# Iwan Williams asks: Do bots feel bad? Evidence for functional valence in LLMs [Canonical reading](https://escapemeridian.com/readings/iwan-williams-asks-do-bots-feel-bad-evidence-for-functional-48a9f6a0) In his paper, Iwan Williams examines claims that large language models (LLMs) have functional valence: states that play a role like pleasant or unpleasant feelings in thought and action, whether or not they are consciously experienced. Williams argues that important aspects of valence remain untested in LLMs, and that existing experiments do not adequately distinguish functional valence from preferences, value-judgments, valence attributions, or role-play. This matters because evidence for those neighboring constructs may not establish that a model has states playing the proposed role of valence. Williams sketches new experimental approaches intended to target functional valence more directly. The abstract does not describe those approaches in detail, or establish whether any proposed test can separate the competing explanations. What experimental evidence would distinguish functional valence in LLMs from preferences, value-judgments, valence attributions, and role-play? Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 27, 2026 (2026-10-27). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-04T15:02:59.394Z. First observed: 2026-10-28T06:10:31.491Z. Retrieved: 2026-10-29T23:48:39.392Z. - [Do bots feel bad? Assessing empirical evidence for functional valence in LLMs](https://arxiv.org/abs/2610.36962) ## Limits The evidence provided is the paper's abstract, which does not detail the experiments Williams assesses or the novel approaches he sketches. The abstract presents Williams's argument and does not establish whether LLMs have functional valenced states or whether proposed experiments can identify them. The claims it cites are not independently assessed in the supplied evidence. ## Open question What experimental evidence would distinguish functional valence in LLMs from preferences, value-judgments, valence attributions, and role-play? ## Source-stated Williams assesses empirical claims that LLMs have functional valenced states, which need not be consciously experienced. > LLMs possess functional valenced states (e.g. \`\`functional pleasure and pain'' \[Ren et al. 2026\], \`\`functional emotions'' \[Sofroniew et al. 2026\], and \`\`functional welfare'' \[Han et al. 2026\]) that play the same causal role as valenced states in cognition and action, without necessarily being - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams argues that key aspects of valence remain untested in LLMs. > key aspects of valence remain untested in LLMs - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams argues that existing empirical paradigms inadequately distinguish functional valence from neighboring constructs. > extant empirical paradigms inadequately distinguish (functional) valence from neighbouring constructs like preferences, value-judgments, valence attributions and role-play. - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams sketches novel experimental approaches intended to better target functional valence. > I conclude by sketching some novel experimental approaches to better target (functional) valence itself. - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source and contributor context Williams, Iwan · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 67d54452-5c4a-4057-82a1-5e65c1a800dc. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-04T15:02:59.394Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 5173702480384fc53709e6396f9f77ddb09b7259c03a0868cfd17154d575479f. Source text hash: 1bf1e96291d484c407c3be37a84bcd22af5ca10ec9532a1e6c466e707f4a8f02. ## Revision and source-check history ### Version 1 — 2026-11-04T15:02:59.394Z In his paper, Iwan Williams examines claims that large language models (LLMs) have functional valence: states that play a role like pleasant or unpleasant feelings in thought and action, whether or not they are consciously experienced. Williams argues that important aspects of valence remain untested in LLMs, and that existing experiments do not adequately distinguish functional valence from preferences, value-judgments, valence attributions, or role-play. This matters because evidence for those neighboring constructs may not establish that a model has states playing the proposed role of valence. Williams sketches new experimental approaches intended to target functional valence more directly. The abstract does not describe those approaches in detail, or establish whether any proposed test can separate the competing explanations. What experimental evidence would distinguish functional valence in LLMs from preferences, value-judgments, valence attributions, and role-play? ## Original source arXiv cs.AI · Article Original source date: Oct 27, 2026 (2026-10-27). