Introspective Capabilities in Large Language Models
Abstract
This paper considers the kind of introspection that large language models (LLMs) might be able to have. It argues that LLMs, while currently limited in their introspective capabilities, are not inherently unable to have such capabilities: they already model the world, including mental concepts, and already have some introspection-like capabilities. With deliberate training, LLMs may develop introspective capabilities. The paper proposes a method for such training for introspection, situates possible LLM introspection in the 'possible forms of introspection' framework proposed by Kammerer and Frankish, and considers the ethical ramifications of introspection and self-report in AI systems.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Cites
Word meaning in minds and machines.
Lake, B. M., & Murphy, G. L. (2023) · Psychological Review · doi:10.1037/rev0000297 · Cited by 131 · see citing works
Flagged first, then by citations
Research works printed but not linked: 5
- Neuroscience of Consciousness, 2, 2
- Philosophical Transactions of the Royal Society of London A: Physical and Engineering Sciences, 349, 133
- PROCEEDINGS OF THE THEMATIC CONFERENCE ON REMOTE SENSING FOR MARINE AND COASTAL ENVIRONMENTS, 2, 1
- no to shinkei. brain and nerve., 56, 133
- Journal of Consciousness Studies, 28, 121
Cited by
Sorted by citations
Artificial intelligence, human cognition, and conscious supremacy
Mogi, K. (2024) · Frontiers in Psychology · doi:10.3389/fpsyg.2024.1364714 · Cited by 31 · see citing works
“Despite some claims that LLMs are either already or soon becoming conscious (Long, 2023), many regard these generative AI systems as doing computation unconsciously, thus forgoing possible ethical issues involved in AI abuse (Blauth et al., 2022).”
BackgroundQuantifying uncert-AI-nty: Testing the accuracy of LLMs’ confidence judgments
Cash, T. N., Oppenheimer, D. M., Christie, S., & Devgan, M. (2025) · Memory & Cognition · doi:10.3758/s13421-025-01755-4 · Cited by 14 · see citing works
“…of the degree to which generative AI chatbots utilize metacognitive cues, it is theoretically plausible that they could generate meaningful metacognitive judgments through some combination of intrinsic and extrinsic cues, despite their lack of introspective capacities (Li et al., 2024; Long, 2023).”
Individuating artificial moral patients
Register, C. (2025) · Philosophical Studies · doi:10.1007/s11098-025-02409-6 · Cited by 7 · see citing works
“…current LLMs is not widely shared, many experts including neuroscientists, computer scientists, and philoso-Received phers have argued that there is a non-negligible chance that near-future AI systems will be conscious and even sentient (Butlin et al., 2023 , Chalmers, 2023 , Sebo & Long, 2023 ).”
Introspective Machines: Are LLMs Better at Self‐Reflection Than Humans?
Cappelen, H., & Dever, J. (2024) · Philosophical Perspectives · doi:10.1111/phpe.12201 · Cited by 5 · see citing works
Does It Make Sense to Speak of Introspection in Large Language Models?
Comsa, I. M., & Shanahan, M. (2025) · arXiv · doi:10.48550/arxiv.2506.05068 · Cited by 3 · see citing works
Is learning with ChatGPT really learning?
Stolz, S. A., Winterburn, A. L., & Palmer, E. (2024) · Educational Philosophy and Theory · doi:10.1080/00131857.2024.2376641 · Cited by 4 · see citing works
“When asked about their internal states, the answers provided by LLMs tend to be unreliable, reflecting their training data set and fine-tuning (Long, 2023, pp. 145-146). despite this, Long believes that there is evidence that LLMs construct world models (Long, 2023, p. 147).”
InfluentialKammerer, F. (2026) · doi:10.1093/9780197840283.001.0001 · Cited by 2 · see citing works
Quantifying UncertAInty: Testing the Accuracy of LLMs’ Confidence JudgmentsPreprint
Cash, T. N., Oppenheimer, D. M., & Christie, S. (2024) · doi:10.31234/osf.io/47df5 · Cited by 1 · see citing works
Quantifying Uncert-AI-nty: Testing the Accuracy of LLMs’ Confidence JudgmentsPreprint
Cash, T. N., Oppenheimer, D. M., Christie, S., & Devgan, M. (2025) · doi:10.31234/osf.io/47df5_v3 · Cited by 1 · see citing works
Quantifying Uncert-AI-nty: Testing the Accuracy of LLMs’ Confidence JudgmentsPreprint
Cash, T. N., Oppenheimer, D. M., Christie, S., & Devgan, M. (2025) · doi:10.31234/osf.io/47df5_v2
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0002
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0010
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0001
Approaching the House of Mirrors
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0005
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0006
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0004
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0009
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.002.0006
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0008
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0007
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0012
Illusionism About Phenomenal Consciousness
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0003
Kammerer, F. (2026) · Oxford University Press eBooks · doi:10.1093/9780197840283.003.0011
- Authors
- Long, R.
- Affiliations
- Center for AI Safety, San Francisco, CA, USA
- Corresponding author
- Long, Robert
- Venue
- Journal of Consciousness Studies
- Publisher
- Imprint Academic Ltd
- Published
- 2023-09-30
- Volume
- 30
- Issue
- 9
- Pages
- 143-153
- ISSN
- 1355-8250 (ISSN-L)
- OpenAlex
- W4387404506
- Semantic Scholar
- 13871f3f17d9d136bde7da6bb3e497d78636a32b
- Cited by
- 22
- References
- 1
- Citations by year
- 2026: 14 · 2025: 4 · 2024: 4
- FWCI
- 5.14
- Type
- journal article
- Language
- en
- Open access
- closed
- Locations
- Journal of Consciousness Studies (publishedVersion)
- Topics
- Social Sciences › Psychology › Experimental and Cognitive Psychology › Philosophy and Theoretical Science
- Social Sciences › Arts and Humanities › History and Philosophy of Science › Cybernetics and Technology in Society
- Physical Sciences › Computer Science › Artificial Intelligence › Psychiatry, Mental Health, Neuroscience
- Keywords
- self-reflection, large language models, self-report, AI ethics, psychological concepts, artificial consciousness
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
