Talk at CogSci 2026

Abstract

Educators often build up complex concepts by teaching simplified versions that are not quite accurate, such as Bohr’s model of the atom or Newtonian mechanics. ‘‘Lying to children’’, while ubiquitous in STEM teaching, poses a challenge to existing cognitive models of pedagogy, which assume that teachers select evidence that truthfully represents a target concept. Why would helpful, knowledgeable teachers lie? We present a theoretical framework that addresses this puzzle by reinterpreting optimal pedagogy through the lens of bounded rationality. When learners face cognitive constraints on belief updating, our model predicts that teachers should prioritize examples that will bring the learner closest to the target concept—even if they do not represent the target concept truthfully; by contrast, classic pedagogy models fail to make this prediction. Our work formalizes an insight that educators have long understood – pedagogical ``lies’’ are not meant to mislead learners, but to meet them where they are.