Ross (1984)
Found that people often retrieve analogs based on surface features rather than deep structure leading to incorrect problem-solving strategies.
Chi Feltovich Gick and Holyoak (1983)
Showed that experts categorize problems by underlying principles, while novices rely on surface features like context or objects.
Cummins (1992)
Found that schema-based questions help novices focus on structural features improving their ability to transfer solutions across problems.
Large Language Model Research (Webb Holyoak)
Showed that GPT-3 and GPT-4 display strong analogical reasoning, sometimes matching or surpassing human performance on abstract tasks.