Deep Learning Machine Teaches Itself Chess

A new break through in the way artificial intelligence learns has occurred. While super computers that can plan and defeat human opponents in chess have been around for a long time, the engine that runs these programs has changed little. These super computers rely on a method of brute force, considering all strategies and possible outcomes and playing the statistical best move over and over again. Famously, IBM's Deep Blue super computer beat our Gary Kasparov nearly 20 years ago using this method. But Kasparov shouldn't feel too bad, I once played chess against a taxiermied cat and lost.

So what's so nifty about this new intelligence engine? It's the way that it learns and evaluates the game of chess. It has not been given the positional data to sort through and make selections on the best of all the available moves, but rather, has developed it's own set of moves based on actual play. The system uses a layered system of nodes in much the same way the human brain does, and these nodes change as required to adapt to changes in the game.

This is the same way our brains are fine tuned over time to recognize the things we are taught, like a person's face or a brand's logo. What happens is over time a very complex neural network is refined and given certain input, the computer can quickly narrow down the best output, thus eliminating the need to search through hundreds of thousands of possibilities and just sticking the ones that it remembers as being the most profitable.

This is basically rapid evolution. We may not have to create an artificial brain, we may just have to create a brain that can learn. This leaves enormous potential for unimaginable intelligence advancements. But I don't think it's time to start worrying about Skynet just yet. We should be worried once a computer knows it's playing chess.