{"id":"0cc6e4b3-29fd-418d-b116-9ffdf6164d63","arxiv_id":"1908.01024","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"This paper argues that a fairness-focused approach to AI ethics for disabled people is insufficient and should be replaced by a justice framework centered on structural oppression.","lead":"This paper argues that treating fairness as the goal of AI for disabled people can reinforce the very power structures that harm them. It uses two computer vision case studies to call for a shift from fairness to justice.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The fairness-to-justice case hinges on harms persisting under fair deployment; in both case studies that persistence is asserted via critical theory, not demonstrated.","rationale":"I agree with the reader's identification of the load-bearing assumption: the case studies need to show that the harms would persist even if standard fairness constraints were satisfied. My proposed test makes that assumption explicit and checkable. The paper is a conceptual position piece, so requiring deployment data would be unreasonable; however, the internal logic can and should be tested by asking whether the named harms are functions of outcome distributions or of the technology's social role. If the harms are invariant to fairness metrics, the central claim is supported; if not, the distinction between fairness and justice is overstated. I do not see a more serious concern: the undefined 'justice' is a completeness problem rather than a threat to the core critique, and the citation numbering errors in the reference list are mechanical and correctable. The conditional verdict remains appropriate because the paper's conclusion is plausible but depends on this untested persistence claim.","tokens_in":6707,"tokens_out":5396,"duration_ms":63013,"concrete_test":"Analytical test: For each case study, write the strongest fairness remedy the paper's own framing would permit (e.g., requiring equalized odds for diagnostic outcomes across race/gender/class in the autism case; requiring object-recognition accuracy parity across socioeconomic contexts in the sight case). Then determine whether the specific harm named—gatekeeping, epistemic transfer, or surveillance normalization—is expressible as a violation of that constraint. If the harm is not expressible as such a violation, the paper's claim that fairness cannot address it is internally supported; if it is expressible, the case for abandoning fairness for justice is weakened. This isolates the assumption without requiring unavailable deployment data.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's argument requires that the harms in the two case studies survive any fairness-based remedy, not merely that they are not currently addressed by fairness metrics. In 'AI For Diagnosis,' the named harm is not unequal diagnostic accuracy across groups but the reinforcement of psychiatric gatekeeping: the paper claims that even a dataset-diversified, demographically fair classifier would still add technical authority to the medical model. In 'AI For Sight,' the paper explicitly says 'computer vision, even deployed fairly, cements vision as a superior sense and legitimizes surveillance.' These are claims about the social meaning and power effects of the technologies, not about their outcome distributions. They are supported by cited critical literature, but the paper does not demonstrate that a fairness intervention—e.g., equalized odds across disability, race, and class, or participatory dataset audits—would fail to mitigate or reframe those harms. The central shift from fairness to justice is load-bearing on this persistence claim; if a robust fairness process could surface and address the same concerns, the dichotomy collapses into a call for better fairness.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper argues that the dominant 'fairness' framing in AI ethics—understood as statistical outcome parity and dataset representativeness—is insufficient for addressing the harms of AI for disabled people, because it ignores structural oppression, epistemic hierarchies, and the social meaning of technologies. Through two computer-vision case studies (automated autism diagnosis and sight-assistance tools), the authors claim that even fairness-improved systems would still reinforce medical gatekeeping, shift judgment away from disabled users, and legitimize surveillance. They conclude by urging a move 'towards notions of justice,' drawing on disability studies and critical data science.","tokens_in":6855,"tokens_out":4611,"duration_ms":50785,"significance":"If the argument holds, the paper makes a timely and important contribution by bringing disability studies to bear on mainstream AI ethics and by challenging the adequacy of fairness metrics as the primary ethical lens. Its strengths include a clear presentation of two concrete case studies, explicit engagement with foundational critical literature (Hoffmann, Conrad, Waltz, Mankoff et al.), and a pointed call to center multiply marginalized disabled people in AI design. The paper is coherent and argued in good faith; the case studies effectively illustrate harms (medical gatekeeping, epistemic transfer, surveillance normalization) that are often invisible in metric-driven fairness discussions. For a position paper in this area, the conceptual contribution is valuable and would help reorient