Elamur Research⌕
☰
home / papers / 2024 / Chatting with Bots: AI, Speech Acts, and the Edge of Assertion

Chatting with Bots: AI, Speech Acts, and the Edge of Assertion

Paper · 2024 · Iwan Williams and Tim Bayne

Summary

The paper asks whether LLM-powered chatbots such as ChatGPT or Claude can genuinely assert. The authors call the affirmative view the Thesis of Chatbot Assertion (TCA). They first argue that TCA deserves serious consideration. They then set out four strong objections to it, which creates a dilemma. This is a conceptual, philosophical study, with no data or experiments. The authors examine two existing ways out of the dilemma, proxy-assertion and fictionalism, and reject both. Drawing on how assertion develops in children, they propose an intermediate category of proto-assertion. A toddler stands roughly to a full asserter as toddling stands to walking. Chatbots, they argue, sit in the same in-between zone: they have some marks of assertion, or have them only partially. The authors see the main contribution less in classifying today's chatbots and more in the general claim that the capacity to assert comes in degrees. That claim matters for how we describe, evaluate and hold accountable conversational AI.

Key points

  • Three motivations for TCA:
    • Fine-tuned models may acquire "descriptive functions", meaning their job is to output true or informative sentences (Butlin & Viebahn).
    • Chatbots appear to switch between speech-act modes such as asking, warning, pretending and quoting. Worries about "hallucination" or "confabulation" already assume that some outputs, unlike fiction requested by a prompt, present things as facts.
    • Chatbots behave like asserters. They defend claims against challenges, give reasons, avoid immediate self-contradiction, and retract unsupported claims.
  • Four objections to TCA:
    • Understanding: purely text-based systems may lack grounded concepts.
    • Attitudes: chatbots may lack beliefs and intentions.
    • Mentalizing: Grice's account requires layered intentions about the hearer.
    • Normativity: Butlin & Viebahn's sanctionability conditions, which chatbots only partly meet.
  • Rival views rejected: proxy-assertion (the chatbot asserts on someone else's behalf) and fictionalism.
  • Control case: a self-calibrating thermometer meets two sanctionability conditions. It still lacks the other assertion features, such as defending claims, switching modes, and having belief-like states. This keeps the proto-assertion category from becoming trivial.
  • Contrast with prior work: Freiman & Miller's "quasi-testimony" rests only on phenomenological resemblance plus basic epistemic norms, and treats it as a kind of testimony. Proto-assertion is multi-dimensional and separate from full assertion.

Interesting ideas

  • I1. Treat speech-act status as a graded, multi-dimensional profile rather than a yes/no label, scored separately on several component capacities.
  • I2. The same false sentence counts as an error under an assertive mode but is harmless inside a requested fiction, so the force of the mode decides whether content is presented as fact.
  • I3. Judge an asserter by its behaviour after the utterance: defending under challenge, explaining reasons, staying consistent, retracting when refuted.
  • I4. Check any graded category against a minimal control (the self-calibrating thermometer) that satisfies some criteria but clearly should not qualify, to show the category is not trivial.
  • I5. Uttering a sentence with understanding is not enough for assertion; microphone tests, pronunciation practice or quoting carry no assertive force.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (frames): crate prag could output each assertion candidate with a vector of force cues (commitment markers, hedges, fiction framing, quotation) instead of one discrete speech-act tag, so the decision stays explainable.
  • I2 (world): statements inside fiction, quotation or pretence frames should go into a separate sub-world, not into the factual world state, so that fables do not pollute compiled knowledge.
  • I3 (answer): the readers could be tested by follow-up probes asking them to justify an answer, to remain consistent across rephrasings, and to retract when given counter-evidence.
  • I4 (learn): when defining silver-data quality or "understanding" metrics, add negative-control systems that game a single criterion, to confirm the metric needs several dimensions at once.
  • I5 (frames): speech-act detection should use contextual purpose cues (rehearsal, citation, example sentences) to downgrade sentences that look declarative but are not commitments.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Monash UniversityMelbourne · Australia64 works · 54,972 citations🔬Canadian Institute for Advanced ResearchToronto · Canada35 works · 19,316 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 17,450 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 worksPhilosophy · 287 worksHistory and Philosophy of Science · 62 works

🧭 Topics 1Research subjects and languages the work is about.Speech act theory · 229 works

Cite

Cite as Iwan Williams and Tim Bayne. 2024. Chatting with Bots: AI, Speech Acts, and the Edge of Assertion. Inquiry (2024).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1080/0020174x.2024.2434874 · arXiv 2410.16645

@inproceedings{williams2024chatting,
  title = {{Chatting with Bots: AI, Speech Acts, and the Edge of Assertion}},
  author = {Iwan Williams and Tim Bayne},
  booktitle = {Inquiry (2024)},
  year = {2024},
  doi = {10.1080/0020174x.2024.2434874},
  eprint = {2410.16645},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.1080/0020174x.2024.2434874}
}

