Rational AI Dependence Through Capability Augmentation: A Reference-Dependent Model Without Requiring Skill Degradation

Generative AI can augment human productivity and cognitive capability. Existing discussions of AI dependence have addressed cognitive offloading, overreliance, skill degradation, delegation of cognitive agency, and the social diffusion of AI use. This paper proposes a complementary pathway: rational AI dependence through capability augmentation. Let unaided human capability be H and additional capability provided by AI be A. Effective capability during AI use is therefore H + A. If repeated AI use causes this augmented state to become established as the normal capability level for the user or the occupational environment, losing access to AI may be perceived not merely as a return to the original baseline but as the loss of capability A. In occupational settings, workload, deadlines, assigned scope, and evaluation criteria may also adapt to AI-assisted productivity, producing not only a subjective sense of loss but an actual shortfall in task performance. This paper presents a minimal model incorporating reference-point adaptation, loss aversion, and adaptation of organizational performance requirements. The increase in the utility advantage of continued AI use produced by post-adoption adaptation is defined as the Augmentation Lock-in Premium. Rationality here refers to the rationality of choice given the reference point and environmental conditions at each point in time. The model does not require degradation of unaided human skill: AI dependence may arise not only because humans become weaker, but also because AI makes them stronger. Generative AI was used as an assistive tool in the preparation of this paper. Its primary uses included developing ideas, organizing the theoretical structure, examining the mathematical model and propositions, searching for and organizing related literature, drafting text, examining falsification conditions and experimental designs, refining expression, and restructuring prose. The selection of the topic, the central idea of “rational AI dependence through capability augmentation,” decisions regarding the adoption or rejection of theoretical elements, the final structure, verification of content, revisions, and the decision to publish were made by the author, kurato. AI-generated outputs were incorporated only after selection, editing, restructuring, and verification by the author. Final responsibility for the content and publication of this paper rests with the author.