Privileged Self-Access Matters for Introspection in AI

Siyuan Song, Harvey Lederman, Jennifer Hu, Kyle Mahowald
8/20/2025
cs.AIcs.CL

No official code found. Best implementation: tsuji-tomonori/llm-temp-introspection (0 stars, last commit 2026-02-27).

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.

View on arXivView PDF

Code Implementations(1)

tsuji-tomonori/llm-temp-introspectionCommunity59%

論文「Privileged Self-Access Matters for Introspection in AI」の日本語追実験プロジェクト

00Shell, PythonOct 5, 20257 months ago

Similar Papers

Does It Make Sense to Speak of Introspection in Large Language Models?

85% match2025

Looking Inward: Language Models Can Learn About Themselves by Introspection

85% match1 repo162024

Dissociating Direct Access from Inference in AI Introspection

84% match2026

Position: Introspective Experience from Conversational Environments as a Path to Better Learning

81% match2026

Emergent Introspective Awareness in Large Language Models

80% match1 repo82026

Latent Introspection: Models Can Detect Prior Concept Injections

80% match1 repo32026

Cite this paper

@article{song2025privileged,
  title  = {Privileged Self-Access Matters for Introspection in AI},
  author = {Siyuan Song and Harvey Lederman and Jennifer Hu and Kyle Mahowald},
  year   = {2025},
  eprint = {2508.14802},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2508.14802v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2508.14802.svg)](https://codeofpaper.com/paper/2508.14802)

Discussion