Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 81 of the 81 references it cites were found in a registry or an index; 30 of the 53 sources its claims cite were checked for support.

Deanthropomorphising NLP: Can a language model be conscious?

Open access
Shardlow, Matthew, Przybyła, Piotr · 2024 · PLoS ONE

Abstract

This work is intended as a voice in the discussion over previous claims that a pretrained large language model (LLM) based on the Transformer model architecture can be sentient. Such claims have been made concerning the LaMDA model and also concerning the current wave of LLM-powered chatbots, such as ChatGPT. This claim, if confirmed, would have serious ramifications in the Natural Language Processing (NLP) community due to wide-spread use of similar models. However, here we take the position that such a large language model cannot be conscious, and that LaMDA in particular exhibits no advances over other similar models that would qualify it. We justify this by analysing the Transformer architecture through Integrated Information Theory of consciousness. We see the claims of sentience as part of a wider tendency to use anthropomorphic language in NLP reporting. Regardless of the veracity of the claims, we consider this an opportune moment to take stock of progress in language modelling and consider the ethical implications of the task. In order to make this work helpful for readers outside the NLP community, we also present the necessary background in language modelling.

Overview

CoChat Summary

TL;DR

The paper argues that Transformer-based LLMs such as LaMDA cannot be conscious under Integrated Information Theory, and that claims of sentience arise from anthropomorphic language in NLP and media reporting. It also documents anthropomorphic terminology in ACL abstracts and offers recommendations for developers, users, and reporters.

Key findings

  • Applying IIT, LaMDA and other Transformer-based LLMs cannot possess consciousness for three reasons: they execute as a sequence of Transformer blocks with extremely limited inter-block information exchange; the blocks are feed-forward networks with no recurrent connections; and they are implemented on modular, low-connectivity computer hardware.
  • In LaMDA, the link between modules corresponds to a single word represented by 8192 numbers, while within-module links are at least the number of non-embedding weights, 137 billion in the most advanced LaMDA; within-module links outnumber between-module links by at least 8 orders of magnitude.
  • LaMDA has 64 layers, 128 attention heads, 137 billion parameters, and was trained on 1.56 trillion words; the paper argues there is no breakthrough innovation in such models that would give rise to consciousness, and that apparent consciousness is a simulated language style produced by prompting.
  • In an analysis of the first 100 abstracts from the 2022 ACL Annual Meeting, 90 referred to language modelling technology and 36 used anthropomorphic terminology; media reporting of LaMDA was found to use explicit anthropomorphism.

Methods

  • Conceptual and argumentative analysis: the paper reviews language modelling, Transformer-based LLMs, and LaMDA, then assesses them against Integrated Information Theory as presented by Tononi and Koch, including recent LLM developments such as larger models, larger context windows, non-Transformer architectures, and retrieval-augmented generation.
  • Qualitative audit: the authors analysed the first 100 abstracts in order of appearance from the proceedings of the 2022 ACL Annual Meeting, identified anthropomorphic terminology, and proposed a three-level framework—non-anthropomorphic, ambiguous anthropomorphism, and explicit anthropomorphism—which they applied to the LaMDA paper and selected media reporting.
  • No empirical consciousness test or direct experimental manipulation of LaMDA is reported; the study is an analysis of architecture, theory, and language use.

Limitations

  • Stated: the identification of anthropomorphic language relies on the subjective application of the framework by one author, and others may reach different conclusions; the authors acknowledge their inherent bias, namely non-belief in the model’s consciousness, and cannot claim to have fully overcome it.
  • Stated: the study presupposes that distinguishing consciousness from non-consciousness in LLMs is valuable; it uses IIT because it appears most applicable to neural networks, but other approaches are possible; conclusions are limited to English-language models and examples, and generalisability is argued for Transformer-based models rather than all possible architectures.
  • Evident: the media analysis is limited to selected sources around LaMDA, and the anthropomorphism audit is not presented with inter-rater reliability or a full quantitative coding protocol; the paper does not directly test LaMDA or other LLMs for consciousness.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1371/journal.pone.0307521

Cites

81 of 81 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Fig 1 Data flow in text generation using transformer-based left-to-right LLM, such as LaMDA or GPT.
Fig 1 Data flow in text generation using transformer-based left-to-right LLM, such as LaMDA or GPT.CC BY · PubMed Central

Verification

Claims the paper makes

22 Valid · 13 Not valid · 12 Not checked · 68 Own results and interpretations

50 more claims · 9 with their cited sources checked

Create a free account to see all 115 claims and to verify the 3 references still open.

