Privileged Self-Access Matters for Introspection in AI
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866911112380809216 |
|---|---|
| author | Song, Siyuan Lederman, Harvey Hu, Jennifer Mahowald, Kyle |
| author_facet | Song, Siyuan Lederman, Harvey Hu, Jennifer Mahowald, Kyle |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2508_14802 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Privileged Self-Access Matters for Introspection in AI Song, Siyuan Lederman, Harvey Hu, Jennifer Mahowald, Kyle Artificial Intelligence Computation and Language 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. |
| title | Privileged Self-Access Matters for Introspection in AI |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2508.14802 |