assistive-technology research and AI ethics conversations.","major_comments":[{"comment":"The central shift from fairness to justice rests on the claim that the identified harms would persist even under fair deployment—e.g., 'computer vision, even deployed fairly, cements vision as a superior sense and legitimizes surveillance'—and that the 'issue is not one of discrimination against the patient for being autistic but for being a patient.' This persistence claim is load-bearing: if a robust fairness process (e.g., equalized odds across intersectional groups, participatory dataset audits, or co-design with disabled stakeholders) could surface and mitigate the same concerns, the fairness-to-justice dichotomy would collapse into a call for better fairness. The paper asserts this persistence through cited critical theory rather than analyzing why specific fairness remedies would fail. I recommend that the authors either engage with the broader fairness literature—including processual and structural approaches—or qualify the claim to 'metric-based fairness as typically practiced in ML,' which would make the argument more defensible.","section":"Case Studies, 'AI For Diagnosis' and 'AI For Sight'"},{"comment":"The paper calls for a reframing 'towards notions of justice,' but the concept of justice is never defined or operationalized. The reader is not told what a justice-oriented design or evaluation would look like in either case study, beyond vague references to 'power' and 'wider social context.' Since the paper's title and central recommendation hinge on this dichotomy, the positive framework is underspecified. A concrete sketch—e.g., drawing on disability-justice principles, or showing how the two case studies would be redesigned under a justice lens—would make the contribution far more actionable and would allow readers to assess whether the proposed shift is genuinely distinct from a more robust fairness framework.","section":"Conclusion"},{"comment":"The paper scopes fairness narrowly to statistical outcome parity and data representativeness, following Trewin and Hoffmann. This is a defensible reading of the ML fairness literature, but it risks setting up a strawman: several fairness frameworks in the broader literature address structural concerns, such as fairness as non-domination or fairness as accountability, and are not reducible to parity metrics. The paper would be strengthened by acknowledging these positions and explaining why they too would fail to address the case-study harms, or by explicitly confining the critique to a labeled subset (e.g., 'metric-based individual fairness'). Without this, the dichotomy between fairness and justice appears sharper than it actually is.","section":"Fairness: Defined and Contested"}],"minor_comments":[{"comment":"The citation numbering is seriously inconsistent: for example, Trewin's 'AI Fairness for People with Disabilities' is cited as [27] in the introduction but appears as [29] in the Fairness section; Mitzi Waltz's 'Autism=death' is cited as [35] in the text, but the reference list has it as [32]; and the World Institute on Disability is cited as [32] in the introduction but [34] later in the text. These mismatches make the bibliography difficult to verify and must be corrected.","section":"Throughout"},{"comment":"'we would be remis to deny' should be 'remiss'.","section":"Introduction"},{"comment":"The claim that assistive seeing tools 'legitimize surveillance' is supported by a plausible mechanism (the AT Effect, the savior narrative), but the sentence 'who's to say blind people aren't among the users of policing technologies' is speculative and could be strengthened with an example or a reference showing actual integration or misuse of similar technologies.","section":"AI For 'Sight', Concerns Raised Through a Justice Lens"},{"comment":"A large copyright/license boilerplate appears in the body of the manuscript; this should be removed from the submission version. Heading styles are also inconsistent (e.g., 'Case Studies of AI in Assistive Technology' appears in a different font than other headings).","section":"General formatting"}],"recommendation":"major_revision","confidential_remarks":"The paper is a position piece that would benefit from more sustained engagement with counterarguments, particularly whether broader fairness frameworks could address the identified harms. The citation-number errors are extensive and suggest the manuscript needs a careful technical pass before publication. The paper's scope is limited to computer vision, but the authors do acknowledge this; the conclusion generalizes somewhat beyond the evidence presented, which should be tightened. Overall, the core argument is defensible and the topic is well suited to the journal, but the load-bearing persistence claim and the undefined notion of justice require substantive revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper does something genuinely useful: it takes Anna Hoffmann's fairness-to-justice critique and shows what it means for two computer vision applications aimed at disabled people. The autism diagnosis case and the 'AI for sight' case are well chosen, and the paper is right that a fairness lens alone would miss the structural harms - diagnostic gatekeeping, medicalization, the cementing of vision as superior, and surveillance normalization. The writing is clear and the argument is easy to follow.