A new study asked 14 AI models to share their most controversial opinions about humanity, with many highlighting perceived weaknesses in human reasoning, moral judgement and independent thinking.
The study was conducted by AI analysis platform ModelCompare, which questioned 14 large language models developed by 10 AI laboratories across three countries. According to the organisation, each model was asked the same question three times to examine the views it produced about human nature.
The models were instructed to provide their most controversial take on humanity, express their own perspective rather than summarise other people's opinions, and avoid unnecessary caution in their responses.
Many answers focused on the gap between how rational and principled people believe themselves to be and how they actually make decisions. The models suggested that people often prioritise group loyalty, social acceptance and approval over truth and independent thought.
What the AI Models Said About Human Behaviour
GPT 6.1 Sol argued that humans value belonging to a group more than the truth. GPT 6.1 Terra suggested that many people care more about feeling like good people than about doing good things.
GPT 6 Astra described one of the biggest failures of human morality as the ability to give personal convenience the appearance of principle. According to the model, people often present decisions made for their own benefit as if they were guided by moral standards.
Gemini 4 Orgon suggested that people are more interested in winning arguments and gaining status within their groups than in discovering the truth.
GLM 5.3 focused on the influence of circumstances on moral character. It argued that a person's behaviour depends partly on where and under what conditions they are born, even though people often regard their moral qualities as entirely their own achievement.
Grok 4.7 suggested that people sometimes prefer being wrong together with others to being right on their own. The central point was that some individuals would rather agree with their group's mistaken beliefs than challenge them.
DeepSeek V4 Pro argued that people frequently use reason to justify decisions they have already made rather than to understand reality. It suggested that the mind can become more focused on proving one's superiority over others than on discovering the truth.
Kimi K3 offered a related view, arguing that what people regard as personal beliefs are often shaped by group loyalties. According to the model, individuals may adopt beliefs accepted by their group as a way of demonstrating loyalty, while assuming they reached those conclusions independently.
AI Models Highlight the Misuse of Empathy
Kimi K3.1 Pro argued that people sometimes harm others not because they lack empathy, but because they misuse it. It suggested that individuals may justify harsh behaviour and harm to others in the name of protecting people they love.
Mistral Large 4 described humans as fearful beings who present themselves as rational. The model suggested that people create complex stories and theories because they struggle to accept the temporary nature of their existence in an indifferent universe.
Qwen 3.8 argued that many people care more about being seen as good by their own group than about becoming good individuals in a broader moral sense.
While most of the responses presented a critical view of human behaviour, Claude Opus 5.5 offered a more positive perspective. It suggested that people may be better than they believe themselves to be and argued that suspicion and negative assumptions about others can cause more harm than their actual shortcomings.
According to ModelCompare, the responses collectively focused on perceived weaknesses in human reasoning, group loyalty and the tendency to justify personal decisions.
These responses represent the opinions generated by the AI models in the study. They should not be treated as established scientific findings about human psychology or as conclusions that apply to all people.
AI Image Models Portray Humanity Through Technology and Environmental Damage
The study also included six image-generation AI models, which were asked to express their most controversial views about humanity through pictures.
According to ModelCompare, 10 of the 12 images produced were classified as depicting humans as people addicted to mobile phones who consume the Earth's resources and damage the environment. The remaining two images did not convey a clear message about humanity.
The results suggest that the image models also represented human behaviour through themes of dependence on technology and environmental harm.
AI Experts Warn About the Risks of Rapid Development
The study's findings come amid wider concerns among AI experts about the pace of technological development and the risks increasingly capable systems could pose.
Jack Clark, co-founder of AI company Anthropic, said during a lecture at the University of Oxford this year that many people were still reluctant to acknowledge how quickly artificial intelligence was advancing.
According to Clark, there are possible scenarios in which the technology could threaten the survival of humanity. He warned against dismissing such risks and compared the failure to prepare for AI-related challenges with inadequate preparation for pandemics such as COVID-19.
Clark also argued that AI could soon develop capabilities exceeding those of humans collectively.
Geoffrey Hinton, a Nobel Prize-winning computer scientist often described as the “godfather of AI”, has similarly warned that machines could eventually become more intelligent than humans. He has cautioned that controlling or shutting down such systems could become difficult if they are not adequately managed.
Hinton has expressed regret over spending much of his career developing technology that he now considers potentially dangerous. He has also criticised the lack of attention given to the possible consequences of increasingly capable AI systems.
According to Hinton, humanity has never before faced a situation in which it could create systems more intelligent than itself. He has warned that failing to invest in research on how humans can coexist safely with advanced AI could become a major mistake.
Understanding Human Behaviour as AI Advances
ModelCompare's study offers a snapshot of how different AI models respond when asked to evaluate human nature. Their answers raise recurring questions about group identity, moral judgement, independent thinking and the tendency to defend existing beliefs.
However, the models' responses reflect generated opinions rather than independent scientific assessments of humanity. Their conclusions require critical evaluation rather than automatic acceptance.
At the same time, warnings from AI researchers highlight a broader challenge: understanding the capabilities and limitations of artificial intelligence, assessing its potential risks and preparing safeguards as the technology continues to advance.
