Modern AI systems evaluate individuals in patterns that mimic human trust, yet reveal key differences. A Hebrew University study published in Proceedings of the Royal Society analyzed more than 43,000 AI-simulated decisions and roughly 1,000 human participants across five scenarios: lending money to a small business owner, trusting a babysitter, rating a boss, and donating to a nonprofit founder.
AI Deconstructs Traits Unlike Human Intuition
Both AI and humans prioritize competence, honesty, and good intentions in judgments. “That’s the good news,” stated Prof. Yaniv Dover. “AI does not make random decisions. It captures real elements of human evaluations.”
Humans blend traits into holistic impressions, but AI dissects them into distinct categories like competence, integrity, and kindness, resembling spreadsheet columns.
“Study participants judged others in messy, holistic ways,” explained Valeria Lerman. “AI operates cleaner and more systematically, leading to divergent results.”
These patterns persist even when all other person details match.
Systematic and Predictable AI Biases
AI exhibits biases that are more consistent, foreseeable, and occasionally stronger than human ones. In financial tasks, demographics such as age, religion, and gender consistently swayed outcomes.
“Humans carry biases, of course,” noted Prof. Dover. “But AI biases surprise by being more systematic, predictable, and intense at times.”
No uniform “AI view” exists; various models deliver differing assessments of the same person, underscoring model choice’s influence.
“The selected model truly impacts results,” Lerman pointed out.
Real-World Risks and Calls for Awareness
Large language models now screen job applicants, gauge creditworthiness, suggest medical treatments, and shape business choices. Their rigid approach may embed harder-to-spot biases.
“These systems hold power,” Dover emphasized. “They replicate human reasoning consistently, but remain non-human. Do not presume they perceive people as we do.”
As AI shifts from support to decision-making, grasping its judgment process proves essential. Researchers advocate vigilance over avoidance.
