Why Google’s computer winning this board game against a human is a big deal
To beat a human at Go, a computer needs to do much more than brute force calculations.
It’s the artificial intelligence breakthrough many were waiting for. A computer just beat a human expert at the board game Go, a development that many in the field of AI say will brings artificial intelligence closer to human-like intuitive decision making.
Computers beat humans at checkers in 1994. Since then computers have defeated humans at games like scrabble, poker and even the American television quiz game Jeopardy. In 1997, IBM’s Deep Blue supercomputer defeated Grand Master Gary Kasparov in the second of their 6-game clashes. Go has been the last and most important bastion to fall to the machines.
In a game of chess, an artificial intelligence system can compute all the possibilities at each stage of the game and pick the best move. A powerful computer like Deep Blue can compute these permutations so quickly that it can get its results within the classical time controls that a normal game of chess is played under. This brute force computing is not possible in Go, simple because the number of possibilities is too big.
The game, invented some 2,500 years ago in China, is played by two players placing black and white stones and a gridded board with the objective of covering as much of the board as possible while avoiding capture by the opponent. It is a great computing puzzle because it is a deterministic perfect information game where no information is hidden and there is no element of chance.
At the same time there are, as DeepMind CEO Demis Hassabis says, “more configurations of the board than there are atoms in the universe.” Even a powerful computer cannot predict the results of the match based on a move made in the initial stages.
DeepMind scientists programmed their computer AlphaGo to predict the game to about only 20 moves ahead and see where that takes it. AlphaGo operates on information from its two neural networks that are modelled after the human brain. One neural network evaluates how the computer is situated on the board and the other decides where it should move. AlphaGo played European Go champion Fang Hui in October and beat him in five straight games.
Not to be left behind, Facebook too announced that its programmers were close to reaching the milestone of defeating a human expert at the game.
Does this mean that computers can take over where only humans have tread? Go programmers at a high profile computer Go tournament said for this Wired article that game-playing computers are solutions to specific problems and defeating a human at a game does not mean a computer can move on to curing cancer. As the report from the tournament says “But in truth, these machines are nowhere close to mimicking the brain, and their creators admit as much.”