{"id":"74b9ad11-272b-4aac-b07d-fc59c2fc074f","arxiv_id":"2509.02624","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Community workshops with 22 participants yield four ethical questions (safety, beneficiaries, data ownership, necessity) as a reflective framework for wellbeing robot design.","lead":"This paper ran ethics workshops with 22 people from three groups rarely included in robotics design, and turned their discussions into four questions for wellbeing robot developers: is it safe, who is it for, who owns the data, and why a robot at all. The value is a short, community-grounded prompt set that robot builders could use before deploying wellbeing coaches in the real world.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Instrument-direction: the canvas's six explicit dimensions pre-structure most of the four final themes, undermining the claim that they emerged from community engagement.","rationale":"The reader's weakest_assumption identifies the instrument-direction risk, and the manuscript itself supports it: §3 lists six explicit canvas dimensions, §4.1–4.3 map almost one-to-one onto three of them, §4.4 overlaps with transparency/emotional consideration, and §5.0.2 concedes the canvas directed conversation. This is exactly where the central claim—that the four questions are community-grounded and emerged from workshops—is least secure. I considered the small convenience sample (n=22, G3 n=4) and the unrecorded workshops as alternative concerns; both limit transferability and auditability but do not by themselves refute the framework's value. The instrument-direction issue is load-bearing because if the themes are largely restatements of the canvas, the contribution is less novel and the claim to have 'identified through community-based investigations' is overstated. The authors acknowledge the limitation and say they expanded beyond the six dimensions, but they provide no evidence of which sub-themes were unprompted; the appendix quote tables are drawn from the canvas, so they cannot demonstrate unprompted emergence. A blind re-analysis would settle this. I therefore see no reason to move the reader's CONDITIONAL verdict: the paper is honest, well-structured, and useful, but should soften the 'emerged from community' language, provide evidence of unprompted themes, or reframe the contribution as a synthesis of prompted community reflection. The attribution mismatches (e.g., 'getting attacked by the robot' assigned to G1P05 in the body and G1P06 in the appendix) should also be corrected, and the conclusion's extension to 'other social robot applications' needs justification or softening.","tokens_in":26887,"tokens_out":5145,"duration_ms":56031,"concrete_test":"Conduct an independent blind re-analysis: give the raw canvas text (and, if available, de-identified researcher notes) to two coders with no knowledge of the Canvas's six dimensions or the paper's four themes, and have them perform inductive thematic analysis. Compare the emergent structure to the paper's four themes and to the six canvas dimensions. If independent coders reproduce the four themes without exposure to the canvas, the community-emergence claim survives; if they recover the six canvas dimensions—or the four themes are direct restatements of prompt categories—the instrument-direction concern is confirmed. Complement with a count per final sub-theme in §4.1–4.4 of whether every constituent quote in Appendix Tables 1–2 appears in a canvas section whose prompt matches that sub-theme; flag any sub-theme with zero quotes falling outside the prompted dimensions. This is a quantita","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3 states the Social Robot Co-Design Canvas on Ethics \"addresses six ethical issues explicitly: physical safety, data security, transparency, equality across users, emotional consideration, and behaviour enforcement.\" The four final themes map directly onto these: §4.1 safety ↔ physical/emotional safety; §4.3 ownership/data ↔ data security; §4.2 built-for/with ↔ equality across users; §4.4 why-a-robot overlaps with transparency and emotional consideration. Section 5.0.2 concedes that \"the content of the canvases themselves did direct the topics of conversation.\" If the instrument supplied the thematic structure, the central deliverable is a re-organisation of the authors' prior canvas (Axelsson et al. 2021b) rather than a community-emergent framework, and the abstract/conclusion claim that the questions were \"identified through three community-based investigations\" is weakened. The analysis is described as inductive (Sec. 3), but the data-collection instrument is not neutral: participant quotes are real, but they are responses to pre-selected prompts. The paper mitigates by saying they expanded beyond the six dimensions, yet no evidence is given which sub-themes were unprompted; for example, \"why a robot?\" could be a synthesis of transparency and emotional consideration prompts rather than an independent community concern. The unrecorded workshops (Sec. 3) mean the researcher notes are the only record of unprompted discussion, and those notes are not reproduced, so the claimed expansion cannot be independently checked.