Privileged Self-Access Matters for Introspection in AI

2025/08/20 by Siyuan Song, Harvey Lederman, Song, Siyuan +5 · 2 voices · 11 citations
Computer Science · Social Sciences · #Consciousness #Ethics and Social Impacts of AI #Explainable Artificial Intelligence (XAI) #Introspection #Online Learning and Analytics #Process (computing)

paper · pdf · doi:10.48550/arxiv.2508.14802

published in arXiv (Cornell University) (Cornell University)

openalex publication_date 2025/08/20 · openalex created_date 2025/10/09 · openalex updated_date 2026/07/28

Abstract

Whether AI models can introspect is an increasingly important practical question. But there is no consensus on how introspection is to be defined. Beginning from a recently proposed ''lightweight'' definition, we argue instead for a thicker one. According to our proposal, introspection in AI is any process which yields information about internal states through a process more reliable than one with equal or lower computational cost available to a third party. Using experiments where LLMs reason about their internal temperature parameters, we show they can appear to have lightweight introspection while failing to meaningfully introspect per our proposed definition.

Citations

• Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
• Does It Make Sense to Speak of Introspection in Large Language Models?
• Tell me about yourself: LLMs are aware of their learned behaviors
• Looking Inward: Language Models Can Learn About Themselves by Introspection

Cited by

• Hallucinations Undermine Trust; Metacognition is a Way Forward
• Emergent Introspection in AI is Content-Agnostic
• Latent Introspection: Models Can Detect Prior Concept Injections
• Metacognitive Sensitivity for Test-Time Dynamic Model Selection
• Steering Awareness: Detecting Activation Steering from Within
• Evidence for Limited Metacognition in LLMs
• Asymmetric Communication: Large Language Models and Language Games
• Mechanistic Interpretability Needs Philosophy
• Training Language Models to Explain Their Own Computations
• Metacognition in LLMs: Foundations, Progress, and Opportunities
• Introspective Coupling: Self-Explanation Training Tracks Behavioral Change Despite Fixed Supervision
• Can LLMs Introspect? A Reality Check

Discussions

• How reliable is what an AI says about itself? The answer depends on whether models can introspect. But, if an LLM says its temperature parameter is high (and it is!)….does that mean it’s introspecting [bsky, 16 points, 1 comments]
• ..reliably report their inner workings. Such progress could greatly improve AI transparency, safety, and alignment with human values. No GitHub link available. Read more: https://arxiv.org/abs/2508.14 [bsky, 0 points, 0 comments]

Related

• Emergent Introspection in AI is Content-Agnostic
• Me, Myself, and π : Evaluating and Explaining LLM Introspection
• Looking Inward: Language Models Can Learn About Themselves by Introspection
• Does It Make Sense to Speak of Introspection in Large Language Models?
• Introspection Fine-Tuning (IFT): Training Small LLMs to Introspect