Adaptive, LLM-Generated Explanations in AI-Assisted Decision-Making: Do They Improve Human Performance and Appropriate Trust?

This research proposal investigates whether adaptive, LLM-generated explanations can improve human decision-making and appropriate trust in AI-assisted decision-making. The proposed study compares three conditions: no explanation, standard LLM-generated explanation, and adaptive LLM-generated explanation. It examines decision accuracy, appropriate reliance, error recovery, trust calibration, confidence, and perceived clarity across different task modalities. The study has not yet been conducted; no participants have been recruited and no experimental data have been collected or analyzed. The manuscript is structured as a Stage 1 Registered Report, with the results and discussion intended for a future Stage 2 study.