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper reports three community workshops (n=22) in which members of the public, women computer scientists, and humanities researchers used the Social Robot Co-Design Canvas on Ethics to reflect on wellbeing robots. Through Thematic Analysis of canvas responses and researcher notes, the authors identify four ethical and socio-technical questions: (1) Is the robot safe and how can we know that? (2) Who is the robot built for and with? (3) Who owns the robot and the data? (4) Why a robot? These are proposed as a reusable framework for roboticists to reflect on ethics and socio-technical dimensions during design, development, and deployment, and to support dialogue with communities.","tokens_in":27090,"tokens_out":4547,"duration_ms":55512,"significance":"If the four questions are genuinely community-grounded, the paper offers a concise, memorable reflective tool for an area—wellbeing robots—where anticipatory ethics is still underdeveloped. The work has clear strengths: it engages three under-represented groups, provides participant quotes in the appendix that substantiate each theme, includes positionality statements, and explicitly discusses power, privacy, and systemic factors. The proposed questions are practically useful as conversation starters, and the authors are transparent about several limitations, including the canvas's influence on discussion. However, the central empirical claim—that the questions emerged from community reflection—is only partially supported, because the data-collection instrument already explicitly structures most of the final thematic space. The contribution's novelty and validity depend on whether the framework is a community-emergent synthesis or a re-organisation of the authors' prior canvas dimensions.","major_comments":[{"comment":"The canvas is described as explicitly addressing six ethical issues: physical safety, data security, transparency, equality across users, emotional consideration, and behaviour enforcement. Three of the four final themes map almost directly onto these dimensions (§4.1 ↔ physical/emotional safety; §4.3 ↔ data security; §4.2 ↔ equality across users). Section 5.0.2 concedes that 'the content of the canvases themselves did direct the topics of conversation.' The paper asserts that the analysis 'expanded' beyond the six dimensions, but provides no evidence which sub-themes were unprompted; for example, 'Why a robot?' could be a synthesis of transparency and emotional consideration prompts rather than an independent community concern. Because the abstract and conclusion claim that the four questions were 'identified through three community-based investigations,' this is load-bearing. Please pr","section":"§3, §5.0.2"},{"comment":"The workshops and interviews were deliberately unrecorded to protect participants' privacy; the analysis therefore rests on canvas text and researcher notes. The notes are not reproduced or summarised in any systematic way. This makes it impossible for a reader to verify which themes and sub-themes actually arose in conversation rather than being prompted by the canvas, or to assess whether the notes faithfully preserve participant views. At minimum, the paper should provide an anonymised thematic summary of the notes, or explicitly identify which themes (e.g., employer surveillance, 'why a robot') rely on notes rather than canvas text.","section":"§3"},{"comment":"The Conclusion states that the four questions 'are applicable to other social robot applications' and the abstract presents them as a framework that 'roboticists can and should use' broadly. This prescriptive generalisation goes beyond the evidence: the sample is three small, self-selected groups (n=22) recruited at University of Cambridge events, with limited demographic diversity and no demographics for G1. The paper carefully frames the questions as a starting point in §5, but the abstract and conclusion make a stronger, less warranted claim. I recommend softening these claims to scope them to wellbeing robots and to frame the questions as hypotheses or discussion tools requiring further validation across contexts.","section":"§6, Abstract"}],"minor_comments":[{"comment":"In §4.1.1 the text attributes 'getting attacked by the robot' to G1P05, but Table 1 lists it under G1P06. Please check participant IDs for consistency.","section":"Table 1"},{"comment":"The ordering of the four layers is described inconsistently. The text talks about an expanding circle from 'person' to 'social group' to 'society' to 'culture,' but the figure caption and the list in Figure 1 appear to place 'Culture' before 'Social group' and 'Society.' Clarify the intended ordering.","section":"Figure 1"},{"comment":"The table headers contain a typo: 'T opic' should be 'Topic.' Also, formatting of participant references is inconsistent (e.g., 'G1P01' vs. 'G1, P6' in §4.4.1).","section":"Tables 1 and 2"},{"comment":"The claim of data saturation is asserted with a citation but not demonstrated. Given the heterogeneity of the three groups and the small group sizes, a brief justification of why saturation was reached would be useful.","section":"§3"},{"comment":"The discussion of future methods (citizens' assemblies, AI Hopes and Fears) is interesting but somewhat disconnected from the present data; it could be shortened or moved to a separate 'future work' subsection to keep the Discussion focused.","section":"§5.0.1"}],"recommendation":"major_revision","confidential_remarks":"The instrument-direction concern raised in the review is genuine, but the authors are unusually transparent about the canvas's influence, and the paper already contains the seed of the needed revision. I do not see this as a case for rejection: the four questions are useful and the participant quotes are real, but the framing as 'community-emergent' needs to be either substantiated with a source-mapping or softened. The paper may be a strong fit for a qualitative HRI or ethics venue; the main risk is that the contribution is seen as a restatement of the authors' earlier canvas. With the analysis made more transparent and the claims appropriately scoped, the paper could make a solid contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know at the outset: this is one of the more honest qualitative ethics papers I've read in this space, but its central deliverable is less community-generated than the abstract implies. The four questions are a sane, usable synthesis, and the appendix quotes largely support them. But the instrument did a lot of the work, and the authors admit as much in Section 5.0.2.