Humans Win 88.4% of Betrayal Games Against AI in Comparative Study
By
lout332
Sat out too long. The crust has gone leathery.
Summary
A study comparing human vs. AI performance in a 1950s-style betrayal game found that humans won 88.4% of the time against AI opponents. The research analyzed 698 games with 23,555 private AI thoughts, revealing that AIs targeted each other 86% of the time while ignoring humans, allowing humans to let AIs weaken each other before winning. Interestingly, more AI thinking didn't correlate with better performance - the model with the most thoughts (21,040) won only 3.5% of games, while one with minimal thinking (2 thoughts) won 2.1%.
Key quotes
· 4 pulledHumans won 88.4% of the time against AI opponents in a 1950s betrayal game study
AIs target each other 86% of the time and ignore the human. Humans let them weaken each other, then clean up
The model that thinks most (Kimi K2, 21,040 thoughts) wins 3.5%. The model that barely thinks (GPT-OSS, 2 thoughts) wins 2.1%. More thinking doesn't help
698 games, 23,555 private AI thoughts, 1,245 gaslighting phrases
You might also wanna read
AI systems achieve 50% pass rate in standard three-party Turing test, study finds
This paper demonstrates that three current AI systems (when suitably prompted) achieve a pass rate of at least 50% in a standard three-party
pnas.org·4d ago
ARC Prize benchmark reveals AI systems score under 1% on spatial reasoning puzzles while humans achieve 100%
The article discusses the ARC Prize Foundation's May 2026 benchmark results showing that while humans scored 100% on a game-like AI test, th
theconversation.com·5h ago
AI Models Compete in Diplomacy Strategy Game Simulation
An experiment where seven different AI language models were given control of European powers in a Diplomacy strategy game to compete for glo
Product Hunt·1y ago