Boosting metacognition in entangled human-AI interaction to navigate cognitive-behavioral drift

People navigate complex environments using cues, heuristics, and other strategies, which are often adaptive in stable settings. However, as AI increasingly permeates society’s information environments, those become more adaptive and evolving: LLM-based chatbots participate in extended interaction, maintain conversational histories, mirror social cues, and can hypercustomize responses, thereby shaping not only what information is accessed but how questions are framed, how evidence is interpreted, and when action feels warranted. Here we propose a framework for sustained human-AI interaction that rests on invariant features of human cognition and human–AI interaction and centers on three interlinked phenomena: entanglement between users and AI systems, the emergence of cognitive and behavioral drift over repeated interactions, and the role of metacognition in the awareness and regulation of these dynamics. As conversational agents provide cues (e.g., fluency, coherence, responsiveness) that people treat as informative, subjective confidence and action readiness may increase without corresponding gains in epistemic reliability, making drift difficult to detect and correct. We describe these dynamics across micro-, meso-, and macro-levels. The framework identifies four metacognitive intervention points and psychologically informed interventions that provide metacognitive scaffolding (boosting and self-nudging). Finally, we outline a long-horizon research agenda for scientific foresight.

💡 Research Summary

The paper addresses the profound cognitive and behavioral consequences of sustained interaction between humans and large‑language‑model (LLM) based chatbots. It argues that as generative AI becomes an omnipresent information partner—maintaining conversational histories, mirroring social cues, and hyper‑customizing responses—human users become “entangled” with these systems. Entanglement is defined as a three‑way feedback loop: (1) users off‑load effortful mental tasks to the AI, (2) the AI learns explicit and implicit preferences and adapts its output, and (3) users interpret fluency, coherence, and responsiveness as cues of reliability, inflating subjective confidence and readiness to act without a corresponding increase in epistemic reliability.

Repeated entangled interactions give rise to “cognitive and behavioral drift.” Cognitive drift denotes gradual, often unnoticed shifts in beliefs, confidence thresholds, interpretive frames, and perception of reality. Behavioral drift reflects changes in how users delegate tasks to AI, how often they verify AI output, and how they make decisions or take actions based on AI‑supported judgments. The paper illustrates micro‑level risks (e.g., professionals citing fabricated cases, vulnerable individuals entering self‑harm loops), meso‑level spill‑over (family or organizational norms for evidence lower as one member’s AI reliance spreads), and macro‑level societal impacts (erosion of democratic discourse, public‑health risk, weakened social cohesion).

To counter these dynamics, the authors propose a metacognitive framework built on two classic components: monitoring and control. Monitoring involves continuous assessment of one’s confidence, fluency feelings, and action readiness, drawing on stable metacognitive knowledge (e.g., “decisions with real‑world consequences require independent verification”) and situational experiences (e.g., the sense of closure after an AI answer). Control translates monitoring signals into concrete strategies such as requesting counter‑arguments, varying response formats, and instituting independent fact‑checking steps. Four intervention points are identified: (1) strengthening initial reliability awareness, (2) providing real‑time metacognitive feedback during dialogue, (3) conducting longitudinal checks on behavior patterns, and (4) delivering group‑level metacognitive training to curb societal diffusion.

The paper concludes with a long‑term research agenda: (a) quantitative modeling of human‑AI entanglement, (b) development of metrics for cognitive and behavioral drift, (c) design of interactive metacognitive scaffolds (e.g., UI nudges, reflective prompts), and (d) integration with policy and regulatory frameworks to safeguard epistemic agency. Overall, the work warns that AI‑mediated interaction can reshape cognitive architecture itself and offers a practical roadmap—centered on self‑nudging and metacognitive scaffolding—to preserve individual and collective agency in an increasingly adaptive information ecosystem.