REVIEW 3 major objections 2 minor
Beyond Technocratic XAI: The Who, What & How in Explanation Design
T0 review · 3 major / 2 minor · reviewed 2026-08-15 · deepseek-v4-flash
Pith's one-line read Three questions turn explainable AI into a design problem: Who needs the explanation, what needs explaining, and how should it be delivered.
desk verdict Plausible framing paper that needs to justify its triad; abstract-only review limits the verdict. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
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.
What would settle it
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.
Extended reading notes
Core claim
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.
Load-bearing premise
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.
Editorial extensions
If this is right
- 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.
Reading between the lines
- 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.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
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.
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 (3)
- [Abstract, para. 1] 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.
- [Abstract, para. 2] 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.
- [Abstract, last sentence] 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.
minor comments (2)
- [Abstract, para. 1] 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.
- [Abstract, para. 2] 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.
Circularity Check
No circularity: the abstract offers a normative proposal with no fitted inputs, equations, or prediction claims.
full rationale
This abstract-only manuscript proposes a three-part design framework (Who, What, How) and argues for ethical considerations in XAI. There is no derivation chain, no equation, no fitted parameter, and no empirical prediction that could reduce to an input. The triad is asserted rather than derived, and its exhaustiveness is not defended in the abstract, but that is a burden-of-proof or completeness concern, not circularity: asserting a proposal does not make the proposal the output of itself. The paper explicitly grounds the framework in prior research and design thinking, and no cited result is used as a load-bearing external theorem. Even though the full text is unavailable, nothing in the abstract supports a circularity finding. Per the review rules, an honest non-finding is appropriate, so the score is 0.
Assumptions & free parameters
assumptions (3)
- domain assumption Explanation in XAI is best understood as a situated design process.
- ad hoc to paper The Who, What, How triad is a useful and sufficient structure for explanation design.
- domain assumption Ethical considerations such as epistemic inequality, social inequity, and accountability are essential to explanation design.
Cite this review
Pith. "Pith review of Beyond Technocratic XAI: The Who, What & How in Explanation Design." pith.science (2026). https://pith.science/paper/OBHZZ7W6
@misc{pith2026250809231,
author = {Pith},
title = {Pith review of: Beyond Technocratic XAI: The Who, What & How in Explanation Design},
year = {2026},
howpublished = {\url{https://pith.science/paper/OBHZZ7W6}},
note = {Machine review of arXiv:2508.09231}
}
read the original abstract
The field of Explainable AI (XAI) offers a wide range of techniques for making complex models interpretable. Yet, in practice, generating meaningful explanations is a context-dependent task that requires intentional design choices to ensure accessibility and transparency. This paper reframes explanation as a situated design process -- an approach particularly relevant for practitioners involved in building and deploying explainable systems. Drawing on prior research and principles from design thinking, we propose a three-part framework for explanation design in XAI: asking Who needs the explanation, What they need explained, and How that explanation should be delivered. We also emphasize the need for ethical considerations, including risks of epistemic inequality, reinforcing social inequities, and obscuring accountability and governance. By treating explanation as a sociotechnical design process, this framework encourages a context-aware approach to XAI that supports effective communication and the development of ethically responsible explanations.
Reviewed August 15, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.