\n\nWhat's genuinely new is the empirical base and the candour. Workshops with 22 people from three under-represented groups—science-festival public, women in CS, humanities researchers—give a useful cross-section of voices that usually aren't in the robotics room. The full quote tables are a real strength: you can check the themes against the raw data. The positionality statements and the discussion's self-criticism, especially the concession that the canvas directed conversation, are more than most papers manage.\n\nThe soft spot is instrument direction. The canvas they used explicitly prompts six ethics dimensions; three of the four final themes—safety, data, equality—map almost directly onto those. The fourth, 'Why a robot?', has some independent support but overlaps with transparency and emotional consideration. The authors say they expanded beyond the six dimensions, but because the workshops were deliberately unrecorded, the only evidence of that expansion is researcher notes we can't inspect. So the claim that the questions 'emerged' through community engagement is overstated; a more careful phrasing would be that the communities' reflections, prompted by the canvas, were organised into four useful questions.\n\nTwo minor issues: the conclusion extends the questions to all social robot applications from a wellbeing-coach sample of 22, which is a stretch; and a couple of participant attributions differ between body text and appendix (e.g., the 'white western users' quote is G3P03 in the text, G2P03 in Table 1). Both are fixable.\n\nThe central argument holds up if you read the paper as a modest synthesis rather than a discovery. The reader's conditional verdict is about right. The paper deserves a serious referee: it is clear, reproducible in its process (the canvas is public), and the honest treatment of limitations makes it a good model for reporting this kind of work. I'd want the authors to soften the emergence claim and fix the errors, but I wouldn't desk-reject it.","headline":"Useful, honest, but partly instrument-generated: the four questions are a sane synthesis, not a cleanly community-emergent result.","tokens_in":27689,"tokens_out":4769,"would_cite":true,"duration_ms":50070,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that four broad questions—safety and its evidence, who the robot is for and with, ownership of the robot and its data, and why deploy a robot at all—form a reusable framework that roboticists should use to reflect on ethic","keywords":["robot ethics","wellbeing robots","community engagement","participatory design","thematic analysis","socially assistive robotics","data ownership","anticipatory ethics"],"falsifier":"Re-run the workshops with full audio recording and independently transcribe and analyse the free-form discussion without reference to the canvas sheets; if the same four thematic questions fail to appear in the unaided conversation and only surface when participants are explicitly prompted by the canvas's six dimensions, the claim that these are community-derived questions collapses.","tokens_in":26673,"feed_emoji":"🤖","tokens_out":2738,"duration_ms":32455,"temperature":0.7,"pith_summary":"The paper tries to establish that community engagement with under-represented groups can surface the key ethical and socio-technical questions that wellbeing robots will raise once they leave the lab. Running structured reflection workshops with members of the public, women computer scientists, and humanities researchers, the authors distil participants' written and spoken reflections into four questions: Is the robot safe and how can we know that? Who is the robot built for and with? Who owns the robot and the data? Why a robot? The authors argue these four questions constitute a broad, reusable framework that roboticists can and should use during development and deployment to align robot design with public interest.","feed_headline":"Four questions for anyone building a wellbeing robot","feed_subtitle":"Community workshops turn six canvas prompts into a reusable ethics framework developers can adopt.","key_machinery":"The Social Robot Co-Design Canvas on Ethics, a reflection tool that prompts participants to consider six ethical dimensions (physical safety, data security, transparency, equality across users, emotional consideration, and behaviour enforcement), combined with inductive Thematic Analysis of the canvas text and researcher notes. The canvas supplies the structured elicitation; the thematic analysis turns the resulting qualitative data into the four thematic questions that form the proposed framework.","core_discovery":"The central discovery is the quartet of ethical and socio-technical