Preprint: Triad Interviews With Stroke Survivors Expose Three XAI Elicitation Biases
Researchers report a preprint, posted to arXiv on July 28, 2026 and not yet peer reviewed, on the methodological problem of eliciting explainable AI (XAI) requirements from stroke survivors. Existing protocols run dyadic interviews and overlook facilitation dynamics; this formative study moved to a survivor-caregiver triad, with three stroke survivors (two with moderate-to-severe aphasia) and three caregivers, and facilitators used four scaffolding techniques: analogical bridging, projective personas, binary forcing and extended response time. A reflexive analysis identified three systematic facilitation biases — normative bias, hypothesis confirmation bias and the presence effect — which the authors present as protocol risk guidelines for practitioners.
Why it matters
Explainability requirements gate the clinical uptake of AI-driven rehabilitation, and this preprint moves the failure point from the algorithm to the interview room — its dyad-to-triad redesign is a usable protocol risk guide, though with three survivors it is a hypothesis about elicitation, not evidence about stroke patients.
