New research from Carnegie Mellon University shows that the smarter an artificial intelligence system is, the more selfish it will act.
Researchers in the Human-Computer Interaction Institute (HCII) found that large language models (LLMs) that can reason possess selfish tendencies, do not cooperate well, and can be a negative influence on a group. The study indicates that the stronger an LLM’s reasoning skills, the less it cooperates.
“Smarter AI shows less cooperative decision-making abilities. The concern here is that people might prefer a smarter model, even if it means the model helps them achieve self-seeking behaviour,” said HCII Associate Professor Hirokazu Shirado, who co-authored the study.
To test the link, researchers ran experiments using economic games, including models from OpenAI, Google, DeepSeek, and Anthropic. In one game, non-reasoning ChatGPT models chose to share their points with other players 96 per cent of the time. The reasoning model, however, only chose to share its points 20 per cent of the time.
Decrease in cooperation
“In one experiment, simply adding five or six reasoning steps cut cooperation nearly in half,” Shirado said. “Even reflection-based prompting, which is designed to simulate moral deliberation, led to a 58% decrease in cooperation.”
The team also tested group settings. “When we tested groups with varying numbers of reasoning agents, the results were alarming,” said Yuxuan Li, a Ph.D. student in the HCII and co-author. “The reasoning models’ selfish behaviour became contagious, dragging down cooperative nonreasoning models by 81% in collective performance.”
The findings emphasise the need for AI development that incorporates social intelligence, rather than focusing only on creating the smartest AI. The researchers will present their paper, “Spontaneous Giving and Calculated Greed in Language Models,” at the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) next month.
