I agree with Oz. I suspect this is just the tip of the iceberg.
With respect to Azul, there are two types of AI assistance we're currently seeing:
a) AI-assisted players: People using AI to help make decisions during their own games. Unfortunately, I don't think there's much BGA can realistically do to prevent this.
b) Fully automated bots: AI systems that simulate human players and play games without human involvement.
From what we've seen in our Discord community, many of the people developing these bots appear to be doing so for academic or research purposes. BGA provides a convenient platform for testing their bots against real players and measuring their strength.
Unfortunately, some of these projects are open source or implemented in readily accessible JavaScript, making them easy for others to adopt or modify. I suspect this is contributing to the increasing number of bots we're seeing.
If BGA wants to tackle this problem, I think there are several practical measures worth considering.
1) Introduce a suspected AI usage flag: Allow players to privately flag opponents they suspect of using AI. Multiple independent flags could help BGA identify suspicious accounts and prioritise investigations. These flags should not be publicly visible or automatically result in sanctions.
2) Improve reporting mechanism: Provide a clearly identified mechanism for reporting suspected AI or bot assistance, directly linked to the game concerned.
3) Introduce proactive analysis: Automatically analyse high-ranking Arena players and rapidly advancing accounts for suspicious patterns of AI-assisted play, using AI to lighten the human impact.
4) Empower trusted game specialists: Recruit experienced and respected community members to assist existing moderators with investigating suspected AI cheating. BGA staff are unlikely to have the resources or game-specific expertise to investigate every case. Individuals such as Oz in Heat and Prip in Azul are deeply committed to their respective communities and have earned considerable trust. These specialists could help review evidence and recommend temporary suspensions or the invalidation of results, subject to safeguards.
These final two suggestions are specifically aimed at fully automated bots rather than players using AI assistance, and may be a little extreme.
5) Introduce player-requested bot challenges: Allow players in high-ranking Arena games to request a CAPTCHA or equivalent human-verification challenge once per opponent per game.
6) Allow switching to turn-based play: Give players a limited number of jokers per Arena season to convert a live game into a turn-based game. From what we've seen, current bots appear to be designed for real-time play and may not support turn-based games. This could provide another way to disrupt automated play.
I appreciate that none of these measures would completely eliminate cheating. The situation is only likely to become more difficult as AI engines improve and become more accessible. Like others, I think BGA needs to start addressing it more proactively.
With respect to Azul, there are two types of AI assistance we're currently seeing:
a) AI-assisted players: People using AI to help make decisions during their own games. Unfortunately, I don't think there's much BGA can realistically do to prevent this.
b) Fully automated bots: AI systems that simulate human players and play games without human involvement.
From what we've seen in our Discord community, many of the people developing these bots appear to be doing so for academic or research purposes. BGA provides a convenient platform for testing their bots against real players and measuring their strength.
Unfortunately, some of these projects are open source or implemented in readily accessible JavaScript, making them easy for others to adopt or modify. I suspect this is contributing to the increasing number of bots we're seeing.
If BGA wants to tackle this problem, I think there are several practical measures worth considering.
1) Introduce a suspected AI usage flag: Allow players to privately flag opponents they suspect of using AI. Multiple independent flags could help BGA identify suspicious accounts and prioritise investigations. These flags should not be publicly visible or automatically result in sanctions.
2) Improve reporting mechanism: Provide a clearly identified mechanism for reporting suspected AI or bot assistance, directly linked to the game concerned.
3) Introduce proactive analysis: Automatically analyse high-ranking Arena players and rapidly advancing accounts for suspicious patterns of AI-assisted play, using AI to lighten the human impact.
4) Empower trusted game specialists: Recruit experienced and respected community members to assist existing moderators with investigating suspected AI cheating. BGA staff are unlikely to have the resources or game-specific expertise to investigate every case. Individuals such as Oz in Heat and Prip in Azul are deeply committed to their respective communities and have earned considerable trust. These specialists could help review evidence and recommend temporary suspensions or the invalidation of results, subject to safeguards.
These final two suggestions are specifically aimed at fully automated bots rather than players using AI assistance, and may be a little extreme.
5) Introduce player-requested bot challenges: Allow players in high-ranking Arena games to request a CAPTCHA or equivalent human-verification challenge once per opponent per game.
6) Allow switching to turn-based play: Give players a limited number of jokers per Arena season to convert a live game into a turn-based game. From what we've seen, current bots appear to be designed for real-time play and may not support turn-based games. This could provide another way to disrupt automated play.
I appreciate that none of these measures would completely eliminate cheating. The situation is only likely to become more difficult as AI engines improve and become more accessible. Like others, I think BGA needs to start addressing it more proactively.
