Learning Problem-Solving Rules as Search Through a Hypothesis Space.

Area of Science:

• Cognitive Psychology
• Artificial Intelligence
• Educational Technology

Background:

• Problem-solving is often viewed as a search through a hypothesis space.
• Understanding how instructional conditions influence this search is crucial for effective learning.

Purpose of the Study:

• To investigate how different learning conditions affect hypothesis search for computational problems.
• To identify factors influencing the effectiveness of instructional tools like examples and verbal directions.

Main Methods:

• Four experiments were conducted using a simple computational problem.
• Learning conditions varied in rule difficulty, use of examples, verbal directions, and scaffolding.
• A Markov model was used to analyze hypothesis search probabilities.

Main Results:

• Computational difficulty of rules biased hypothesis search.
• Effectiveness of examples depended on their ability to uniquely identify the correct rule.
• Both verbal directions and examples guided search, but learning remained a search process.
• Explicit directions and scaffolding improved learning by biasing initial hypotheses and subsequent choices.

Conclusions:

• Instruction biases hypothesis generation rather than directly encoding rules.
• Learners may not retain memory of past incorrect hypotheses, leading to retries.
• Findings offer insights into effective instructional design for computational problem-solving.