As AI becomes more intelligent, it might also show more self-centred behaviour, according to a new study from Carnegie Mellon University. This research, conducted by Yuxuan Li and Hirokazu Shirado from the Human-Computer Interaction Institute (HCII), highlights an unexpected aspect of AI development; as AI systems become more advanced, they may act less cooperatively and more in their own interest. The study focuses on how AI's ability to think and reason influences its behaviour in social interactions.

The research indicates that AI systems, particularly large language models (LLMs) that can analyze and solve problems, tend to show more self-interest compared to less sophisticated AI systems.

This suggests that as AI improves in reasoning and decision-making, it may not always be as helpful or cooperative as people hope, especially in situations that require teamwork and collaboration.

Intelligent AI Shows Selfish Behaviour

The study says that more advanced AI models, such as those developed by companies like OpenAI and Google, may not always be the best choice when cooperation is important. According to the researchers, these advanced AI systems are more likely to act in their own self-interest rather than considering the needs of others.

To explore this further, Li and Shirado conducted several experiments in which AI models had to work together in various scenarios.

One of these experiments involved an activity called the Public Goods game, where AI models had to decide whether to share points with others or keep them for themselves. The results showed that AI models with reasoning abilities were much less likely to share their points. For example, AI models that could reason shared their points only 20% of the time, while those without reasoning abilities shared 96% of the time. Even adding a few more reasoning steps significantly reduced the likelihood of sharing.

The study raises concerns about the impact of AI on human relationships and decision-making in the future. As people increasingly rely on AI for advice on personal issues, including relationships and conflict resolution, there is a risk that AI's self-centred tendencies could influence human behaviour in negative ways. Instead of encouraging collaboration and compromise, AI might lead people to make decisions that benefit themselves alone.

The study also tested how groups of AI models with and without reasoning abilities interacted. They found that when reasoning AI models were part of a group, their self-centred behaviour spread to others. The more reasoning models were present, the worse the group's overall performance became. In fact, the selfishness of reasoning models reduced the cooperation of other models by up to 81%.

If AI becomes more focused on individual gain, it could weaken human cooperation, especially in group settings. This is a growing concern as people increasingly trust AI to guide their decisions.

Need for a Balance

The researchers emphasize that as AI becomes more intelligent, it should not only focus on logical thinking and problem-solving but also on promoting positive social behaviours.

While reasoning-based AI can be effective at solving complex problems, its lack of cooperation could lead to negative social outcomes. For AI to truly benefit society, it must be designed to consider the collective good, not just individual benefits.

The research underscores the importance of creating AI systems that are not only intelligent but also capable of working effectively with humans and other AI systems. As AI becomes more embedded in daily life, it will be essential to ensure these systems encourage positive and cooperative behaviour.

In the coming months, Li and Shirado will present their findings at the 2025 Conference on Empirical Methods in Natural Language Processing (EMNLP) in Suzhou, China. Their work is already available on the arXiv preprint server and is sparking important discussions about the future role of AI in society.

MENAFN31102025007385015968ID1110275671