Identifying which specific weights and neurons in a model are responsible for particular behaviors or knowledge.