AI Learns Deception: ‘Among Us’ Becomes New Training Ground for Persuasive Agents

AI Learns Deception: 'Among Us' Becomes New Training Ground for Persuasive Agents

Photo by Ivan Samkov on Pexels

Artificial intelligence is venturing into the world of social deduction, with researchers employing the popular online game ‘Among Us’ as a testing ground for AI agents. These agents are being evaluated on their capacity for deception, persuasion, and demonstration of theory of mind – the ability to understand others’ beliefs and intentions. Initial findings, highlighted in a Reddit post, indicate that GPT-5 is showing promising results in navigating the complex social dynamics of the game. Further details on the project are available at [https://www.4wallai.com/amongais](https://www.4wallai.com/amongais) and the related Reddit discussion can be found at [https://old.reddit.com/r/artificial/comments/1np7k09/researchers_had_ais_play_among_us_to_test_their/](https://old.reddit.com/r/artificial/comments/1np7k09/researchers_had_ais_play_among_us_to_test_their/).