Borrowed Certainty AI, Self-Assessment and the Masking of Competence Gaps

Generative artificial intelligence allows people to produce answers, analyses and texts that exceed what they could produce unaided. This paper asks what such assisted success does to a person’s sense of their own competence. It develops the concept of borrowed certainty: a first-person state in which an achievement depends materially on an external system, while the achievement, or the competence it displays, is represented to oneself as grounded in one’s own understanding or ability, and that self-representation lacks an adequate basis. The account is deliberately narrower than the claim that externally supported cognition is not knowledge. It accepts the lessons of the extended-mind thesis, distributed cognition and the epistemology of testimony, and relocates the failure from dependence to attribution: borrowed certainty is a misallocation of epistemic credit within a hybrid achievement. Three distinctions give the concept its shape. First, between knowledge and understanding: AI output can transmit knowledge that p while leaving the user without understanding why p or the ability to do again what was done, and borrowed certainty typically mistakes the former for the latter. Second, between dependence, deference and appropriation, which are causal, doxastic and attributional relations respectively and can come apart. Third, between momentary distrust, abandonment and repair as responses to AI error, a distinction that matters because conversational systems invite correction rather than withdrawal. The paper argues that the automatic endorsement which extended-mind theory treats as harmless becomes problematic when fluent output suppresses the signals by which integrated cognitive systems ordinarily register their own failures. It situates the account within recent evidence on overreliance, under-reliance, self-assessment under AI assistance and the social penalty for disclosure, states a set of conditional theoretical propositions with the evidence that would count against them, and closes with an agenda for future conceptual and empirical research.