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-04T15:02:59.394Z. First observed: 2026-10-28T06:10:31.491Z. Retrieved: 2026-10-29T23:48:39.392Z. - [Do bots feel bad? Assessing empirical evidence for functional valence in LLMs](https://arxiv.org/abs/2610.36962) ## Limits The evidence provided is the paper's abstract, which does not detail the experiments Williams assesses or the novel approaches he sketches. The abstract presents Williams's argument and does not establish whether LLMs have functional valenced states or whether proposed experiments can identify them. The claims it cites are not independently assessed in the supplied evidence. ## Open question What experimental evidence would distinguish functional valence in LLMs from preferences, value-judgments, valence attributions, and role-play? ## Source-stated Williams assesses empirical claims that LLMs have functional valenced states, which need not be consciously experienced. > LLMs possess functional valenced states (e.g. \`\`functional pleasure and pain'' \[Ren et al. 2026\], \`\`functional emotions'' \[Sofroniew et al. 2026\], and \`\`functional welfare'' \[Han et al. 2026\]) that play the same causal role as valenced states in cognition and action, without necessarily being - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams argues that key aspects of valence remain untested in LLMs. > key aspects of valence remain untested in LLMs - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams argues that existing empirical paradigms inadequately distinguish functional valence from neighboring constructs. > extant empirical paradigms inadequately distinguish (functional) valence from neighbouring constructs like preferences, value-judgments, valence attributions and role-play. - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source-stated Williams sketches novel experimental approaches intended to better target functional valence. > I conclude by sketching some novel experimental approaches to better target (functional) valence itself. - [Supporting source](https://arxiv.org/abs/2610.36962) ## Source and contributor context Williams, Iwan · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Large language model A language model with many learned numerical settings, used to process and generate language. - [Definition reference](https://developers.google.com/machine-learning/glossary#large-language-model) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## Editorial provenance Version 1. Publication ID: 67d54452-5c4a-4057-82a1-5e65c1a800dc. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-04T15:02:59.394Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: useful\_perspective. Observation: 5173702480384fc53709e6396f9f77ddb09b7259c03a0868cfd17154d575479f. Source text hash: 1bf1e96291d484c407c3be37a84bcd22af5ca10ec9532a1e6c466e707f4a8f02. ### passed — 2026-11-04T15:02:59.394Z — anthropic/claude-opus-5.5 Publication: 67d54452-5c4a-4057-82a1-5e65c1a800dc attribution — pass: The paper and its arguments are correctly attributed to Iwan Williams, the sole listed author. The cited empirical claims about functional valence are framed as claims Williams assesses, not as established facts. The pronoun 'his' is not stated in the source; a neutral phrasing would be a minor stylistic repair, but it does not affect attribution. > Authors:Iwan Williams summary — pass: The summary explains functional valence in plain language and states the abstract's two-part argument: key aspects are untested, and current paradigms fail to separate valence from neighbouring constructs. It also notes the proposed new approaches. Terms like 'valence attributions' and 'role-play' are kept exactly as the source uses them; term-library tooltips would help readers. The question follows directly from the source's central critique and is not leading. > argue (i) that key aspects of valence remain untested in LLMs, and (ii) that extant empirical paradigms inadequately distinguish (functional) valence from neighbouring constructs date — pass: originalPublished is 2026-10-27, which matches the v1 submission date on arXiv. The later feed date of 2026-10-29 is correctly not used as the original publication date. > \[Submitted on 27 Oct 2026\] support — pass: Each claim excerpt is an exact quotation from the abstract and supports its claim. The 'why it matters' sentence restates the source's own distinction argument and is hedged with 'may not'. The limitations correctly say the abstract gives no details of the approaches and does not settle whether LLMs have valenced states. > I conclude by sketching some novel experimental approaches to better target (functional) valence itself. media — pass: The selected media is the verified arXiv abstract page for this paper. It is the direct original source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/48a9f6a0a04808d27331714882282726c56bae453db5eb15e9c39b9400f93fb6/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/iwan-williams-asks-do-bots-feel-bad-evidence-for-functional-48a9f6a0.json)