MIRROR: Cognitive Inner Monologue Between Conversational Turns for Persistent Reflection and Reasoning in Conversational LLMs

2025-05-31

AI Summary

This work addresses three critical failure modes of large language models (LLMs) in dialogue: acquiescence bias, neglect of salient information, and inconsistent constraint prioritization. To mitigate these issues, we propose MIRROR, a cognitive architecture comprising two synergistic modules—Thinker (featuring an introspective inner-monologue manager and a cognitive controller) and Talker—that jointly enable cross-turn reflective reasoning, memory integration, and goal-directed inference. MIRROR introduces a human-inspired parallel introspection mechanism, orchestrating modular internal reasoning, persistent cross-turn cognitive state maintenance, and tri-dimensional coordination among goals, reasoning, and memory to drive cognition-informed response generation. Evaluated on the CuRaTe safety benchmark, MIRROR achieves a 156% relative performance improvement over baselines; when integrated across multiple models, it attains a stable overall accuracy exceeding 80%, representing an average gain of 21 percentage points.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Natural Language Processing: Safety and Robustness

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Human intelligence relies on inner monologue to process complex information through simultaneous reflection, memory retrieval, and response formulation. We introduce MIRROR (Modular Internal Reasoning, Reflection, Orchestration, and Response), a cognitive architecture that systematically implements these parallel reasoning capabilities in large language models. MIRROR operates as a unified system with two distinct functional layers: the Thinker and the Talker. The Thinker encompasses: (1) the Inner Monologue Manager, coordinating reasoning threads across cognitive dimensions (Goals, Reasoning, and Memory); and (2) the Cognitive Controller, synthesizing these threads into a coherent internal narrative maintained across conversation turns. The Talker component then leverages this integrated narrative for context-aware responses. Evaluated on the CuRaTe benchmark--testing personalized dialogue with safety-critical constraints, conflicting preferences, and multi-turn consistency--LLMs utilizing the MIRROR architecture achieve up to 156% relative improvement in critical safety scenarios involving three persons with conflicting preferences, maintaining an average accuracy of ~>80% on all scenarios. Across scenario-specific comparisons, GPT-4o, Gemini 1.5 Pro, Claude 3.7 Sonnet, Llama 4 variants, and Mistral 3 variants with the MIRROR architecture outperformed baseline models by 21% on average (15.5 percentage points absolute). MIRROR directly addresses three critical LLM failure modes: sycophancy, attentional deficits to critical information, and inconsistent prioritization of conflicting constraints. This work bridges cognitive science and AI by implementing modular internal reasoning inspired by human cognition, creating a persistent internal model that significantly enhances multi-turn conversation capabilities.

Problem

• Enhancing conversational LLMs with human-like inner monologue for reflection and reasoning
• Addressing LLM failure modes: sycophancy, attentional deficits, and inconsistent constraint prioritization
• Improving multi-turn dialogue consistency and safety in complex personalized scenarios

Innovation

• Modular cognitive architecture for parallel reasoning
• Thinker-Talker layers for internal narrative synthesis
• Persistent reflection across conversational turns