3.3 How Would You Teach A Computer To Play Chess?

How Would You Teach A Computer To Play Chess?

Ask yourself the question: how would you teach a computing machine to play chess?

There are two possibilities off the bat. The first one is that because a computer has the ability to store information, why not store data about every game of chess ever played? You could give the computer all of the chess moves that were possible, stored in tables. But, in the game of chess, after both players make their first move, 400 possible board setups exist. After the second pair of turns, there are 197,742 possible games, and after three moves, 121 million. To store all of those possibilities written down on a piece of paper would be impossible – to store them in a database would require a massive amount of storage. Eventually you’re going to be limited by storage capacity.

If this is the first possibility, than the second one is to teach the computer the rules of chess.Think about it. We humans don’t play Chess based on how many different configurations of Chess we’ve memorized. We play chess based on an understanding of the rules. When I look at the chess pieces, I think “A pawn can only do this… a knight can only do this…” when you program an A.I., you teach it the rules, and then you let it go. It’s first day, the first steps, is that it knows what each piece does, but it doesn’t have memories of playing. It has to create those memories of playing and store them. When it puts that piece down and the piece gets taken, its pieces keep getting taken, it’s stored an entire game of “OK, we’re not going to do that!”

What is remarkable about Turing and Shannon, and why their contributions to computation theory laid the groundwork for modern computing, is precisely that it takes a person with an understanding of how thinking works in order to write a program that knows how to think.

Essentially, if you try to train an artificial intelligence by giving it everything you know, you will have a very inefficient and storage intensive computer. If you can figure out a way to teach the computer the rules and have it learn from its own mistakes, you have a much more efficient method of achieving your desired goal. Now we have projects like OpenAI, where computers are playing multiple games of multiplayer battle games like Defense of the Ancients (DoTA) 2 at the same time.

If you want to teach a computer how to play chess, you have to teach it how to think.

So, did Turing ever create a computer that could play chess? Unfortunately, the computers of the time were too slow to implement the program he wrote. Instead, Turing wrote an algorithm that could theoretically be implemented on a computer that could play two moves in advance of the player. Check it out here.