AI Machine Teaches Itself to Become a Top Chess Player in 72 Hours

istock
istock / istock
facebooktwitterreddit

These days, a computer that has mastered chess is about as surprising as a toaster that can heat up bread. But most chess programs, since the introduction of IBM’s Deep Blue in 1996, have relied on brute force to work out moves. When Deep Blue beat chess master Garry Kasparov in the ‘90s, the computer searched through approximately 200 million positions per second, whereas Kasparov could consider around five per second. Deep Blue was able to beat Kasparov not because it was a great strategic chess player, but because it had the processing power to consider and eliminate options with incredible speed.

But now a new artificial intelligence machine is revolutionizing computer chess by actually learning. According to MIT Technology Review, the AI machine, known as “Giraffe,” taught itself chess in just 72 hours. Giraffe uses a neural network—inspired by the human brain—which consists of several layers of nodes whose connections change as the system learns. 

This means scientists can “teach” Giraffe chess by inputting data derived from real chess games. Giraffe observes the data and learns to recognize which moves are strong and which are weak. Instead of considering millions of positions for each move, the machine employs strategy and is able consider fewer positions, just like a human chess player, because it can rule out moves that don’t make sense from the start.

Giraffe’s creator, Matthew Lai, tested the machine’s progress as it learned chess over the course of 72 hours. He used a database called the Strategic Test Suite, which grades the machine’s understanding of different strategies, like “the understanding of how bishop and knight’s values change relative to each other in different situations.” According to Lai, Giraffe's chess abilities peaked after 72 hours of "training," by which time the machine placed within the top 2.2 percent of tournament chess players.

“Unlike most chess engines in existence today," Lai explained, "Giraffe derives its playing strength not from being able to see very far ahead, but from being able to evaluate tricky positions accurately, and understanding complicated positional concepts that are intuitive to humans, but have been elusive to chess engines for a long time.”

[h/t: MIT Technology Review]