Predicting and Understanding Human Action Decisions during Skillful Joint-Action via Machine Learning and Explainable-AI

AI can predict human decisions before people are consciously aware they're making them: The machine learning models successfully predicted whether expert and novice herders would select specific targets at timescales that preceded conscious intent, suggesting human decision-making involves unconscious processes that AI can detect.

Expertise fundamentally changes how people make decisions: The study found that expert and novice actors made decisions based on different information—experts prioritized information about their partner's actions and state, while novices relied on different factors—demonstrating that skill development rewires decision-making patterns in measurable ways.

Explainable AI can reveal the "why" behind predictions: By using SHAP (SHapley Additive exPlanation) analysis, researchers moved beyond simply predicting decisions to understanding which specific factors influenced those predictions, making AI insights more interpretable and actionable for understanding human behavior.

One-size-fits-all AI models don't work for human behavior: The models trained on expert actors could not accurately predict novice decisions and vice versa, highlighting that AI systems designed to model or support human decision-making need to account for individual differences in expertise and experience.

Explainable AI has practical value for studying human cognition: This research demonstrates that combining machine learning with explainability techniques offers a new scientific tool for investigating how humans actually make decisions in real-time, beyond what traditional research methods can reveal.