{"id":"d3cc564b-e4d2-4a1f-a813-776347626b7d","arxiv_id":"2507.05187","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper introduces CDAVP, a socio-technical framework for letting communities define AI value profiles that users activate by context, bounded by democratically legitimated meta-rules.","lead":"A conceptual paper proposes replacing one-size-fits-all AI alignment with a framework in which communities define machine-readable value profiles and individual users choose which profiles apply. A smart generalist should read it to see a concrete alternative to centralized AI governance that puts value contestation at the center of design.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central claim rests on two unbuilt mechanisms: a formal value-profile language and a legitimate inter-user conflict-resolution procedure. Section 3.3 silently delegates the latter to 'the platform,' which could recreate the centralized value-setting CDAVP is meant to replace.","rationale":"The reader's verdict and reasoning are sound. The paper is a conceptual framework, not an implemented system, and it openly declares that formal languages, empirical studies, and psychological/social effects are future work (Section 8). This transparency is a strength, but it means the central claim is a research program, not a demonstrated result. The most load-bearing assumption is the one the reader flags: value profiles must be machine-readable and interpretable by AI, and conflicts must be moderated in a legitimate way. I would sharpen the reader's point by observing that Section 3.3's 'platform applies transparent, pre-defined conflict resolution strategies' is not merely an implementation detail—it is the point where CDAVP's distributed power could be silently re-centralized. Who defines the strategies? If the platform does, the framework retains the top-down value-setting it criticizes in Section 2. If communities do, then a shared space with conflicting profiles still requires a decision procedure that cannot be derived from any single community's values. This is not an internal inconsistency in the paper—it is an honest gap—but it is the hinge on which the strongest claim turns. The three scenarios do not test this hinge: Section 4.1.1 demonstrates only a single hard constraint, Section 4.2.1 reduces a community value to weighted priorities without defending that representation, and Section 4.2.2 leaves the inter-user conflict unresolved. Because the paper presents itself as a framework and the gaps are acknowledged, the appropriate verdict remains CONDITIONAL, not ACCEPT or REJECT. The reader's proposed condition—demonstrate a formal representation and legitimate conflict resolution—is the right one. My concrete test is deliberately minimal: write down the procedure for one shared-space conflict and show it does not hand one actor unilateral normative power. If that test passes, the strongest claim becomes plausible; if it fails, the claim collapses into metaphor.","tokens_in":8853,"tokens_out":4669,"duration_ms":56467,"concrete_test":"Pick the Section 4.2.2 shared-space example and specify a minimal formal conflict-resolution procedure: given two users' activated value profiles in a group chat, define the exact algorithm that decides which content is shown or blocked, and identify which actor (users, communities, platform, or meta-rule) chooses that algorithm. Then verify two properties: (1) the procedure terminates and satisfies the declared meta-rules; (2) if the platform chooses the algorithm, the result is not already a unilateral value imposition—i.e., the algorithm's choice is independently contestable by the affected users. If no such procedure can be written without giving one party unilateral power, the framework's pluralism collapses at the point of inter-user conflict.","verdict_should_be":"UNCHANGED","load_bearing_attack":"For the strongest claim to hold, communities must be able to encode values in profiles that AI systems interpret faithfully, and conflicts between profiles of different users in shared spaces must be resolved in a way that is itself pluralistic. Section 3.3 states only that 'the platform applies transparent, pre-defined conflict resolution strategies'; it does not say who defines those strategies, how they are legitimated, or how they avoid imposing one community's values on another. Without this, the system either requires a centralized authority at the exact point of disagreement—contradicting the paper's critique of top-down value definition—or it fails to manage the pluralism it promises. Section 8 explicitly defers formal languages and open standards, so the machine-readable profile pillar is also a promissory note. The scenarios in Section 4 do not close the gap: the dark-pattern rule in Section 4.1.1 is a single hard constraint, not a profile; the predictive-policing example reduces values to weighted priorities, a choice that is itself normative; and the content-moderation scenario leaves the shared-space conflict unresolved. The central claim is therefore conditional on feasibility assumptions the paper itself identifies as future work.