Now You See Me: Designing Responsible AI Dashboards for Early-Stage Health Innovation

TL;DR

This work contributes actionable guidance for designing Responsible AI governance dashboards that support decision-making and accountability in early-stage health innovation, and suggests that ecosystem-level coordination can enable more scalable and diverse AI innovation in healthcare.

Abstract

Innovative HealthTech teams develop Artificial Intelligence (AI) systems in contexts where ethical expectations and organizational priorities must be balanced under severe resource constraints. While Responsible AI practices are expected to guide the design and evaluation of such systems, they frequently remain abstract or poorly aligned with the operational realities of early-stage innovation. At the ecosystem level, this misalignment disproportionately affects disadvantaged projects and founders, therefore limiting the diversity of problem-areas under consideration, solutions, stakeholder perspectives, and population datasets represented in AI-enabled healthcare systems. Visualization provides a practical mechanism for supporting decision-making across the AI lifecycle. When developed via a rigorous and collaborative design process, structured on domain knowledge and designed around real-world constraints, visual interfaces can operate as effective sociotechnical governance artifacts enabling responsible decision-making. Grounded in innovation-oriented Human-Centered Computing methodologies, we synthesize insights from a series of design studies conducted via a longitudinal visualization research program, a case study centered on governance dashboard design in a translational setting, and a survey of a cohort of early-stage HealthTech startups. Based on these findings, we articulate design process implications for governance-oriented visualization systems: co-creation with stakeholders, alignment with organizational maturity and context, and support for heterogeneous roles and tasks among others. This work contributes actionable guidance for designing Responsible AI governance dashboards that support decision-making and accountability in early-stage health innovation, and suggests that ecosystem-level coordination can enable more scalable and diverse AI innovation in healthcare.

Figures (6)

• Figure 1: Overview of the two-stage visualization design-study approach: co-created industry characterization (Stage 1) produces reusable domain artifacts that surface stage-specific obligations, evidence dependencies, stakeholder responsibilities etc., which then define and constrain dashboard design (what is shown, how it is encoded, and how roles interact) (Stage 2) and support context- and task-specific views for users-decision-makers.
• Figure 2: Project timeliness and scope: gray denotes past activities, while blue shades cover the base study learnings reported here pertinent to the design of RAI dashboards. First, learnings are synthesized from real-world projects and informed the Case-Study, and both subsequently led to the formulation of Survey questions.
• Figure 3: Projects span European jurisdictions with different AI regulatory regimes (incl. the UK, EU, EEA, Ukraine), with 43 % of teams women-led, and several solutions focused on or using datasets from underserved populations, such as workers in care homes.
• Figure 4: The 21 participating organizations were each connected to one or more innovation clusters through funders, spin-outs, or accelerator programs; these ecosystem-level linkages were not planned a priori and emerged through analysis at a later stage.
• Figure 5: Case study overview: operationalizing Responsible AI through the governance dashboard design. Top row, before (1a–1e): the migraine decision-support team relied on fragmented, manual governance work. System characterization and risk framing were handled through ad hoc documentation and informal interpretation (1a–1b), while technical model-facing artifacts (for example, model saliency maps) were difficult for non-technical or time-constrained stakeholders to interpret and use in decision-making (1c). Reporting was assembled post hoc across multiple tools and formats (1d), producing slow, reviewer-heavy evidence packages, cumbersome reporting (1e) and limited internal alignment on what counted as “sufficient” Responsible AI evidence at a given stage. Bottom row (2a–2e, “after”): using the two-stage Sustainable Design Study framework and industry precondition mapping, the team rapidly configures governance views by TRL and target jurisdiction (2a), grounded in explicit stakeholder roles, obligations, and evidence dependencies represented as structured domain artifacts (2a-ii–2b). This enabled project-level governance dashboards that connect regulatory obligations to concrete evidence links and responsibility allocation for different users (2c and 2e) and ongoing external reporting enabling faster feedback.