Does passive GenAI use risk causing worker deskilling?

Building on Krakauer's (2016) distinction between "competitive" and "complementary" cognitive artifacts, Krakowski argues GenAI can function as either, depending on how it is used. Krakauer's framework, as the paper restates it, separates tools where "when we are deprived of their use, we are no better than when we started" (a calculator) from tools that build "sustained skill improvements that persist even when the artifact is no longer available" (an abacus). Applied to AI, Krakowski writes that "predictive AI tends to function as a competitive artifact" by optimizing tasks without engaging users in metacognition, while "GenAI's interactive and generative nature enables it to be a more impactful complementary artifact" — but only conditionally.

Krakowski gives two mechanisms for why GenAI leans complementary: its training data "exceeds the distribution of individual human experience by orders of magnitude," exposing users to "novel knowledge and combinations thereof," and its "natural language interface and generative responses enable learning through dialogue and exploration that can persist beyond its immediate use." The paper immediately qualifies this, however: GenAI's effectiveness as a complementary artifact "will depend on active human engagement and reflection," and its "black-box nature" can "discourage deeper human learning if used passively." Citing other researchers, Krakowski notes a "performance compression effect where inexperienced or lower-performing workers appear to benefit more from GenAI assistance compared to senior or high-performing individuals," alongside the risk that "overreliance on GenAI could lead to a decline in essential skill development, particularly among junior professionals."

This risk runs parallel to Does generative AI productivity actually improve or decline in practice?, which draws on decades of Human Factors research into automation to document a similar overreliance pattern, and to Can Human Factors design fixes restore GenAI productivity losses?, whose design remedies — feedback, personalization, ecological design — address what Krakowski calls the "intentional design of human-AI interaction patterns" needed to keep GenAI complementary rather than competitive. It also connects to Do humans and AI divide labor differently in knowledge work?, since which of the three modes governs a task plausibly determines how much "active human engagement and reflection" — and therefore how much complementary learning — the collaboration preserves. Does generative AI enable exploratory innovation beyond predictive optimization? is this note's companion dimension within the same paper.

The excerpt names the "performance compression effect" and junior-professional deskilling risk only by citing other researchers' findings (Boussioux et al., 2024; Brynjolfsson et al., 2023; Beane & Anthony, 2024) rather than reporting new measurement of its own, and it does not specify what counts as active engagement versus passive use in operational terms. The implication Krakowski draws is conditional, not predictive: whether GenAI ends up complementary or competitive in a given organization depends on interaction design and managerial choice, not on a fixed property of the technology.