{"id":"76c5092f-d824-4a3e-8e63-4f96dcce7303","arxiv_id":"2508.09231","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":3.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper frames explanation in XAI as a situated design process organized around Who, What, and How, with explicit ethical safeguards.","lead":"This paper proposes that explaining AI decisions should be treated as a design process, with three questions guiding every explanation: Who needs it, What needs explaining, and How it should be delivered. It argues that current XAI focuses too much on tools and not enough on the context and ethics of communication.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Triad exhaustiveness is asserted rather than argued; without full text, the framework's normative force remains unsubstantiated.","rationale":"The reader's weakest assumption was that the triad may not be exhaustive or sufficiently justified. I agree that this is the central load-bearing concern. The abstract gives no evidence for the triad's completeness, and the paper's contribution is conceptual rather than empirical, so the only way to assess it is to examine the full argument. Since the full text is unavailable, the appropriate verdict remains UNVERDICTED. My concern does not change the verdict because the paper is already unverdictable from the abstract alone; however, if the full text merely asserts the triad without a systematic derivation or comparison against alternative dimensions, the framework should not be accepted as a robust design method. The concrete test I propose would settle whether the triad actually has the claimed organizing power, and the paper's failure to address such a test would raise its correctness risk.","tokens_in":678,"tokens_out":2607,"duration_ms":30925,"concrete_test":"Apply the Who/What/How framework to a set of explanation needs drawn from the XAI literature (e.g., decision justification, model debugging, regulatory audit, fairness complaint). For each need, check whether the triad captures all design-relevant distinctions or whether a separate temporal or contextual dimension must be introduced. If any need requires a dimension orthogonal to the triad, the framework is incomplete; if the paper itself never considers such an adversarial test, the centrality of the triad remains unproven.","verdict_should_be":"UNVERDICTED","load_bearing_attack":"The central claim is that explanation design in XAI should be organized around three questions: Who, What, and How. For this framework to be load-bearing, the triad must be non-redundant, sufficient to produce actionable design guidance, and not arbitrarily omit other equally important dimensions. The abstract offers no derivation of these three questions; it lists them as a proposal and then asserts their relevance. If the full paper does not justify why exactly these three categories are the right decomposition, the framework risks being an arbitrary re-labeling of known XAI concerns. In particular, the paper claims ethical considerations are essential, but the abstract gives no indication how ethics is integrated into Who, What, or How; if the ethics discussion is appended as a separate admonition, the central claim that the triad structures the design process is weakened. The weakest link is therefore the completeness and distinctness of the triad, with the possibility that dimensions such as 'when' (timing) or 'where' (deployment context) are equally necessary for situated explanation design.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a conceptual framework for explanation design in explainable AI (XAI), organized around three questions: Who needs the explanation, What needs explaining, and How the explanation should be delivered. It argues that explanation should be understood as a situated sociotechnical design process, and that ethical considerations—such as epistemic inequality, social inequity, and accountability—should be central to this process. The present review is based solely on the abstract, as the full text was not provided.","tokens_in":980,"tokens_out":2231,"duration_ms":24951,"significance":"If the framework is developed with adequate argumentative support, it could provide a useful structuring device for practitioners and researchers in XAI, redirecting attention from purely technical interpretability toward audience, context, and responsibility. The emphasis on ethics and accountability is timely and reflects a growing concern in the field. However, because the available material is limited to the abstract, the substantive significance of the framework cannot yet be assessed, and its contribution relative to existing XAI taxonomies and design frameworks remains unclear.","major_comments":[{"comment":"The central organizing claim is that the triad of Who, What, and How is sufficient to structure explanation design, yet the abstract states this as a proposal without supporting argument. The completeness and distinctness of these three categories is load-bearing: if other dimensions, such as timing (when an explanation is needed) or deployment context (where the explanation is used), are equally important, then the framework would be incomplete. The full text must provide a principled derivation of the triad or an explicit argument for why other dimensions are subordinate.","section":"Abstract, para. 1"},{"comment":"The ethical