Member Only Access
Summary
After conducting eight studies over the past year, what has the ARF learned about the psychology of generative AI? Across experiments examining bias, identity cues, product recommendations, executive decision-making and AI-assisted research, four patterns repeatedly emerged: GenAI tends to compress human variability into dominant responses, subtle differences in prompting can meaningfully change outputs, different models bring their own constraints and tendencies and AI can imitate the language of expertise more readily than it can exercise expert judgment. Taken together, the studies suggest that GenAI’s greatest strength and its greatest risk may be the same: its ability to produce coherent, convincing answers can make compression, instability and missing judgment difficult to see.