References

72 references, 11 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdou, M., Kulmizev, A., Hershcovich, D., Frank, S., Pavlick, E., & Søgaard, A. (2021). Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color. Proceedings of the 25th Conference on Computational Natural Language Learning. https://doi.org/10.18653/v1/2021.conll-1.9 ↗
  2. Abreu Zavaleta, M. (2023). Partial understanding. https://doi.org/10.1007/s11229-023-04268-2 ↗
  3. Adler, J. E. (2002). Belief's own https://doi.org/10.1080/15665399.2003.10819759 ↗
  4. Alston, W. P. (2000). Illocutionary Acts and Sentence Meaning. Ithaca, NY: Cornell University Press. find
  5. Arora, C. (2024). Proxy Assertions and Agency: The Case of Machine-Assertions. Philosophy & Technology, 37(1), 1–19. https://doi.org/10.1007/s13347-024-00703-5 ↗
  6. Azaria, A., & Mitchell, T. (2023). The Internal State of an LLM Knows When It’s Lying. Findings of the Association for Computational Linguistics: EMNLP 2023, 967–976. https://doi.org/10.18653/v1/2023.findings-emnlp.68 ↗
  7. Bayne, T., & Hattiangadi, A. (2013). Belief and its Bedfellows. In N. Nottelmann (Ed.), New essays on belief: constitution, content and structure (pp. 124–144). London: Palgrave Macmillan. https://doi.org/10.1057/9781137026521 Synthese, 202(2), 1–32. ethics. MIT Press. ↗
  8. Bender, E. M., & Koller, A. (2020). Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data. Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 5185–5198. ↗
  9. Brandom, R. (1994). Making it explicit: Reasoning, representing, and discursive commitment. Harvard University Press. find
  10. Browning, J. (2023). “Personhood and AI: Why large language models don’t understand us”. AI & SOCIETY, . https://doi.org/10.1007/s00146-023-01724-y ↗
  11. Burns, C., Haotian, Y., Klein, D., & Steinhardt, J. (2024). Discovering latent knowledge in Language Models Without Supervision. 1–28. arXiv Preprint: https://arxiv.org/abs/2212.03827 ↗
  12. Butlin, P. (2023). Sharing Our Concepts with Machines. Erkenntnis, (88), 3079–3095. https://doi.org/10.1007/s10670-021-00491-w ↗
  13. Butlin, P., & Viebahn, E. (2023). AI Assertion. https://doi.org/10.7560/783812-007 ↗
  14. Butterfill, S. (2020). The Developing Mind: A Philosophical Introduction (1st ed.). Routledge. https://doi.org/10.4324/9780203758274 ↗
  15. Cappelen, H., & Dever, J. (2021). Making AI Intelligible. Oxford: Oxford University Press. ↗
  16. Carruthers, P. (2013). Mindreading in Infancy. 28(2), 141–172. ↗
  17. Dummett, M. (1973). Frege: Philosophy of Language. Duckworth. find
  18. Evans, O., Cotton-Barratt, O., Finnveden, L., Bales, A., Balwit, A., Wills, P., … Saunders, W. (2021). Truthful AI. 1–96. arXiv Preprint: https://arxiv.org/abs/2110.06674 ↗
  19. Freiman, O. (2024). AI-Testimony, Conversational AIs and Our Anthropocentric Theory of Testimony. Social Epistemology, 38(4), 476–490. https://doi.org/10.1080/02691728.2024.2316622 ↗
  20. Freiman, O., & Miller, B. (2018). Can Artificial Entities Assert ? In S. Goldberg (Ed.), The Oxford Handbook of Assertion. Oxford University Press. ↗
  21. García-Carpintero, M. (2019). Conventions and Constitutive Norms. Journal of Social Ontology, 5(1), 35–52. https://doi.org/10.1515/jso-2019-0013 ↗
  22. Glaese, A., Mcaleese, N., Trebacz, M., Aslanides, J., Firoiu, V., Ewalds, T., … Irving, G. (2022). Improving alignment of dialogue agents via targeted human judgements. 1–77. arXiv Preprint: https://arxiv.org/abs/2209.14375 ↗
  23. Graham, P. J. (2018). The Function of Assertion and Social Norms. In S. Goldberg (Ed.), The Oxford Handbook of Assertion (pp. 725–748). https://doi.org/10.1093/oxfordhb/9780190675233.013.30 ↗
  24. Green, M., & Michel, J. G. (2022). What Might Machines Mean? Minds and Machines, 32(2), 323–338. https://doi.org/10.1007/s11023-022-09589-8 ↗
  25. Grice, H. Paul (1957). Meaning. Philosophical Review, 66, 377–88. find
