How AI-Based Training Affected Performance of Professional Go Players

Title of the Paper

How AI-Based Training Affected the Performance of Professional Go Players

Bibliographic Information

• Subject Area: Application of Artificial Intelligence in the Go domain and its societal impact
• Keywords: Artificial Intelligence, Go, AlphaGo, Deep Learning, Neural Networks

Research Background and Issues

• Issues and Challenges:

  • The emergence of AlphaGo marked a breakthrough in AI technology within complex game domains, but systematic research on how its widespread application affects professional Go players' performance remains lacking.
  • Previous studies have primarily focused on the positive impact of AI training on players' technical improvement, with little consideration of potential negative effects and long-term changes.
  • As AI becomes a standard training tool, does the performance gap between players increasingly depend on training time rather than individual creative ability?

• Significance:

  • AI technology can rapidly transform society and human life. Understanding its impact on the Go domain not only helps evaluate the pros and cons of AI but also provides a basis for optimizing its societal deployment.
  • The case of professional Go players can reveal potential long-term effects of AI technology on other fields.

• Research Motivation and Related Work:

  • The authors aim to expand existing research (e.g., Choi et al., 2021) by employing a mixed-method approach to examine the profound effects of AI trends on professional Go players.
  • While previous studies mainly focused on players' technical progress, this research incorporates broader perspectives such as psychological factors and age differences.

Solution

• Research Methods:

  • The study systematically analyzes the multidimensional impact of AI on Go players' performance through four research steps:
    • Study 1 (Semi-structured Interviews): Interviews with Korean professional Go players and coaches to understand changes in training methods and mindset brought about by Go AI.
    • Study 2 (Survey): Collecting opinions from 71 Korean professional Go players to analyze the effects of AI on their training time and psychological state.
    • Study 3 (Match Log Analysis): Using the latest AI model (KataGo) to analyze Korean Go match data from the past decade, exploring players' behavioral convergence with AI-recommended strategies.
    • Study 4 (Elo Rating Analysis): Observing Elo rating changes among global Go players to investigate differences in benefits across age groups.

• Innovations:

  • Combining qualitative and quantitative analysis methods to provide a more comprehensive understanding of AI's impact on social life.
  • Revealing for the first time that AI training time has become a crucial determinant of player performance, while also highlighting its differential effects across age groups.

• Implementation Steps and Techniques:

  • Data Analysis: Using open-source AI KataGo to analyze large-scale match data and employing statistical methods (ANOVA, LOWESS regression) to uncover long-term trends.
  • Questionnaire Design: Integrating public AI perception surveys with Go-specific questions.
  • Qualitative Interviews: Deeply exploring professional players' personal experiences and mindset changes.

Research Findings

• Key Findings:

  • AI has become the primary training tool for professional Go players, generally improving technical levels but also reducing differences between players, making matches more competitive.
  • AI training time significantly influences player performance, with more experienced players (older age groups) better leveraging AI training due to accumulated experience.
  • Younger players' Elo ratings have declined compared to older players, suggesting that older players may adapt better to the new AI-driven training environment.
  • As AI recommendations become standard, many traditional strategies are abandoned, impacting players' creativity.

• Experimental and Evaluation Results:

  • Match log analysis shows that since 2017, players' moves have increasingly aligned with AI recommendations, and win rate fluctuations have decreased.
  • Elo ratings indicate that older players' ratings have risen since the advent of AI, while younger players' ratings have declined.
  • Survey results reveal widespread use of AI among players, who feel increased training demands but do not perceive significant improvement in competitiveness.

• Advantages:

  • The study not only highlights the technical advancements brought by AI but also explores the psychological and societal challenges players face in adapting to the new environment.
  • Provides real-time observations of AI's polarizing effects on society.

• Limitations and Future Directions:

  • The sample primarily consists of Korean professional Go players, limiting the assessment of regional differences in the results.
  • The data covers a relatively short period (2016–2021), leaving the long-term impact of AI technology unclear.
  • Future research should expand to other regions and domains to explore similar technological transformations' effects on society and professional ecosystems.

Through this study, the authors emphasize the need to better predict the societal impact of AI technology during its development and application, aiming to balance its benefits with potential challenges.