considerations are presented as an additional emphasis rather than as consequences of the triad. If ethics is not integrated into the Who, What, and How questions—for example, through an account of whose interests shape each answer—then the central claim that the triad structures a responsible design process is weakened. The full text should show how the ethical criteria flow from or transform the three questions, rather than appearing as a separate admonition.","section":"Abstract, para. 2"},{"comment":"The abstract claims that the framework 'supports effective communication,' which is an empirical outcome, but no evidence, case study, or evaluative protocol is indicated. As a conceptual proposal this may be acceptable, but the paper should specify what would count as evidence for the framework's effectiveness, especially given the framework's prescriptive role in guiding practitioners.","section":"Abstract, last sentence"}],"minor_comments":[{"comment":"The phrase 'situated design process' is central to the argument but is not defined; consider adding a one-sentence clarification for readers unfamiliar with design-thinking terminology.","section":"Abstract, para. 1"},{"comment":"The list of ethical concerns mixes outcomes and values ('epistemic inequality, reinforcing social inequities, and obscuring accountability and governance'); aligning the grammatical structure of the list would improve clarity.","section":"Abstract, para. 2"}],"recommendation":"uncertain","confidential_remarks":"This review is based solely on the abstract because no full text was provided. If the full manuscript is available, I would need to examine how the triad is derived, how the ethical analysis is operationalized, and whether the framework is compared with existing XAI taxonomies. The current abstract gives no basis for a verdict beyond 'uncertain.' I would also recommend checking that the paper engages with prior work on explanation as communication and with established accounts of accountability in sociotechnical systems."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe abstract describes a sensible but not startling reframing: explanation design in XAI should ask who needs the explanation, what they need explained, and how to deliver it. The paper seems to be a position piece, not an empirical study, so the bar is whether the framing earns its place. What I can say from the abstract: the emphasis on situated design and on ethics (epistemic inequality, accountability) is welcome and aligns with current thinking in XAI and HCI. The three questions are familiar from communication theory and design thinking; the novelty is applying them squarely to XAI and foregrounding sociotechnical consequences. That is a reasonable contribution for a workshop or conference paper.\n\nThe soft spot, without the full text, is the exhaustiveness and non-redundancy of the triad. The abstract lists the three questions but doesn't justify why these three, or how they generate actionable guidance. 'When' and 'where' could matter as much as 'how' in a situated design process. If the paper just asserts the triad and then appends ethics as a separate admonition, the central claim weakens. I also note no empirical validation is offered in the abstract, which is fine for a conceptual paper only if the argument is rigorous enough to stand on its own.\n\nThe biggest uncertainty is structural: the review is based on the abstract alone. I can't assess whether the full text actually defends the triad or engages with alternative frameworks. If it does, this deserves a serious referee. If it doesn't, it's a well-intentioned but thin re-labeling. My default is to give the paper the benefit of the doubt, because the questions it asks are real and the audience (XAI practitioners) could benefit from a common language.\n\nFor a peer-review decision: I would accept it for review, conditional on the full argument being available. A good referee should push on triad completeness, on how ethics is operationalized, and on what practical difference the framework makes compared with existing audience-centered XAI work. If the answers are in the paper, this is a solid contribution. If not, a major revision is needed.\n\nBring it to a reading group? Maybe, if someone wants to debate whether the triad is complete. I wouldn't cite it yet.\n\nBest.","headline":"Plausible framing paper that needs to justify its triad; abstract-only review limits the verdict.","tokens_in":1348,"tokens_out":1827,"would_cite":false,"duration_ms":19113,"reading_group":"maybe","serious_thinker":"unclear","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Three questions turn explainable AI into a design problem: Who needs the explanation, what needs explaining, and how should it be delivered.","keywords":["explainable AI","explanation design","situated design process","Who-What-How framework","sociotechnical systems","ethics of AI","transparency","accountability"],"falsifier":"A user study in which explanation success varies with a factor that cannot be placed under Who, What, or How—for example, if the same explanation delivered at different times relative to a decision consistently changes recipient comprehension or trust—would show that the triad is not exhaustive.","tokens_in":499,"feed_emoji":"🧭","tokens_out":2164,"duration_ms":25105,"temperature":0.7,"pith_summary":"This paper argues that the