Category-based induction from similarity of neural activation

Area of Science:

• Cognitive Science
• Neuroscience
• Philosophy of Mind

Background:

• The relationship between similarity and inductive inference has been debated since David Hume.
• Testing this relationship is challenging due to potential circularity in similarity judgments.

Purpose of the Study:

• To develop and test a model of category-based induction.
• To investigate whether similarity, measured independently, predicts inductive inference.

Main Methods:

• A one-parameter model was created to predict conditional probabilities.
• Brain activation patterns in the ventral visual stream were recorded during a categorization task.
• Representational similarity analysis was used to quantify pattern similarity.
• Model predictions were compared against human judgments of conditional probability.

Main Results:

• The model's predictions, based on neural similarity, correlated well with subjects' judgments.
• This suggests that neural similarity can predict inductive reasoning.

Conclusions:

• Similarity, when assessed without direct reference to the inductive task, appears to drive inductive inference.
• These findings provide empirical support for Hume's thesis on similarity and induction.