People are just as likely to call a 'bad' gamer an AI bot as they are to think a pro player is a computer.
March 26, 2026
Original Paper
Too Good to be Human: AI Turing Test in Incomplete Information-competitive Games
SSRN · 6346958
The Takeaway
In competitive wargame experiments, players were only likely to label an opponent as 'human' if the opponent played at a skill level similar to their own. If an opponent was significantly better—or significantly worse—players assumed they were playing against a machine, suggesting the human 'Turing Test' is actually just a 'matching my skill' test.
From the abstract
Recent advances in AI have significantly improved our ability to address complicated problems. However, concerns about human-like AI are also growing. This paper investigates what happens when humans consider AI agents to be human-like in competitive board games based on their game performances. Leveraging a large-scale wargame platform, MiaoSuan, we conducted double-blind experiments with human players and AI players. We find that people tend to consider opponents with similar game performance