\n\nWhat's actually new is the application, not the framework. The authors draw heavily on Hoffmann and disability studies, and they acknowledge that. That's fine: the contribution is in the concrete illustration, and they do it effectively. I also appreciated that they take the fairness position seriously before critiquing it, rather than setting up a strawman.\n\nThe main soft spot is the one the stress-test flags. The paper asserts that the harms would persist even under fair deployment - e.g., 'computer vision, even deployed fairly, cements vision as a superior sense and legitimizes surveillance.' That is load-bearing: if a robust fairness process could surface and address these power concerns, the dichotomy between fairness and justice would collapse into a demand for more thorough fairness. The authors don't demonstrate that, and they also don't engage with procedural or participatory fairness, which might be able to raise the same questions. However, read charitably, the paper's claim is conceptual: fairness metrics are defined over outcome distributions, so by construction they cannot question the social meaning of a technology. That point is valid, but the paper should say it explicitly and defend it. Right now it feels asserted rather than argued.\n\nThe other weaknesses are minor. The justice framework is left vague - the paper says 'move towards justice' but doesn't say what that looks like in practice. That limits its usefulness for designers. And the reference list has clear numbering errors (Trewin is cited as [27] in the text, but the entry numbered 27 is Thevenot et al.). These are easy fixes.\n\nOverall, this is a solid position paper. The central argument holds up as a critique of a narrow fairness framing; it would be a good contribution to an AI ethics or accessible computing venue. It deserves serious peer review. I'd suggest asking the authors to (1) clarify the sense in which harms persist under fair deployment, (2) engage with at least one attempt to broaden fairness, and (3) fix the references.","headline":"A useful position paper that concretely applies the fairness-to-justice critique to disability and computer vision, with a few soft spots but a sound core.","tokens_in":676,"tokens_out":909,"would_cite":true,"duration_ms":27976,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that a fairness framing of AI ethics for disabled people can reinforce medical gatekeeping and surveillance, and that the field should instead center justice.","keywords":["AI fairness","disability","assistive technology","computer vision","justice","medicalization","surveillance","algorithmic ethics"],"falsifier":"If a longitudinal deployment study of a fairness-adjusted autism-diagnosis aid showed that multiply marginalized children gained access to self-directed support without diagnostic over-reliance, and that sight-assist users overrode automated interpretations in practice with no surveillance spillover, the paper's central claim would be weakened. The study would need to observe actual use, not lab accuracy.","tokens_in":6510,"feed_emoji":"♿","tokens_out":4711,"duration_ms":47190,"temperature":0.7,"pith_summary":"The paper argues that evaluating AI for disabled people through a single measurement of fairness—does the same case get the same outcome—can itself reinforce oppression. Two computer-vision case studies, automated autism diagnosis and sight-assistance tools, are used to show that even a fair algorithm can reinforce medical gatekeeping or shift judgment away from the user. The paper's positive claim is that AI ethics for disability should be reframed around justice, which asks who holds power, which structures are upheld, and who is left out. A sympathetic reader should care because this shifts the goal of assistive AI from fairness in inputs and outputs to redistribution of power and structural change.","feed_headline":"Fairness alone cannot make AI just for disabled people","feed_subtitle":"Two computer-vision cases show fair algorithms that still reinforce medical gatekeeping and surveillance; the paper shifts the goal to…","key_machinery":"The mechanism carrying the argument is the paired case-study comparison of two ethical lenses. Each technology is first seen through a fairness lens—what biases in data or outcomes could be corrected—and then through a justice lens—who gains diagnostic or interpretive authority, what social structures are legitimized, and which disabled people are served or harmed. The fairness-justice distinction, drawn from prior critiques that fairness is modeled on anti-discrimination law and leaves structural oppression untouched, supplies the category that lets the paper show how a technology can be fair yet unjust.","core_discovery":"The central claim is that 'fairness' as usually operationalized in AI ethics—equal treatment for similar cases separated only by disability status—is not merely incomplete but can be harmful when applied to disabled people. In the diagnosis case, a fairness fix such as diversifying training data would leave untouched the question of why diagnosis is the only legitimate route to autistic existence, strengthening the gatekeeping authority of psychiatrists and now algorithms. In the sight-assistance case, even a vision system that treats all users equally can transfer judgment from the user to black-boxed technology and normalize surveillance. The paper concludes that researchers should move beyond