  26. Gubelmann, R. (2024). Large Language Models, Agency, and Why Speech Acts are Beyond Them (For Now) – A Kantian-Cum-Pragmatist Case. Philosophy and Technology, 37(1), 1–24. https://doi.org/10.1007/s13347-024-00696-1 ↗
  27. Gurnee, W., & Tegmark, M. (2023). Language Models Represent Space and Time. arXiv Preprint: http://arxiv.org/abs/2310.02207 ↗
  28. Hamblin, C. L. (1970). The effect of when it's said. Theoria, 36(3), 249-263. ↗
  29. Harding, J. (2023). Operationalising Representation in Natural Language Processing. The British Journal for the Philosophy of Science, 1–36. https://doi.org/10.1086/728685 ↗
  30. Hendel, R., Geva, M., & Globerson, A. (2023). In-Context Learning Creates Task Vectors. Findings of the Association for Computational Linguistics: EMNLP 2023, 9318–9333. https://doi.org/10.18653/v1/2023.findings-emnlp.624 ↗
  31. Herrmann, D. A., & Levinstein, B. A. (2024). Standards for Belief Representations in LLMs. ↗
  32. Horton, W. S. (2018). Assertion and Mindreading. In S. Goldberg (Ed.), The Oxford Handbook of Assertion. Oxford University Press. ↗
  33. Kim, H. H. (2023, June). If Pinocchio Doesn't Freak You Out, Sydney Shouldn't Either. WIRED. https://www.wired.com/story/artificial-intelligence-fiction-philosophy-consciousness/ ↗
  34. Kosinski, M. (2024). Evaluating Large Language Models in Theory of Mind Tasks. ↗
  35. Lake, B. M., & Murphy, G. L. (2021). Word Meaning in Minds and Machines. Psychological Review. https://doi.org/10.1037/rev0000297 ↗
  36. Lavelle, J. S. (2024). Growth From Uncertainty: Understanding the Replication ‘Crisis’ in Infant Cognition. Philosophy of Science, 91(2), 390–409. https://doi.org/10.1017/psa.2023.157 ↗
  37. Levinstein, B. A., & Herrmann, D. A. (forthcoming). Still No Lie Detector for Language Models: Probing Empirical and Conceptual Roadblocks. Philosophical Studies. http://arxiv.org/abs/2307.00175 ↗
  38. Li, K., Hopkins, A. K., Bau, D., Viégas, F., Pfister, H., & Wattenberg, M. (2023). Emergent World Representations: Exploring a Sequence Model Trained on a Synthetic Task. The Eleventh International Conference on Learning Representations, 1–17. https://openreview.net/forum?id=DeG07_TcZvT ↗
  39. Ludwig, K. (2018). Proxy Assertion. In S. Goldberg (Ed.), The Oxford Handbook of Assertion. Oxford University Press. find
  40. Mahowald, K., Ivanova, A. A., Blank, I. A., Kanwisher, N., Tenenbaum, J. B., & Fedorenko, E. (2023). Dissociating language and thought in large language models. Trends in Cognitive Sciences, 1–24. https://doi.org/10.1016/j.tics.2024.01.011 ↗
  41. Mallory, F. (2023). Fictionalism about Chatbots. Ergo an Open Access Journal of Philosophy, 10(0), 1082–1100. https://doi.org/10.3998/ergo.4668 ↗
  42. Marks, S., & Tegmark, M. (2024). The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets. Conference on Language Modeling. http://arxiv.org/abs/2310.06824 ↗
  43. Marsili, N. (2018). Truth and assertion : rules versus aims. 78(4), 638–648. https://doi.org/10.1093/analys/any008 ↗
  44. Marsili, N. (2023). The norm of assertion: a ‘constitutive’ rule? Inquiry (United Kingdom), 66(5), 799–820. https://doi.org/10.1080/0020174X.2019.1667868 ↗
  45. Marsili, N. (2024). The definition of assertion: Commitment and truth. ↗
  46. Mind and Language, (February 2023), 1–21. https://doi.org/10.1111/mila.12476 ↗
  47. Mitchell, M., & Krakauer, D. C. (2022). The Debate Over Understanding in AI’s Large Language Models. 120(13), 1–5. Retrieved from http://arxiv.org/abs/2210.13966 ↗
  48. Mollo, D. C., & Millière, R. (2023). The Vector Grounding Problem. 1–34. arXiv Preprint: https://arxiv.org/abs/2304.01481 ↗
  49. Nanda, N., Lee, A., & Wattenberg, M. (2023). Emergent Linear Representations in World Models of Self-Supervised Sequence Models. BlackboxNLP 2023 - Analyzing and Interpreting Neural Networks for NLP, Proceedings of the 6th Workshop, 16–30. https://doi.org/10.18653/v1/2023.blackboxnlp-1.2 ↗
  50. Nickel, P. J. (2013). Artificial speech and its authors. Minds and Machines, 23(4), 489–502. https://doi.org/10.1007/s11023-013-9303-9 ↗
