The Rise of Chess AI Part 1: A Short History of Chess and Modern Artificial Intelligence.

In popular culture, the defeat of reigning world chess champion Garry Kasparov at the computer chips of Deep Blue in 1997 marks the beginning of the ‘Computer Age’ in chess, however Deep Blue had primarily used a search algorithm not a machine learning algorithm (Campbell, Hoane and Hsu, 2002). This is an important distinction to make, as it was a crude processes, wherein the AI did not learn how to play chess but rather used complex algorithms and purpose built hardware in order to search for the best outcome – meaning tree trimming wasn’t well developed and inefficient; restricted to roughly 8’000 patterns (Campbell, Hoane and Hsu, 2002). It is due to this and other promotional reasons that lead to IBM at the time being hesitant to call Deep Blue a true instance of ‘Artificial Intelligence’ (Korf, 1997).

Furthermore, the contest between AI and World Chess Champions had remained close for many years following. With multiple events featuring multiple top chess players defeating AI, the last of which occurred in 2005 (Levy, 2005) when the Chess Grandmaster Ruslan Ponomariov became the last human to defeat a top chess AI (Fritz) under tournament conditions. Since 2005 multiple moments of human brilliance has led to the defeat of various top “Chess Engines” (AI) however these did not occur under tournament conditions. The ways in which modern Chess Engines operate has also leading to a larger divide in skill between Humans and AI (McIlroy-Young et al., 2020) and their subsequent sharp rise in ability has led to an alteration in the ways in which we conceptualize chess.

The competitive nature of Chess, whether it be Humans playing Humans, AI playing AI or Humans playing AI has been the focus of countless statistical studies and multiple ranking metrics have been created surrounding the game (Grabner, 2014; Blanch, Aluja & Badia, 2016). The most impactful of which is the statistical ranking ELO, which is an ‘objective and valid indicator of players’ expertise level in terms of an international performance ranking system’ (Grabner, 2014 p. 27). ELO computes an expected score E based on the differences of rP - rO between the ratings of a player P and the ratings of various opponents O and updates rP according to the difference between the actual score and E (Regan & McC. Haworth, 2011). This ranking system however has also been used to track and display the performance of various different chess AI (McIlroy-Young et al., 2020; AI impacts, 2018), both in comparison to other AI and in comparison to Humans.

Figure 1: Chess AI Rating Over Time (AI impacts, 2018)

The above graph displays the ELO rating of the top performing Chess AI since 1985, it isn’t till approximately 2005 that the rating of the top Chess AI surpassed that of the top performing Chess Grandmasters colloquially known as “Super Grandmasters” which generally hold an ELO rating of 2700+ with some reaching the high 2800 range (Ninan, 2021). This graph and statistical methods behind it adequately present the timeline of AI’s growth and subsequent mastery of chess, fully displaying that despite Deep Blue’s success AI had not truly managed to become dominant till several years later.

References:

AI impacts. (2018). Historic trends in chess AI. Retrieved 13 November 2021, from https://aiimpacts.org/historic-trends-in-chess-ai/

Blanch, A., Aluja, A., & Badia, J. (2016). Evaluating the association of age with standard and blitz Elo chess ratings: Sex and country variability. Personality And Individual Differences, 101, 468. doi: 10.1016/j.paid.2016.05.093

Campbell, M., Hoane, A., & Hsu, F. (2002). Deep Blue. Artificial Intelligence, 134(1-2), 57-83. https://doi.org/10.1016/s0004-3702(01)00129-1

Grabner, R. (2014). The role of intelligence for performance in the prototypical expertise domain of chess. Intelligence, 45, 26-33. doi: 10.1016/j.intell.2013.07.023

Korf, R. (1997). Does Deep Blue use Artificial Intelligence?1. ICGA Journal, 20(4), 243-245. https://doi.org/10.3233/icg-1997-20404

Levy, D. (2005). II People vs Computers World Chess Team Match. Chessbase. Retrieved 13 November 2021, from https://en.chessbase.com/post/bilbao-the-humans-strike-back.

McIlroy-Young, R., Sen, S., Kleinberg, J., & Anderson, A. (2020). Aligning Superhuman AI with Human Behavior. Proceedings Of The 26Th ACM SIGKDD International Conference On Knowledge Discovery & Data Mining. https://doi.org/10.1145/3394486.3403219

Ninan, S. (2021). To GM or not to GM: Inside calls for FIDE to change Grandmaster requirements. ESPN.

Regan, K., & McC. Haworth, G. (2011). Intrinsic Chess Ratings. Association for the Advancement of Artificial Intelligence.