According to the paper, learning and problem solving depend critically on how internal representations are structured. The authors argue that while modern data-driven AI systems perform well predictively, their learned representations often lack explicit structure, which may limit interpretability and generalization.
To address this, the paper outlines a structural theory of cognitive representation. It ties together problem solving, contexts, and invariance, and introduces the idea of a 'knowledge space' as a central construct. The abstract does not provide details on how the theory is formalized or evaluated.
Because this is a single arXiv abstract, the claims are preliminary and not yet peer-reviewed. No comparison with other theories is available from this source.