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes Community-Defined AI Value Pluralism (CDAVP), a socio-technical framework in which self-organizing communities define explicit, machine-readable value profiles that users contextually activate, and AI applications interpret those profiles while a set of non-negotiable, democratically legitimated meta-rules bounds acceptable pluralism. It argues that this architecture, grounded in HCI concepts of contestability, infrastructuring, seamful design, and values levers, provides a necessary pathway to algorithmic accountability. Three application scenarios (autonomous UI design, predictive policing, content moderation) are used to illustrate the framework and differentiate it from existing approaches such as reactive contestation, participatory design, centralized oversight, and decentralized protocols.","tokens_in":9028,"tokens_out":4582,"duration_ms":51402,"significance":"The paper is a clearly written conceptual synthesis with a coherent internal argument and an unusually honest statement of its own limitations. Its strengths include the explicit grounding in established HCI/CSCW literature, the careful differentiation from prior work (e.g., reactive contestation, participatory design, collective constitutionalism), and the concrete acknowledgement in Section 8 that formal languages and empirical studies remain future work. As a position paper, it offers a useful vocabulary for discussing pluralistic AI governance and may seed future design research. However, its central claim is conditional on two unbuilt mechanisms: a formal value-profile representation that AI systems can interpret faithfully, and a legitimate method for resolving conflicts between profiles in shared spaces. The scenarios in Section 4 illustrate the idea but do not demonstrate that either mechanism is feasible. The paper should therefore be read as a research agenda rather than an implemented framework, and the conclusion should be tempered accordingly.","major_comments":[{"comment":"The framework's load-bearing premise is that value profiles are 'rich, machine-readable representations' (Section 3.1) that AI applications can 'transparently interpret' (Section 3.3), yet no formalism, schema, or prototype is presented, and Section 8 explicitly defers 'formal languages and open standards' to future work. Because the ex-ante contestation promise depends on profiles being faithfully interpretable by AI systems, the paper needs at least a minimal sketch of a profile language (e.g., how values, priorities, constraints, and rights/duties are represented and combined) or an explicit reframing of the contribution as a research program whose feasibility is unproven.","section":"3.1, 3.3, 8"},{"comment":"Section 3.3 states that 'the platform applies transparent, pre-defined conflict resolution strategies' for conflicts between profiles of different users, but it does not specify who defines those strategies, how they are legitimated, or how they avoid imposing one community's values on another. This reintroduces centralized value-setting at the exact point of disagreement, which is the problem CDAVP claims to solve. The paper should specify a governance mechanism for conflict moderation, for example democratic deliberation over meta-rules, federation, or user-level opt-in to predefined conflict-resolution procedures, and explain how it preserves pluralism.","section":"3.3"},{"comment":"In the predictive policing scenario, the institutional value profile is illustrated by 'Priority_Weight(Violent_Crime)=10, Priority_Weight(Minor_Infractions)=1.' This encoding is itself a normative choice, and the paper does not say who sets these weights, how they are audited, or how 'value-conformity' becomes a 'measurable and competitive criterion' without falling prey to the same contestation it criticizes in fairness metrics. The scenario needs either a concrete governance model for weight-setting or an explicit acknowledgment that the weights are placeholders for a political process that is outside the framework's current scope.","section":"4.2.1"},{"comment":"The content moderation scenario proposes a 'federal model' in which users activate community profiles for different contexts, but it does not explain how a shared space (e.g., a group chat) should handle users who activate conflicting profiles. This is precisely the inter-user conflict case left open in Section 3.3, and the scenario's claim that users 'curate their own moderation standards' is only meaningful for individual feeds, not for common spaces. The paper should either provide a worked example of pluralistic conflict resolution in a shared space or state clearly that the framework currently addresses only the individual-activation case.","section":"4.2.2"}],"minor_comments":[{"comment":"The framework's acronym appears inconsistently as 'CDA VP' (e.g., in the abstract and Section 1) and 'CDAVP'; please use a single spelling.","section":"Throughout"},{"comment":"Table 1 presents a 2x3 matrix with six cells, but only three scenarios are analyzed; the paper should state explicitly why the other cells (e.g., Pluralistic News Consumption, Human-AI Co-Creativity) are outside the scope.","section":"Table 1"},{"comment":"The dark-pattern rule 'The process to cancel a subscription must not require more steps than the sign-up process' is a single constraint, not a full value profile; the paper should clarify how such rules relate to the richer profile model described in Section 3.1.","section":"4.1.1"},{"comment":"Section 3.4 invokes 'Collective Constitutional AI' to motivate democratic legitimacy of meta-rules, but it does not discuss how such constitutional processes would scale to global AI platforms or how disagreement over the meta-rules themselves would be resolved.","section":"3.4"},{"comment":"The generative AI disclosure is a welcome transparency practice, though the statement could be shortened without losing its value to the reader.","section":"Statement on the Use of Generative AI"}],"recommendation":"major_revision","confidential_remarks":"This is a concept paper that fits a CSCW/FAccT-style venue, but the phrase 'necessary pathway' in the conclusion overstates what the evidence presented supports. The main revision challenge is to either close or explicitly embrace the representation and conflict-resolution gaps; without that, the submission reads more as a manifesto than as a framework with demonstrated load-bearing mechanisms."