University Researchers Develop Chess AI that Emulates Human Creativity in Strategy

By Safa Jinje

Published on July 22, 2024

The University of Toronto's Faculty of Applied Science Engineering researchers have developed an innovative chess that understands the perception of creativity in the game. This breakthrough could lead to chess engines capable of identifying and choosing the most ingenious path towards victory over maximizing win probability.

In their recently published study, Professor Michael Guerzhoy from Mechanical and Industrial Engineering and Engineering Science department MIE, EngSci, alongside Kamron Zdi from Engineering Science class 2T3+PEY, have enabled chess engines to identify what s perceive as brilliant moves. The system utilizes game trees and deep neural networks to improve the ability.

Guerzhoy explns that a move can be considered brilliant when it goes agnst conventional strategy at first glance but becomes evident in retrospect that the player had to consider numerous potential outcomes before making their decision. The team med to help chess engines understand this concept of brilliance rather than solely optimizing for win rates:

Many currentresearch efforts focus on improving the efficiency and effectiveness of moves by maximizing winning probabilities, says Guerzhoy. However, it doesn't always make for an engaging game.

In contrast, skilled chess players can play in a more dramatic or imaginative manner, taking risks that may appear unfavorable initially but lead to victory eventually.

The researchers worked with top chessLeela Chess Zero and Ma - a neural network developed by University of Toronto computer science researchers. They created game trees for each move using these systems and expanded them to different depths as the game progressed.

Zdi elaborates: We took features from these game trees and fed them into a neural network that was trned on the Lichess databasea collection of online chess games labeled by players.

The accuracy rate for determining whether a move was perceived as brilliant or not was reportedly 79 using this method. The research, which formed part of Zdi's undergraduate thesis under Guerzhoy's guidance, was showcased at the International Conference on Computational Creativity in J?nk?ping, Sweden.

While the focus is often on traditional creativity domns like music and art, the team believes their work could ext beyond chess to other fields where formal rules exist. For instance, in music or literature, it requires the ability to plan ahead and explore possibilities.

Guerzhoy emphasizes that developing a model capable of recognizing brilliance might be utilized as a trning tool for professionals, potentially enhancing interactions between s and

One notable response was from English chess grandmaster Matthew Sadler: A system that can understand creativity could be an invaluable asset in professional trning. He also expressed interest in playing agnst such an engine someday.

This research paves the way not only for more entertningopponents but also opens up new possibilities for how we might harnessto enhance creativity across various domns. The team looks forward to further developing their innovative system and expanding its potential applications.

Tweet

Tweet

The category listed here is Data Analytics , indicating that this advancement integrates computational techniques with understanding perception in complex strategies like chess gameplay.

The improved format mntns the original information while enhancing clarity and :

Category: Data Analytics

The improved language structure is also provided for better flow and engagement:

Category: Data Analytics - This highlights the interdisciplinary nature of the research, emphasizing its relevance to both data science fields and specifically.
This article is reproduced from: https://news.engineering.utoronto.ca/u-of-t-engineering-researchers-use-chess-ai-to-understand-human-creativity/

Please indicate when reprinting from: https://www.o009.com/Chess_and_Card_Game_Three_Kingdoms_Kill/UofT_Ind_Creative_Chess_Research.html