Decision Referrals in Human-Automation Teams

Explore a research seminar from GERAD Research Center that delves into the optimization of human-automation collaboration in decision-making processes. Learn how computerized decision support systems can be designed to effectively share tasks with human operators across various applications including defense surveillance, business risk assessment, and medical diagnostics. Understand the critical challenge of determining when automation should handle tasks independently versus referring them to human operators, particularly considering human cognitive workload limitations. Examine a mathematical model for binary classification tasks that accounts for declining human performance under increased workload, and discover how optimal referral policies can be developed using measurable human error probabilities rather than complex cognitive models. Investigate how signal detection theory from psychology applies to this framework, and analyze extensions that incorporate human trust dynamics in automation. Master the implementation of efficient algorithms for computing cost-minimizing referral policies and the conditions under which myopic optimization approaches achieve global optimality.