Prediction, Explanation, and Control: The Use of Mental Models in Dynamic Environments.

Area of Science:

• Cognitive Science
• Human-Computer Interaction
• Artificial Intelligence

Background:

• Human cognitive abilities for prediction, explanation, and control can arise from unified representations or distinct processes.
• Investigating the role of probabilistic mental models in these tasks is crucial for understanding cognitive flexibility.

Purpose of the Study:

• To explore how individuals utilize probabilistic mental models for prediction, explanation, and control under varying complexity and uncertainty.
• To examine task performance, correlations between abilities, and skill transfer across tasks.

Main Methods:

• Developed a novel experimental paradigm using a chatbot defined by a finite-state machine.
• Assessed participant performance in prediction, explanation, and control tasks with full and hidden information.
• Conducted cross-training experiments to evaluate skill transfer and correlations between abilities.

Main Results:

• Control tasks were most robust to complexity, while explanation was most challenging with full information.
• Hidden information led to performance equalization and an alternative neglect bias.
• Cross-training revealed asymmetries: control training generalized to prediction and explanation, but not conversely.

Conclusions:

• Findings suggest a complex interplay between mental models and task-specific heuristics, especially with partial information.
• Control-based training fosters generalizable cognitive abilities, indicating a potential unifying role for mental models.
• Highlights new research directions in cognitive abilities and general-purpose mental model acquisition.