Deep Blue and Chess: A Landmark in Artificial Intelligence?

By

Home / Deep Blue and Chess: A Landmark in Artificial Intelligence?
Statue of Alan Turing. One of the computer pioneers, who was one of the first to propose that computers could play Chess. Photo: Stephen Richards / CC 2.0

Statue of Alan Turing. One of the computer pioneers, who was one of the first to propose that computers could play Chess. Photo: Stephen Richards / CC 2.0

May 11th 1997 stands as a landmark in Artificial Intelligence (AI). This was the day that Deep Blue® defeated the reigning World Chess champion Garry Kasparov.

Whether Deep Blue was intelligent is a matter of debate, and the actual defeat of Kasparov is not without its controversy as well.

Deep Blue: History

We can trace Deep Blue back to 1985, with the work of Feng-hsiung Hsu, Thomas Anantharaman and Murray Campbell at Carnegie Mellon University.

Their computer, Chip Test, was a Chess program burnt onto a computer chip. This hardware solution runs much faster than if the program had run from the computer’s RAM, or Random Access Memory.

Being implemented with hardware, Chip Test was able to search through 500,000 positions a second.

Deep Thought was developed from Chip Test and was able to search 700,000 positions per second.

In 1989, Deep Thought moved to IBM®, with Deep Thought 2 becoming the first product of this new enterprise. It had four major improvements over the original Deep Thought: multi-processing, enhanced hardware, improved search algorithms, and an extended book, or set of opening moves that follow a predetermined path, until one of the players deviates from it and takes the game into unknown territory. The extended book allowed the software to make good moves at the start of a game, even when no opening book was available.

The first version of Deep Blue (Deep Blue I) saw the light of day in 1996. It ran on a 36-node IBM computer and had 216 Chess chips. Each chip was able to search between 1.6 and 2 million Chess positions a second, enabling the whole system to consider 50-100 million positions a second.

The programmers of Deep Blue II (now just referred to as Deep Blue) recognized some of the shortcomings in Deep Blue I and attempted to correct them. This included a new Chess chip which had a better evaluation function that was now able to take into account around 8000 features, rather than the 6400 of Deep Blue I and a number of changes after observing games against Kasparov and other Grandmasters. Deep Blue II was able to search between 2 and 2.5 million positions per second. The number of positions Deep Blue II actually searched depended on the specific game state.

Leave a Comment