Modeling how, when, and what is learned in a simple fault-finding task.

Area of Science:

• Cognitive Science
• Computational Modeling
• Human-Computer Interaction

Background:

• Understanding human learning and problem-solving is crucial for designing effective training and support systems.
• Previous models often struggle to capture the nuances of human learning, especially its transfer across different tasks.
• Fault diagnosis in simple systems provides a tractable domain for studying cognitive processes like learning and strategy adaptation.

Purpose of the Study:

• To develop and validate a process model capable of learning and predicting human fault-finding performance.
• To investigate how different types of learning (procedural, declarative, episodic) contribute to problem-solving efficiency.
• To explain the mechanisms behind learning curves and the impact of learning transfer on performance.

Main Methods:

• Developed a novel process model that simulates learning through fault diagnosis in a control panel task.
• Systematically compared the model's learning predictions with empirical data from human participants.
• Analyzed model and human performance metrics including time course, sequence of behaviors, strategy, fault difficulty, and learning transfer.

Main Results:

• The model accurately predicted human learning trajectories, fault-finding strategies, and time-on-task.
• The model demonstrated effective learning transfer across different faults and problem-solving episodes.
• Model performance correlated well with human data, accounting for individual differences and learning curve shapes.

Conclusions:

• The developed process model provides a robust framework for understanding human learning during problem-solving.
• Simulating multiple learning types within a single episode explains recognition-based performance and learning curve dynamics.
• This approach offers insights into cognitive mechanisms underlying skill acquisition and expertise development.