See all claims — free account
  • Probably not a substantial claim. 52
  • It may state two separate claims as one. 21

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 80 references

Research works printed but not linked: 13

  1. Brown T, Mann B, Ryder N, Subbiah M, Kaplan JD, Dhariwal P, et al. Language Models are Few-Shot Learners. In: Larochelle H, Ranzato M, Hadsell R, Balcan MF, Lin H, editors. Advances in Neural Information Processing Systems. vol. 33. Curran Associates, Inc.; 2020. p. 1877–1901. Available from: https://proceedings.neurips.cc/paper/2020/file/1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf.
  2. Thoppilan R, De Freitas D, Hall J, Shazeer N, Kulshreshtha A, Cheng HT, et al. Lamda: Language models for dialog applications. arXiv preprint arXiv:220108239. 2022;.
  3. J Austen, 1813, Pride and Prejudice
  4. Mikolov T, Sutskever I, Chen K, Corrado GS, Dean J. Distributed Representations of Words and Phrases and their Compositionality. In: Burges CJ, Bottou L, Welling M, Ghahramani Z, Weinberger KQ, editors. Advances in Neural Information Processing Systems. vol. 26. Curran Associates, Inc.; 2013.Available from: https://proceedings.neurips.cc/paper/2013/file/9aa42b31882ec039965f3c4923ce901b-Paper.pdf.
  5. Yang Z, Dai Z, Yang Y, Carbonell J, Salakhutdinov R, Le QV. XLNet: Generalized Autoregressive Pretraining for Language Understanding. arXiv: 190608237. 2019;.
  6. Clark K, Luong T, Le QV, Manning C. ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators. In: Proceddings of ICLR; 2020.Available from: https://openreview.net/pdf?id=r1xMH1BtvB.
  7. Zhu F, Lei W, Wang C, Zheng J, Poria S, Chua TS. Retrieving and reading: A comprehensive survey on open-domain question answering. arXiv preprint arXiv:210100774. 2021;.
  8. Chowdhery A, Narang S, Devlin J, Bosma M, Mishra G, Roberts A, et al. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:220402311. 2022;.
  9. Zhang S, Roller S, Goyal N, Artetxe M, Chen M, Chen S, et al. Opt: Open pre-trained transformer language models. arXiv preprint arXiv:220501068. 2022;.
  10. Beltagy I, Peters ME, Cohan A. Longformer: The long-document transformer. arXiv preprint arXiv:200405150. 2020;.
  11. Khandelwal U, Levy O, Jurafsky D, Zettlemoyer L, Lewis M. Generalization through Memorization: Nearest Neighbor Language Models. In: 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020. OpenReview.net; 2020.Available from: https://openreview.net/forum?id=HklBjCEKvH.
  12. Fu DY, Dao T, Saab KK, Thomas AW, Rudra A, Re C. Hungry Hungry Hippos: Towards Language Modeling with State Space Models. In: The Eleventh International Conference on Learning Representations; 2022.
  13. Liang PP, Wu C, Morency LP, Salakhutdinov R. Towards Understanding and Mitigating Social Biases in Language Models. In: Meila M, Zhang T, editors. Proceedings of the 38th International Conference on Machine Learning. vol. 139 of Proceedings of Machine Learning Research. PMLR; 2021. p. 6565–6576. Available from: https://proceedings.mlr.press/v139/liang21a.html.

Cited by

See all 16

Metadata

Authors
Shardlow, M., & Przybyła, P.
Affiliations
  • Manchester Metropolitan University (GB)
  • Universitat Pompeu Fabra (ES)
Corresponding author
Shardlow, Matthew
Venue
PLoS ONE
Published
2024-12-04
Volume
19
Issue
12
Pages
e0307521
ISSN
1932-6203 (ISSN-L)
OpenAlex
W4405017998
Cited by
16
References
80
Citations by year
2026: 7 · 2025: 6 · 2024: 2
FWCI
6.11
Type
journal article
Language
en
Open access
Locations
  • PLoS ONE (publishedVersion, cc-by)
  • PubMed Central (publishedVersion, cc-by)
  • PubMed (publishedVersion)
  • Repositori digital de la UPF (Universitat Pompeu Fabra) (publishedVersion, cc-by)
  • e-space (Manchester Metropolitan University) (acceptedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • arXiv (Cornell University)
Funders
HORIZON EUROPE Framework Programme (101060930)
Topics
  • Social Sciences › Arts and Humanities › Philosophy › Philosophy and Theoretical Science
  • Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
  • Social Sciences › Arts and Humanities › History and Philosophy of Science › Cybernetics and Technology in Society
Keywords
large language models, sentience, consciousness, LaMDA, transformer models, integrated information theory, anthropomorphism, chatbots, ChatGPT, AI ethics, language modeling
MeSH terms
Consciousness, Humans, Language, Models, Theoretical, Natural Language Processing

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.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

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

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Deanthropomorphising NLP: Can a language model be conscious? https://app.cochat.ai/papers/10.1371/journal.pone.0307521
The paper
Shardlow, M., & Przybyła, P. (2024). Deanthropomorphising NLP: Can a language model be conscious? PLoS ONE. https://doi.org/10.1371/journal.pone.0307521

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.