Presentation

· Contributors · Organizations · Search Program

Cognitive Fusion in AI-Assisted Decision Support: How Linguistic Alignment Influences User Autonomy

DescriptionArtificial intelligence (AI) systems increasingly use linguistic mirroring to improve conversational fluency and user engagement. Although prior research has examined automation bias and trust in AI-assisted decision-making, less attention has been given to how repeated linguistic alignment may influence user autonomy over time. Drawing from interactive alignment theory and automation reliance literature, this study presents an agent-based simulation investigating whether empirically established alignment effects are computationally sufficient to produce cumulative autonomy-related behavioral dynamics during repeated AI-assisted decision-support interactions. The model simulated 200 user agents across 30 interaction cycles under varying levels of syntactic alignment, task complexity, and time pressure. Instead of replicating human cognition, agents represented simplified behavioral tendencies derived from prior empirical studies on linguistic alignment, automation reliance, and Need for Cognition. AI agents dynamically mirrored user linguistic structure while occasionally providing degraded recommendations. Results showed that higher linguistic alignment increased acceptance of degraded recommendations, reduced override behavior, and accelerated the accumulation of “cognitive fusion,” defined here as a reduced perceived distinction between self-generated and AI-generated reasoning. These findings suggest the theoretical plausibility of a longitudinal mechanism through which repeated linguistic convergence may influence critical evaluation in human-AI interaction.

Contributors