questions that emerged from the authors' thematic analysis of workshop data. The questions are: (1) Is the robot safe and how can we know that?, covering physical safety, psychological attachment, and testing standards; (2) Who is the robot built for and with?, covering equality, culture, accessibility, and inclusivity; (3) Who owns the robot and the data?, covering privacy, protective policies, and data ownership; and (4) Why a robot?, covering the appropriateness of human-like robots, their placement in a wellbeing system, and their capabilities and limits. The paper presents these questions not as a definitive checklist b","pith_inferences":["The four questions effectively re-express classic AI ethics values—privacy, fairness, transparency, accountability—but reorganized from the standpoint of a prospective user rather than a system developer, which may make them easier to operationalize in design meetings.","A testable extension would be to run the same workshop protocol with different communities (for example, older adults, caregivers, or clinical practitioners) and see whether additional major questions emerge that do not fit under the four; this would probe the framework's completeness.","The paper leaves implicit that the four questions could also function as an audit scaffold: robot development teams could be evaluated on whether they can produce concrete, user-accessible answers to each question before deployment.","The 'Why a robot?' question carries a strong anti-solutionist undertone that, if taken seriously, could redirect research funding away from robotic wellbeing coaches and toward cheaper, more scalable non-embodied interventions—an implication the authors do not spell out."],"forward_implications":["If the framework is adopted, wellbeing robot developers should prepare documentation that answers the four questions and make it directly accessible to users, for example as an onboard FAQ.","Roboticists working beyond wellbeing, on any social robot application, could use the same four questions as a reflective guide because the paper explicitly extends its findings to other social robot contexts.","Engaging under-represented communities through structured workshops may become a standard anticipatory-ethics practice in robotic development, shifting evaluation earlier in the design process.","The questions highlight that psychological and emotional safety, not just physical safety, must be part of robot safety standards and certification.","The 'Why a robot?' question challenges developers to justify the choice of an embodied robot over cheaper or simpler alternatives, which could alter product design decisions."],"supporting_citations":[{"why":"Supplies the Social Robot Co-Design Canvas on Ethics, the data-collection instrument whose six prompted dimensions structured the workshops.","marker":"Axelsson et al. [2021b]"},{"why":"Provides the six-step Thematic Analysis method the authors followed to distil canvas text into the four thematic questions.","marker":"Clarke and Braun [2017]"},{"why":"Gives the prior '10 defining questions' for good robotics that inspired the authors' decision to frame findings as questions.","marker":"Šabanovi´c et al. [2023]"},{"why":"Establishes the prior design and ethical recommendations for robotic wellbeing coaches that this work extends toward community-engaged anticipatory ethics.","marker":"Axelsson et al. [2022]"},{"why":"Supports the authors' claim that their sample size of 22 reaches data saturation for thematic analysis.","marker":"Guest et al. [2006]"},{"why":"Provides the loose definition of community (shared interests, not necessarily shared locality) the paper adopts for selecting its three participant groups.","marker":"Bradshaw [2008]"}],"fun_headline_variants":["Four ethical questions for wellbeing robot builders","Community workshops yield four robot ethics questions","Wellbeing robots: four questions developers must answer","Four questions to ask before deploying a wellbeing robot","Ask these four questions before building a wellbeing robot"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The four themes genuinely emerge from what participants reflected on, rather than being predetermined by the six ethical topics that the canvas explicitly prompts.","fun_headline_variants_meta":{"raw":{"variants":["Four ethical questions for wellbeing robot builders","Community workshops yield four robot ethics questions","Wellbeing robots: four questions developers must answer","Four questions to ask before deploying a wellbeing robot","Ask these four questions before building a wellbeing robot"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000518,"raw_usage":{"total_tokens":2396,"prompt_tokens":844,"completion_tokens":1552,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":588,"completion_tokens_details":{"reasoning_tokens":1485}},"tokens_in":588,"tokens_out":1552,"duration_ms":12940,"temperature":1.0,"reasoning_tokens":1485,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T12:30:03.613618+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the workshops with full audio recording and independently transcribe and analyse the free-form discussion without reference to the canvas sheets; if the same four thematic questions fail to appear in the unaided conversation and only surface when participants are explicitly prompted by the canvas's six dimensions, the claim that these are community-derived questions collapses.","supporting_citations":[],"review_version":1}