How Artificial Intelligence has changed over two decades: Taking Chess as an Example

How Artificial Intelligence has changed over two decades: Taking Chess as an Example Image

Artificial Intelligence has changed over the past two decades. What has changed is the fundamental approach to it. We will use Chess as an example.

“Look ahead” is the concept where if I move this piece here then that is the new state of the board. It is thinking one move ahead in human terms. Computers in the 1990s used this “look ahead” but tried to look several moves ahead. You can picture yourself doing this in a game. This is the fundamental approach that brought Deep Blue to beat Gary Kasparov the best chess player in the world in the 1990s. You can “look ahead” to see all the different outcomes a few moves ahead and try to make the best moves that way. Now in the 1990s a computer would look ahead several moves to see what moves might be better. This is similar to how some humans might approach chess, thinking several moves ahead. This is not how Artificial Intelligence approaches the task of playing Chess now.

Artificial Intelligence as it is often applied to Chess looks very different now. Instead of trying to look several moves ahead. That approach isn’t used and something else is done instead. What is done now most often is pattern recognition. So now you show a computer millions and millions of positions of chess boards each time telling the computer the winner in each position. After seeing millions and millions of chess positions the computer sees patterns. So if you show it a random chess board it will know what positions it saw before that looked like that current position and who actually won that game. Meaning the computer recognizes the current chess board the current position as close to a few different patterns it saw before, which is pattern recognition. Then for example in those patterns it saw before most of which White won in the Chess game. So it would recognize that pattern is better for White.

Keep in mind the computer doesn’t have any concept of why the board is better for one player or another it just sees the patterns. Based on the patterns it says: “Well this chess position looks similar to 25 boards I saw before and on those 25 boards white won 70% of the time. Thus the position is better for white 70% of the time”. So beyond the pattern recognition there is no deeper understanding of chess but this pattern matching approach can inform the chess playing computer which of its possible next moves look the best. This approach has born more fruit then the look ahead approach from the 1990s. Computers are very good at finding patterns and this approach is the a more modern approach when solving problems like Chess. The only real restriction to this pattern matching technique is that the input size needs to be very large, as in millions and millions of chess positions mentioned above.

Chad Jones Photo
Chad Jones Photo
About the Author

Chad Jones

Chad is the Founder and CEO at Push and was a former Apple Engineer before returning to Saskatchewan to revolutionize the mobile development world. Chad is passionate about creating efficient, well-designed software.