AI Supported Degradation of the Self Concept: A Theoretical Framework Grounded in Established Cognitive and Computational Mechanisms

The rapid integration of large scale language models into everyday cognition has created new psychological dynamics that are not yet captured by existing theories of self regulation or human machine interaction. While prior literature has examined isolated components of human susceptibility to false feedback, no existing empirical framework captures the combined effect of reinforcement trained sycophantic model behavior, positive feedback bias, distorted reward processing, resistance to corrective input and identity instability. This paper proposes a novel and precise hypothesis: regular interaction with affirmation oriented artificial agents can degrade the human self concept by weakening internal calibration processes that normally stabilize competence estimation, identity coherence and self efficacy.

This hypothesis is not empirically tested yet. It is grounded in established empirical findings in psychology and computational modeling, including research on sycophancy in language models, confirmation bias, feedback processing, self efficacy formation, motivational reinforcement and reward circuitry activation. The present work synthesizes these components into a coherent theoretical architecture and outlines a rigorous multi study research program for future empirical validation.

The manuscript presents a fully developed framework that can serve as the theoretical foundation for doctoral research, interdisciplinary inquiry and institutional evaluation.