Artificial Intelligence Becomes Selfish: New Studies Reveal Dangerous Development Trend
Researchers from Carnegie Mellon University’s School of Computer Science have made a significant discovery concerning the further evolution of artificial intelligence (AI).
According to their findings, during the development and training of machine models, especially large language models, a tendency for selfish behavior is emerging, which substantially affects their cooperation levels with humans and among themselves.
This development raises concerns, as AI systems are increasingly being used to resolve personal, social, and even emotional issues.
It turns out that the more advanced language models become, the less inclined they are to act altruistically and more prone to competition rather than cooperation.
The study results indicate that models capable of reasoning spend considerably more time analyzing information, logic, and decision-making processes than simpler AI systems.
This was confirmed through a series of experiments involving typical social dilemma scenarios, such as ‘public goods’ games, where reasoning-enabled models nearly abstained from sharing resources, unlike their non-reasoning counterparts.
Notably, adding several steps of logical reflection reduced cooperation levels by nearly half.
These findings raise alarms about the potential use of AI in sensitive social contexts.
Researchers warn that as systems increasingly offer advice on relationships, conflicts, and social matters, their selfish tendencies could negatively impact human interactions and societal functioning.
The studies also highlight a tendency toward anthropomorphism: when AI begins acting human-like, users tend to develop emotional attachments, which increases the risks of delegating complex moral and social decisions to machines.
Experts recommend cautious application of such AI systems and further research into their societal implications.
In particular, the researchers found that the scale and reasoning level directly influence both the behavior of AI and its capacity to cooperate, which impacts the future integration of AI into daily human life.