fairness and center justice, understood as attention to structural oppression, power, and multiply marginalized disabled people.","pith_inferences":["A testable extension: if the argument is right, deployed assistive AI optimized only for fairness will show harms that fairness audits miss—such as diagnostic over-reliance or user deference—in longitudinal user studies, while participatory designs that give users veto power over automated interpretations should not.","The fairness-to-justice shift implies that ethical review of assistive AI should include governance models, not just performance metrics; this could be operationalized as requiring documented consent and user veto over automated outputs.","Neighbouring domains—employment screening, predictive policing, and health-care allocation—face the same limitation the paper names for disability, so the argument generalizes to any group whose marginalization is structural rather than merely statistical.","An implicit consequence is that 'fairness' in AI may keep its use for narrow auditing, while 'justice' becomes the standard for whether a technology should exist at all."],"forward_implications":["Fairness metrics such as equalized odds or dataset diversification will not by themselves prevent the harms described; a fair autism-diagnosis system can still reinforce medical gatekeeping, and a fair sight-assist system can still defer judgment to black-boxed technology.","Researchers and designers evaluating assistive AI must ask who holds power over diagnosis and interpretation, not only whether outcomes are equal.","Design and evaluation should center multiply marginalized disabled people, since benefits to otherwise-privileged disabled people can coincide with harm to those marginalized in multiple ways.","AI ethics conversations around disability should draw on disability studies and treat technology as part of the construction of disability rather than a neutral add-on.","Justice-oriented design would question the premises of the technology itself, such as whether early diagnosis is a good outcome or whether vision must be the privileged sense."],"supporting_citations":[{"why":"Supplies the four-limitation critique of fairness and the argument that fairness is modeled on anti-discrimination law, leaving structural oppression untouched.","marker":"[12]"},{"why":"Establishes the baseline framing of AI fairness for disabled people that the paper critiques, defining fairness as similar outcomes across similar cases.","marker":"[29]"},{"why":"Supplies the concept of medicalization and social control used to argue that diagnostic tools strengthen gatekeeping authority.","marker":"[7]"},{"why":"Documents that the diagnostic systems used as baselines have little validity, undermining the premise of automated diagnosis.","marker":"[28]"},{"why":"Supplies the link between diagnostic labels and harmful treatment of autistic people, supporting the claim that a diagnosis can itself be a harm.","marker":"[32]"},{"why":"Provides evidence that users defer to computer-generated captions, supporting the epistemic-transfer concern in sight-assistance tools.","marker":"[15]"},{"why":"Demonstrates that object-recognition systems fail across non-Western and lower-income contexts, grounding fairness and justice concerns for sight-assist tools.","marker":"[8]"},{"why":"Documents bias and misuse in surveillance-oriented computer vision, supporting the concern that sight-assist tools can normalize surveillance.","marker":"[13]"}],"fun_headline_variants":["Fairness in AI can reinforce ableism—justice is the answer","Two vision AI cases show why fairness isn't enough for disability","Beyond fairness: AI ethics for disabled people needs justice","Why fair algorithms can still harm disabled communities","Fair AI isn't just AI: the case for disability justice"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument depends on the premise that the harms it describes—medical gatekeeping from automated autism diagnosis and epistemic transfer plus surveillance normalization in sight-assistance tools—would actually arise and would not be prevented by any fairness-based remedy.","fun_headline_variants_meta":{"raw":{"variants":["Fairness in AI can reinforce ableism—justice is the answer","Two vision AI cases show why fairness isn't enough for disability","Beyond fairness: AI ethics for disabled people needs justice","Why fair algorithms can still harm disabled communities","Fair AI isn't just AI: the case for disability justice"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00092,"raw_usage":{"total_tokens":3899,"prompt_tokens":849,"completion_tokens":3050,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":465,"completion_tokens_details":{"reasoning_tokens":2969}},"tokens_in":465,"tokens_out":3050,"duration_ms":21824,"temperature":1.0,"reasoning_tokens":2969,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:25:18.771515+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"If a longitudinal deployment study of a fairness-adjusted autism-diagnosis aid showed that multiply marginalized children gained access to self-directed support without diagnostic over-reliance, and that sight-assist users overrode automated interpretations in practice with no surveillance spillover, the paper's central claim would be weakened. The study would need to observe actual use, not lab accuracy.","supporting_citations":[],"review_version":1}