AlphaGo would have won all 5 games…if it were part human
AlphaGo’s 4–1 defeat of the World Go champion Lee Sedol is no small feat. With more possible boards than atoms in the universe, brute computing power is far away from being able to play a perfect game. Its the advanced move analysis and board analysis neural networks that have enabled this AI’s evolution to move faster than the predicted “10 years” people thought it would take to best a human grandmaster.
By now you have probably read that AlphaGo is weaker playing as first player then playing second. A long time ago, there was a distinct advantage for the first player, but this has since been leveled by giving the 2nd player a substantial headstart in points. Lee Sedol’s victory came when he played first. After the victory he asked if he specifically could play second — he wanted to see if he could beat the computer when it wasn’t weak.
This was a great example of a true master trying to overcome an obstacle and Lee Sedol’s play is unquestionably the greatest in the world at the moment
But i think AlphaGo could have won the 4th match.
Despite being behind after the critical 78th move Lee Sedol made, AlphaGo still had a substantial lead in time left to play. Each player has a set amount of time to play the game during the match and AlphaGo has a dedicated neural network to handle time-management. Lee Sedol had used up all his time and had to make moves in 60 SECONDS!! Can you imagine how hard it would be to keep up that level of play with that kind of time constraint!?! Im sure you could…. but i don’t think AlphaGo did.
AlphaGo was programmed to resign once it saw a 20% chance of winning. But what would have happened if it understood that its opponent would make increasingly bad decisions as the game went on? Wouldn’t it have made sense to play the game out as long as possible, to force the time constrained human opponent to make several bad moves?
What do you think?