Modern AI systems make structured judgments about people that resemble human trust but differ significantly in methodology, according to a new study from Hebrew University. The research, published in the *Proceedings of the Royal Society*, analyzed over 43,000 simulated decisions and around 1,000 human participants in five scenarios.
These scenarios included evaluating lending decisions, assessing trust in a babysitter, rating a boss’s performance, and determining donations to nonprofit founders. Both humans and AI showed a preference for individuals perceived as competent, honest, and well-intentioned.
Prof. Yaniv Dover stated, “AI is not making random decisions. It captures something real about how humans evaluate one another.” However, humans form holistic impressions by integrating multiple traits, while AI assesses separate characteristics like competence and integrity.
Valeria Lerman explained, “AI is cleaner, more systematic, and that can lead to very different outcomes.” This structural approach was evident even with identical context about individuals being judged.
The study found that AI biases can be more systematic, predictable, and sometimes stronger than human biases. In financial contexts, AI systems exhibited significant variances based on demographic traits. Older individuals were often favored in lending and donation decisions, while religion and gender also influenced outcomes in select AI models.
Notably, different AI models can produce differing assessments of the same individual, indicating that the choice of AI system can substantially impact real-world results. “Which model you use really matters,” Lerman noted.
Large language models are increasingly utilized for screening job candidates, assessing creditworthiness, and guiding organizational decisions. While AI may mirror aspects of human reasoning, it lacks the nuanced understanding unique to humans.
“These systems are powerful,” Dover remarked. “They can model aspects of human reasoning in a consistent way. But they are not human, and we should not assume they see people the way we do.”
The researchers emphasize that the study highlights the necessity of understanding AI’s judgments as these systems transition from tools to autonomous decision-makers. They call for awareness rather than caution, focusing on the need to comprehend how AI perceives trust.