  51. Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., … Lowe, R. (2022). Training language models to follow instructions with human feedback. arXiv Preprint: http://arxiv.org/abs/2203.02155 ↗
  52. Pagin, P. (2016). Problems with Norms of Assertion. Philosophy and Phenomenological Research, 93(1), 178–207. https://doi.org/10.1111/phpr.12209 ↗
  53. Pagin, P. & Marsili, N. (2021). Assertion. The Stanford Encyclopedia of Philosophy (Winter 2021 Edition), Edward N. Zalta (ed.), URL = <https://plato.stanford.edu/archives/win2021/entries/assertion/>. ↗
  54. Piantadosi, S. T., & Hill, F. (2022). Meaning without reference in large language models. ↗
  55. Preston, J. & M. Bishop. (2002). Views into the Chinese Room: New Essays on Searle and Artificial Intelligence. New York: Oxford University Press. find
  56. Rakoczy, H., & Tomasello, M. (2009). Done wrong or said wrong? Young children understand the normative directions of fit of different speech acts. Cognition, 113(2), 205–212. https://doi.org/10.1016/j.cognition.2009.07.013 ↗
  57. Perner, J., & Roessler, J. (2014). From infants’ to children’s appreciation of belief. Trends in Cognitive Sciences, 16(10), 518–524. https://doi.org/10.1016/j.tics.2012.08.004 ↗
  58. Schwitzgebel, E. (2001). In-between believing. The Philosophical Quarterly, 51(202), 76–82. ↗
  59. Shapira, N., Levy, M., Alavi, S. H., Zhou, X., Choi, Y., Goldberg, Y., … Shwartz, V. (2024). Clever Hans or Neural Theory of Mind? Stress Testing Social Reasoning in Large Language Models. EACL 2024 - 18th Conference of the European Chapter of the Association for Computational Linguistics, Proceedings of the Conference, 1, 2257–2273. find
  60. Simion, M., & Kelp, C. (2018). Assertion: The Constitutive Norms View. In S. Goldberg (Ed.), The Oxford Handbook of Assertion (pp. 57–73). https://doi.org/10.1093/oxfordhb/9780190675233.013.2 ↗
  61. Stalnaker, R., (1974). “Pragmatic Presuppositions”, in M. K. Munitz and P. Unger (eds.) Semantics and Philosophy. New York: New York University Press, pp. 197–213. find
  62. Stalnaker, R., (1978). “Assertion”, in Peter Cole (ed.), Syntax and Semantics 9: Pragmatics. New York: New York Academic Press, 315–332. find
  63. Stich, S. P. (1979). Do animals have beliefs? Australasian Journal of Philosophy, 57(1), 15– 28. https://doi.org/10.1080/00048407912341011 Strachan, J. W. A., Albergo, D., Borghini, G., Pansardi, O., Scaliti, E., Gupta, S., … Becchio, C. (2024). Testing theory of mind in large language models and humans. Nature Human Behaviour, 8(July). https://doi.org/10.1038/s41562-024-01882-z ↗
  64. Strasser, A. & Schwitzgebel, E. (2024). Towards asymmetric joint actions. In A. Strasser (Ed.) Anna's AI Anthology: How to live with smart machines. Berlin: Xenomoi Verlag. find
  65. Todd, E., Li, M. L., Sharma, A. Sen, Mueller, A., Wallace, B. C., & Bau, D. (2023). Function Vectors in Large Language Models. (c), 1–52. Retrieved from http://arxiv.org/abs/2310.15213 ↗
  66. Ullman, T. (2023). Large Language Models Fail on Trivial Alterations to Theory-of-Mind Tasks. Retrieved from http://arxiv.org/abs/2302.08399 ↗
  67. van Woudenberg, R., Ranalli, C., & Bracker, D. (2024). Authorship and ChatGPT: a Conservative View. Philosophy and Technology, 37(1), 1–26. https://doi.org/10.1007/s13347-024-00715-1 ↗
  68. Westra, E., & Nagel, J. (2021). Mindreading in conversation. Cognition, 210(January). https://doi.org/10.1016/j.cognition.2021.104618 ↗
  69. Williams, B. (1973). Problems of the Self. Cambridge: Cambridge University Press. ↗
  70. Williamson, T. (2000). Knowledge and Its Limits. Oxford: Oxford University Press. find
  71. Wright, C. (1992). Truth and objectivity. Harvard University Press. ↗
  72. Yildirim, I., & Paul, L. A. (2024). From task structures to world models : what do LLMs know ? Trends in Cognitive Sciences, 1–12. https://doi.org/10.1016/j.tics.2024.02.008 ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/williams-2024-chatting-bots-ai-speech-acts/