AI WILL SERVE AS A MORE VALUABLE SPARRING PARTNER FOR “ORGANIC” CHESS PLAYERS
"AI can only approximate the precision of a mathematically calculated algorithm for finding the strongest move, but not fully achieve it," Dmitri Dzhanhirov
AI WILL SERVE AS A MORE VALUABLE SPARRING PARTNER FOR “ORGANIC” CHESS PLAYERS
Interview with Dmitri Dzhanhirov
by Anna Fiero
“I believe that AI will not be able to surpass chess programs in strength, because intellectual activity can only approximate the precision of a mathematically calculated algorithm for finding the strongest move, but not fully achieve it”.
Dmitri Dzhanhirov is a candidate for the Master of Sports in Chess for the USSR and Ukraine and a National Master in the USA. For many years, he was the First Vice President of the Kyiv Chess Federation and organized 13 annual international Nabokov Memorial chess tournaments in Kyiv, featuring norms for International Master and Grandmaster titles. As part of these tournaments, he held chess problem-solving contests, serving as a jury member.
Image created by ChatGPT
AF: It is commonly believed that AI has demonstrated itself most prominently in chess, far surpassing the strongest human players. Is that true?
DD: This is indeed a popular opinion, but let’s analyze it. In chess, humans and chess programs, and AI, follow the fundamental principle of finding the strongest move in a given position. However, behind this seemingly simple task lie extremely complex algorithms for solving it. Chess programs have undergone significant evolution—from the first program that played on a 6x6 board (without bishops) in 1952, to the first full-fledged chess program in 1957, to the first World Computer Chess Championship in 1974, where the strongest programs played at roughly the level of a third-category player. This journey continued through the first blitz victory against a world champion in 1994, the first classical game victory over a world champion in 1996, and the first match victory in a six-game series against a world champion (in all cases, Garry Kasparov was the defeated player). In the 21st century, the strength of chess programs has far surpassed human capabilities, shifting the dynamic from competition to collaboration—players use chess programs to analyze their games and specific positions, prepare openings, and refine their endgame skills. At this point, computers have analyzed all endgames with seven or fewer pieces. To train, chess players set chess programs to levels matching or slightly exceeding their playing strength. Initially, it was believed that the development of computer programs would follow the path of mimicking human chess thought—creating an AI-like system (although the term AI existed, it was rarely used in reference to chess programs). In the 1960s–80s, with slow computer speeds, brute-force calculation ("brute force") was not sufficient for deep analysis of positions. Some researchers, notably former world champion Mikhail Botvinnik, attempted to model a "human" type of analysis, where many possible moves were preemptively discarded, significantly reducing the breadth of the decision tree and allowing deeper calculations. Botvinnik set an ambitious goal: to create a program that played "grandmaster-level" chess. His team improved the accuracy of position evaluation functions and developed the "Pioneer" program, which could solve complex chess studies (artificial positions requiring "White to win" or "White to draw") that other computer programs could not handle. However, in practical play, "Pioneer" was weaker than existing chess engines.
AF: As we understand, the "Pioneer" project, which was a prototype of AI, was shut down in 1994 after Botvinnik retired.
DD: It wasn’t about Botvinnik himself. Unfortunately, the initially promising project turned out to be a dead end for computer chess development. The exponential increase in processor speed ended the debate about whether chess programs could model human thinking—brute force triumphed over AI prototypes, with calculation depth surpassing 20, then 30, and now 40 half-moves (a "half-move" is one move by White or Black; a "move" in chess consists of a White move followed by a Black response or vice versa). Programmers, in collaboration with chess players, were left to improve position evaluation functions at the end of calculations. But at this point, there was no longer any talk of AI.
AF: Let’s clarify—can modern AI play chess?
DD: Recently, from January 8–14, 2025, well-known chess and AI popularizer Levy Rozman organized a tournament between seven popular chatbots and one of the strongest chess engines, Stockfish. In the final, Stockfish defeated ChatGPT. However, the defining characteristic of this tournament was not Stockfish’s expected victory but rather that chatbots, after relatively reasonable opening play, began making nonsensical or even illegal moves: - Their pieces made impossible moves; - Chatbots attempted to move opponent’s pieces or "capture" their own; - Chatbots placed new pieces on the board, etc. It is important to note that opening databases and recommendations for early-game play are easily accessible online, and it is not difficult to train chatbots to reference endgame tablebases ("Lomonosov tables"). Researchers have even asked ChatGPT to checkmate in two moves in simple king-and-queen vs. king positions, yet the chatbot not only failed but also attempted to use additional pieces or announce a checkmate that wasn’t there. However, the critical part of a chess game occurs after the opening phase and before the transition to a seven-piece endgame. Here, no access to chess literature or databases can substitute for the experience gained from playing and analyzing numerous games. If we abandon brute force, we need algorithms that allow chatbots to learn to play in unfamiliar positions. Perhaps, if AI developers take up this challenge, they will find a worthy solution. And if chatbots learn to play chess through experience, it will validate the claim that chatbots are indeed AI.
AF: What potential applications do you see for blockchain technology in chess?
DD: Currently, I see two major directions for blockchain applications in chess. First, a decentralized chess platform where players can organize their own tournaments and matches—most importantly, on a commercial basis, with prize funds formed from entry fees. Naturally, this involves cryptocurrency wallets, integrating chess into the crypto industry. Players can engage in numerous chess variants, create new ones, take lessons with coaches, give lectures, and more. Today, the pioneer in this field is the crypto project Bit-Chess, which could become a strong competitor to major chess platforms like Chess.com and Lichess. The second direction is anti-cheating measures, particularly preventing the use of chess engines during play. In 2020, Algorand developed a blockchain-based anti-cheating system. The platform records and analyzes over a hundred parameters for each move, including time taken, biometric data of the player, and engine evaluation of the move. This data is stored in a decentralized database, allowing highly accurate detection of cheaters.
AF: Can the relationship between humans and artificial intelligence be compared to that of a creator and their creation?
DD: In many versions of this theme, the human has often turned out to be weaker than their own creation—this is an eternal story that has inspired and continues to inspire writers. In this particular case, I believe that AI will not be able to surpass chess programs in strength, because intellectual activity can only approximate the precision of a mathematically calculated algorithm for finding the strongest move, but not fully achieve it. However, what’s different is that the chess played by AI will be more “human” in spirit rather than purely computer-like. This means that AI, unlike standard computer programs, will be able to perform coaching functions and serve as a more valuable sparring partner for “organic” chess players.