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear [Colleague],\n\nThis is a framework paper, and it should be read as one. It assembles contestability by design, values levers, infrastructuring, Collective Constitutional AI, and the Value Kaleidoscope into a single three-pillar architecture: community-defined machine-readable value profiles, user-controlled contextual activation, and system-level conflict moderation inside a frame of non-negotiable meta-rules. That integration is new, even though each brick is borrowed. The writing is clear, the citations are used appropriately, and the author is candid about what is missing.\n\nThe main soft spot is exactly where the stress-test note points. The central claim—that CDAVP allows communities and users to contest the values embedded in AI—depends on two mechanisms that are not built: a formal language for value profiles that an AI can interpret faithfully, and a conflict-resolution procedure between users' profiles in shared spaces that is itself pluralistic rather than centralized. Section 3.3 says the platform applies transparent, pre-defined strategies, but does not say who defines them or how they are legitimated. That is a potential re-creation of the top-down value-setting the paper argues against. The scenarios do not close the gap: the dark-pattern rule is a single constraint, the policing example reduces values to weighted priorities, and the content-moderation case leaves inter-user conflict unresolved. The paper honestly lists formal languages and empirical studies as future work, but that makes the verdict conditional rather than acceptance.\n\nI am not sure the stress-test is unfair; it lands. That said, for a framework paper in HCI/CSCW, this is a legitimate contribution. It gives the community a vocabulary and an architecture to argue with. The author acknowledges the risks (fragmentation, power asymmetries, cognitive load) and the generative-AI disclosure is a plus.\n\nI would send this to peer review. A serious referee would ask for a concrete example of a value profile schema or at least a clearer specification of the meta-rule legitimation process, but the paper is not incoherent and the idea is worth engaging. It is not a finished result; it is a research program. I would cite it if I were working on contestability or value-sensitive design.","headline":"A coherent and honest framework paper that synthesizes known ideas; its central promise depends on unbuilt machinery, but it deserves a serious referee.","tokens_in":9572,"tokens_out":1873,"would_cite":true,"duration_ms":21509,"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 argues that AI alignment should abandon the search for one universal value set and instead build infrastructure in which self-organizing communities define machine-readable value profiles, users activate them by context, and…","keywords":["Human-Computer Interaction","CSCW","Contestability","Infrastructuring","AI Governance","Value Pluralism","Value Profiles","Algorithmic Accountability"],"falsifier":"A practical falsifier is a controlled prototype test in which members of two culturally distinct communities encode the same scenario into value profiles and an AI must apply both under one meta-rule; if the profiles cannot be made mutually interpretable, or if the AI's output fails the communities' own judgments without developer rewrites, the framework's central representation-and-moderation claim collapses.","tokens_in":8605,"feed_emoji":"⚖️","tokens_out":8589,"duration_ms":96823,"temperature":0.7,"pith_summary":"The paper argues that the standard AI alignment target—a single, centrally defined set of values—cannot be legitimate for everyone and therefore cannot support meaningful contestability. It proposes Community-Defined AI Value Pluralism (CDAVP), a socio-technical framework that gives self-organizing communities the power to define explicit, machine-readable value profiles and gives individual users control over which profiles guide an AI in a given context. AI applications would interpret these profiles transparently and moderate conflicts between them, within a floor of non-negotiable, democratically legitimated meta-rules. If this works, contestability shifts from challenging single decisions after the fact to shaping the rules that generate decisions in the first place, making algorithmic accountability a design property rather than a promise.","feed_headline":"Let communities define the values AI follows","feed_subtitle":"No single aligned state: community profiles, user choice, and non-negotiable meta-rules make AI contestable.","key_machinery":"The central object is the community-defined value profile: a rich, machine-readable representation of a community's shared values, extending beyond preference lists to include rights and duties and drawing on established theories of basic human values and moral intuitions. The framework carries its argument through three pillars built on that object—community definition and forking, user-controlled contextual activation, and transparent conflict moderation—plus a non-negotiable meta-rule frame that sets the boundaries of acceptable pluralism. The profile does the load-bearing work: it makes values explicit, portable, auditable, and switchable, and it is what allows ex-ante contestation to happen.","core_discovery":"The paper's central claim is that contestability should be the organizing principle of AI systems, not a feature added to an otherwise aligned model. CDAVP distributes the definition of values across three layers: communities maintain rich machine-readable profiles that encode preferences, rights, and duties; users choose which profiles are active in each context, preserving agency across their multiple identities; and platforms apply transparent, pre-defined, privacy-preserving conflict moderation under a frame of universal meta-rules derived from human