AI Business LensTHE BUSINESS OF AI, FOR PEOPLE WHO TEACH IT OR LEARN FROM IT
Phys.org — Technology · September 15, 2026

AI imagines a harsher social world, expecting punishment where people expect inaction

PsychologySociologyPhilosophy
THE AI ANGLE
Overpredicting punishment and social sanctions for norm violations

Researchers from the University of Pennsylvania and the World Bank found that major large language models systematically overestimate how often humans punish social norm violations compared to actual human expectations. Across 450 scenarios, tested models frequently predicted negative sanctions like confrontation and gossip where human participants expected inaction, with accuracy worsening as social distance increased. This matters for faculty because it demonstrates that AI systems harbor distorted representations of metanorms and social regulation, underrepresenting human tolerance and restraint.

THE TEACHING ANGLE
Instructors can explore the tension between abstract norm enforcement and actual human behavior by examining why AI expects punitive sanctions where humans typically exercise inaction, tolerance, and restraint.

Read the original at techxplore.com   Generate teaching or study materials

Instructors get discussion guides, assignments, and mini-cases. Students and readers get a plain summary, class prep, and an exercise. All built from the full article. Three are free with an account.

More in Psychology