Similarity as an explanatory construct

Theories can be found throughout cognitive science that give an explanatory role to similarity. Such theories can be contrasted with those that model thought using abstract rules. We lay out four possible explanatory roles for similarity. We then review the computational pros and cons of similarity- and rule-based models and outline the empirical work that speaks to the psychological plausibility of the two frameworks. We conclude that an adequate model of human thought must take advantage of both the flexibility of similarity-based inference and the compositionality and certainty associated with rule-based inference.