Mental Models and the Illusions of Understanding

Shiffrin et al. (2025) argue that scientific understanding is almost always illusory to varying degrees, a claim supported by the depth of apparently simple modelling tools like linear regression. This commentary maps the illusions identified by Shiffrin et al. (SSK) to the CoreSense Theory of Understanding (ToU). By defining understanding as the possession of actionable models allowing sound inferences, the ToU provides a formal diagnosis for these illusions. We categorize them into four model-theoretic failure modes and argue that while partial understanding is structurally unavoidable, the explicit formalization of models can mitigate these epistemic risks in both human science and autonomous robotics.