rights and democratic deliberation. The paper argues that this turns contestability by design into something proactive: users contest the rules rather than merely react to outputs, and institutions such as police departments become accountable by publishing standardized value profiles that can be audited and benchmarked before deployment.","pith_inferences":["Editorial extension: if value profiles become standardized and auditable, an independent conformity-assessment market could emerge, with third parties certifying that platforms apply declared community profiles as promised.","Editorial extension: the paper's own future-work list implies that the framework's viability hinges on a formal profile language; a natural next test is whether a small grammar of prioritized rules with weights and exceptions can express enough real-world community values to handle ordinary moderation conflicts.","Editorial extension: the user-controlled activation pillar suggests a testable empirical prediction—visible profile-switching controls will increase users' perceived agency and trust compared with fixed global moderation—which the paper does not itself test."],"forward_implications":["If CDAVP works, users can block manipulative dark patterns before they appear, because activated value profiles act as hard machine-readable constraints on generative interface designers.","Public institutions such as police departments could publish standardized value profiles, making their strategic priorities auditable and enabling value-conformity benchmarking before systems are procured.","Content moderation could split into a universal legal floor plus a user-curated gray zone, letting individuals apply different moderation standards in different contexts on the same platform.","Designers would shift from specifying final interfaces to building deliberation tools, conflict-moderation algorithms, and control interfaces, making ethics a property of the ecosystem rather than of a single artifact.","Algorithmic accountability becomes achievable because every AI decision can be traced to an explicit, human-defined value profile and a transparent meta-rule frame."],"supporting_citations":[{"why":"Defines contestability by design, the baseline concept that CDAVP extends from reactive decision-challenge to proactive rule-shaping.","marker":"[1]"},{"why":"Supplies the idea that AI-facing value representations should include pluralistic values, rights, and duties, not just preferences.","marker":"[38]"},{"why":"Provides the model of democratically produced constitutional rules that CDAVP relies on to legitimize its non-negotiable meta-rules.","marker":"[19]"},{"why":"Gives the one-time participatory rule-definition baseline that CDAVP positions as needing ongoing dynamic infrastructure.","marker":"[22]"},{"why":"Defines algorithmic accountability, the outcome the framework claims to make achievable.","marker":"[12]"},{"why":"Grounds the psychological breadth of community value profiles in a theory of basic human values.","marker":"[35]"},{"why":"Grounds the moral-intuition components of profiles such as fairness, loyalty, and sanctity.","marker":"[18]"},{"why":"Introduces the values-levers concept that CDAVP reorients toward designing participatory mechanisms rather than final outputs.","marker":"[36]"},{"why":"Provides the infrastructuring perspective that frames CDAVP as open, evolving, participatory infrastructure rather than a finished product.","marker":"[39]"},{"why":"Motivates forking as an evolutionary mechanism that lets subgroups contest a community's value profile by splitting off.","marker":"[33]"}],"fun_headline_variants":["AI values from communities, not corporations","Contestable AI: users pick which values apply","Value pluralism as infrastructure for AI","Community-defined values: a roadmap for contestable AI"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that community values can be written as explicit machine-readable profiles that AI applications can interpret faithfully enough to govern behavior and to moderate conflicts between users; the paper itself assigns the formal profile language to future work (Section 8), so this premise is currently asserted rather than demonstrated.","fun_headline_variants_meta":{"raw":{"variants":["AI values from communities, not corporations","Contestable AI: users pick which values apply","Value pluralism as infrastructure for AI","Community-defined values: a roadmap for contestable AI"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000226,"raw_usage":{"total_tokens":1473,"prompt_tokens":956,"completion_tokens":517,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":572,"completion_tokens_details":{"reasoning_tokens":460}},"tokens_in":572,"tokens_out":517,"duration_ms":6225,"temperature":1.0,"reasoning_tokens":460,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T19:30:14.230795+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A practical falsifier is a controlled prototype test in which members of two culturally distinct communities encode the same scenario into value profiles and an AI must apply both under one meta-rule; if the profiles cannot be made mutually interpretable, or if the AI's output fails the communities' own judgments without developer rewrites, the framework's central representation-and-moderation claim collapses.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Grounds the moral-intuition components of profiles such as fairness, loyalty, and sanctity."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the values-levers concept that CDAVP reorients toward designing participatory mechanisms rather than final outputs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates forking as an evolutionary mechanism that lets subgroups contest a community's value profile by splitting off."}],"review_version":1}