main weakness of explainable AI is not a shortage of techniques but a shortage of design thinking about the recipient. It proposes a normative framework that treats explanation as a situated design process structured by three questions: Who needs the explanation, What they need explained, and How that explanation should be delivered. The authors claim that ethical considerations—epistemic inequality, reinforcing social inequities, and obscuring accountability—must be built into explanation design, not added afterward. For practitioners, this means explanation choices should follow from audience and context rather than from the latest interpretability algorithm.","feed_headline":"Ask who, what, and how before explaining AI","feed_subtitle":"A three-part design framework makes AI explanations context-aware and ethically responsible.","key_machinery":"The central object is the Who-What-How triad, a three-question design heuristic for explanation. 'Who' identifies the recipient's knowledge, role, and needs; 'What' specifies the target of explanation, such as a prediction, a feature, or a model boundary; 'How' covers the format, medium, and level of detail. The triad carries the argument by converting explanation from a one-size-fits-all technique into a set of design decisions, and by forcing ethical considerations to be addressed at each step.","core_discovery":"The central claim is that explanation in XAI should be reframed as a sociotechnical design process rather than a technical output problem. The paper's contribution is a three-part framework: asking Who the explanation is for, What part of the model's behavior needs explaining, and How the explanation is communicated. The authors argue that these three questions, informed by design-thinking principles, make explanation design context-aware and ethically responsible. They further claim that neglecting these questions produces explanations that can create epistemic inequality, perpetuate social inequities, and obscure accountability and governance.","pith_inferences":["The Who-What-How triad may be extendable with other dimensions, such as When an explanation is given or Where it appears; a recipient's needs could depend on timing and context in ways the triad does not explicitly name.","The framework could be operationalized as a design checklist or interview protocol, and its value tested by comparing user comprehension and trust when explanations are chosen via the triad versus chosen by interpretability technique alone.","If the triad becomes standard practice, evaluation benchmarks for XAI might shift from model-centric metrics to audience-specific measures of understanding, which would be a larger methodological change than the paper states explicitly.","The emphasis on situated design implies that no single explanation is correct for all audiences, so explanations may need to be adapted per stakeholder group in deployed systems."],"forward_implications":["Practitioners who adopt the framework will justify explanation choices by audience and context rather than by technical convenience or model fidelity.","The quality of an explanation will be judged by whether it meets the recipient's needs, not only by how accurately it reflects the model.","Ethical risks such as epistemic inequality, social inequity, and obscured accountability become explicit inputs to explanation design rather than afterthoughts.","Transparency is reframed as a communication outcome achieved through design, not as a property that a model either has or lacks.","The framework gives a practical structure for building and deploying explainable systems in real-world settings with diverse stakeholders."],"supporting_citations":[],"fun_headline_variants":["Who, what, how: The 3 questions for ethical AI explanations","Don't explain AI until you answer who, what, and how","Ethical AI explanations come from design thinking, not tech","Reframe AI explanations as a sociotechnical design task"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The framework assumes that the needs of explanation recipients can be meaningfully captured by the three categories Who, What, and How, so that any equally important dimension left out—such as timing or physical context—would make the framework incomplete and possibly misleading.","fun_headline_variants_meta":{"raw":{"variants":["Who, what, how: The 3 questions for ethical AI explanations","Don't explain AI until you answer who, what, and how","Ethical AI explanations come from design thinking, not tech","Reframe AI explanations as a sociotechnical design task"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000614,"raw_usage":{"total_tokens":2774,"prompt_tokens":787,"completion_tokens":1987,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":403,"completion_tokens_details":{"reasoning_tokens":1916}},"tokens_in":403,"tokens_out":1987,"duration_ms":14195,"temperature":1.0,"reasoning_tokens":1916,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:31:41.598462+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A user study in which explanation success varies with a factor that cannot be placed under Who, What, or How—for example, if the same explanation delivered at different times relative to a decision consistently changes recipient comprehension or trust—would show that the triad is not exhaustive.","supporting_citations":[],"review_version":1}