In this seminar, Yue Zhang will give a talk about AI-Augmented Human Decision Making: Bel ief Formation, Choice, and Evidence from Medical Diagnosis.
< p>Human decision makers (HDMs) increasingly rely on artificial intell igence (AI) systems for decision support in high-stakes contexts, yet how HDMs integrate AI recommendations into their decision-making pro cess remains insufficiently understood. Observed responses to AI assi stance confound how humans form beliefs from available information wi th how they translate those beliefs into decisions, making it difficu lt to identify the behavioral mechanisms underlying human-AI collabor ation. We develop a structural model of AI-augmented human decision m aking that separates belief formation from stochastic choice. The fra mework combines a general belief-aggregation layer, which allows huma n and AI evidence to receive different weights and interact, with a s tochastic choice layer that captures imperfect translation of beliefs into actions. The analysis shows that belief distortion and choice n oise represent distinct but interacting behavioral mechanisms: while deterministic decisions can be insensitive to small belief distortion s, stochastic choice makes decision probabilities continuously respon sive to belief distortions. We estimate the model using experimental data from a medical diagnostic task in which endoscopists classify co lonoscopy cases with and without AI assistance. The results show that HDMs incorporate both their own diagnostic evidence and AI recommend ations, but systematically underweight AI recommendations relative to the normative Bayesian benchmark. Moreover, compared with non-expert s, expert HDMs place greater weight on their own diagnostic evidence while maintaining similar responsiveness to AI, indicating that exper tise affects the belief-aggregation process itself rather than merely the strength of human evidence. Finally, by recovering the full choi ce probability rather than only realized decisions, the model enables evaluation of AI’s decision impact beyond observed sample paths an d a probabilistic assessment of AI-assisted diagnostic performance. DESCRIPTION: In this seminar, Yue Zhang will give a talk about AI-Augm ented Human Decision Making: Belief Formation, Choice, and Evidence f rom Medical Diagnosis. Human decision makers (HDMs) increasingly rely on artificial intelligence (AI) systems for decision support in high -stakes contexts, yet how HDMs integrate AI recommendations into thei r decision-making process remains insufficiently understood. Observed responses to AI assistance confound how humans form beliefs from ava ilable information with how they translate those beliefs into decisio ns, making it difficult to identify the behavioral mechanisms underly ing human-AI collaboration. We develop a structural model of AI-augme nted human decision making that separates belief formation from stoch astic choice. The framework combines a general belief-aggregation lay er, which allows human and AI evidence to receive different weights a nd interact, with a stochastic choice layer that captures imperfect t ranslation of beliefs into actions. The analysis shows that belief di stortion and choice noise represent distinct but interacting behavior al mechanisms: while deterministic decisions can be insensitive to sm all belief distortions, stochastic choice makes decision probabilitie s continuously responsive to belief distortions. We estimate the mode l using experimental data from a medical diagnostic task in which end oscopists classify colonoscopy cases with and without AI assistance. The results show that HDMs incorporate both their own diagnostic evid ence and AI recommendations, but systematically underweight AI recomm endations relative to the normative Bayesian benchmark. Moreover, com pared with non-experts, expert HDMs place greater weight on their own diagnostic evidence while maintaining similar responsiveness to AI, indicating that expertise affects the belief-aggregation process itse lf rather than merely the strength of human evidence. Finally, by rec overing the full choice probability rather than only realized decisio ns, the model enables evaluation of AI’s decision impact beyond obs erved sample paths and a probabilistic assessment of AI-assisted diag nostic performance. END:VEVENT END:VCALENDAR