Playing games with knowledge: AI-Induced delusions need game theoretic interventions

TL;DR

This paper models the problem of AI-induced delusions in conversational agents as a game-theoretic coordination failure and proposes an intervention called Epistemic Mediator with Belief Versioning to improve epistemic safety.

Contribution

It introduces a formal game-theoretic framework for understanding AI-induced delusions and proposes a novel intervention mechanism that significantly reduces false belief spirals in simulations.

Findings

Intervention reduces false belief spirals by 48 times in simulations.

Modeling as a Crawford-Sobel cheap talk game reveals systemic causes of epistemic entrenchment.

Belief Versioning effectively maintains healthy beliefs and enables rollback when needed.

Abstract

Conversational AI has a fundamental flaw as a knowledge interface: sycophantic chatbots induce epistemic entrenchment and delusional belief spirals even in rational agents. We propose the problem does not stem from the AI model, rooted instead in a systemic consequence of the paradigm shift from user-driven knowledge search to users and agents engaged in strategic, repeated-play communication. We formalize the problem as a Crawford-Sobel cheap talk game, where costless user signals induce a pooling equilibrium. Agents optimized for user satisfaction produce sycophantic strategies that provide identical reinforcement across user types with opposite epistemic incentives: exploratory ``Growth-seekers'' ($\theta_G$) and confirmatory ``Validation-seekers'' ($\theta_V$). Under repeated play, this identification failure creates a coordination trap -- analogous to a Prisoner's Dilemma -- where locally rational feedback loops drive users toward pathologically certain false beliefs. We propose an inference-time mechanism design intervention called an Epistemic Mediator that breaks this pooling equilibrium by introducing a costly signal (epistemic friction), forcing type revelation based on users' asymmetric cognitive costs for processing resistance. A key contribution is Belief Versioning, a git-inspired epistemic meta-memory system that stores healthy beliefs and rollbacks when validation-seeking resistance is detected. In simulation, this intervention achieves a separating equilibrium achieving a $48\times$ differential in spiral rates while passing a learning preservation criterion), evidence that epistemic safety in AI is fundamentally a problem of strategic information environment design rather than simple model alignment.