Teaching Recombinable Motifs Through Simple Examples

Area of Science:

• Cognitive Science
• Psychology
• Artificial Intelligence

Background:

• Effective teaching imparts abstract knowledge, not just facts.
• Understanding how humans teach and learn abstractions is crucial for educational development.

Purpose of the Study:

• To investigate how human teachers convey abstract knowledge using examples.
• To model pedagogical strategies using Bayesian approaches in a necklace-building task.

Main Methods:

• Applied Bayesian models of pedagogy to a necklace-building task.
• Conducted two experiments (N=151 and N=295) with human teachers and learners.
• Analyzed necklace simplicity using algorithmic complexity and a pedagogical sampling model.

Main Results:

• Human teachers create simpler necklaces than random chance, prioritizing examples with fewer interpretations.
• Simpler examples lead to better motif recovery by human learners, aligning with the pedagogical sampling model.
• Human learners performed better with human teachers than AI-generated examples, indicating unmodeled human expectations.

Conclusions:

• Simplicity in examples is a key strategy for teaching abstract concepts effectively.
• Bayesian pedagogical models can explain teachers' preference for simpler examples.
• Human teachers possess nuanced expectations about learners that current AI models do not fully capture.