This is an amazing breakthrough in artificial intelligence. Mathew Lai at Imperial College London has created an artificial intelligence machine that taught itself to play chess at grand master level in only 72 hours. The grand masters of chess are seen as the standard for the best human players for any game, in this case it is chess. Chess is a great game to test a deep learning machine against because it has all of the information (the pieces) visible to both players, and none of the game is up to chance. A good example of the opposite is poker because it has both hidden information and elements of chance. This is why humans are still, on average, better than computers at games similar to poker, whereas computers have been much better than the majority of humans for a long time.
Why is this Deep Learning Machine Such an Achievement?
Unlike its predecessors, Giraffe does not use a blind search to find the best result. All of its predecessors were designed to iterate through every possible move, based on its starting positions, that it could possibly make and then decide what the best outcome would be. It does this by looking at the state, then creating a new state of what would be if they moved a piece and then thinking of another state and so on. It essentially builds a tree, creating branches to possible states based on each possible move. It does this for every piece and every moveable position. The majority of them also compare the possible moves against a database of grand master past games. This sometimes takes years of training the machine for it to know what is the best move based on a state. Needless to say, this is not how a human plays nor would we be able to do it as quickly. So how are some people still able to beat these computers? They study the board, not the pieces. This was the concept behind Lai’s deep learning machine. He did this by using a a neural network. Neural networks are designed to work similar to the human mind. These networks are trained using test cases to improve their ability to make decisions. This is an amazing advancement in artificial intelligence and neural networks. This will also set a new precedent for AI in gaming. This type of innovative thinking is what will push us into the future.
