Explicit Control versus Implicit Inference In interactive AI systems

This paper investigates the tension between explicit user customization and implicit behavioral inference in conversational AI systems, framing this dynamic as a fundamental concern for user autonomy and epistemic accuracy. It defines "inferential ossification" as a critical failure mode in which systems conflate transient situational needs with permanent dispositional traits, thereby restricting the user's agency and depth of interaction. To evaluate and mitigate these risks, the study introduces a five-dimensional framework spanning legibility, correctability, ownership, paternalism, and homogenization and proposes hybrid design patterns such as editable inferences and mandatory explicit overrides. Finally, it formulates actionable policy and design recommendations for regulators and user advocates to ensure adaptive AI architectures balance automated convenience with human agency.