SHSanaz HosseiniUX Researcher Resume ↓
← All work PhD contribution · framework

PALEI: designing AI explanations with communities

Participatory AI Literacy and Explainability Integration: a six-stage framework for designing resident-facing AI tools before the model is finalized, with a documented artifact at every stage.

My roleCreator; doctoral research
Timeline2022 – May 2027
CasesWildfire risk (primary) · smart transit
StatusStages 1–3 complete · Stage 4 sessions complete · 5–6 next
PALEI framework diagram
What's new

Integration, not another set of principles

PALEI builds on participatory design, explainable AI, AI literacy, and human–AI interaction. Its contribution is how these are integrated and operationalized: community evidence becomes AI design requirements, competing explanation strategies, and evaluation and governance criteria, each documented as an artifact.

Early community involvement

Residents help define what the tool should explain before the model and key technical decisions are fixed.

AI literacy throughout

Every stage considers what residents must understand about outputs, uncertainty, limits, and data practices.

Evidence-to-design traceability

Findings become documented requirements, design choices, and evaluation criteria you can trace back.

Six stages

Each stage has an AI-literacy role and a documented output

01

Contextual framing, AI readiness, and foundational literacy

Define what the proposed AI would do, use, and not do, and who should be involved.

Outputs: AI System Flow Map, Scenario Cards, Readiness and Fairness Audit

Complete
02

Participatory problem definition and AI value alignment

Residents discuss the problem and the AI itself: trust, privacy, fairness, how results should be shown.

Output: AI Value Map, turning input into design requirements before any model exists

Complete
03

Needs assessment, value translation, and explanation personas

Translate requirements into explanation needs, including where people may misread a result.

Outputs: Explanation Personas and seven shared design principles

Complete
04

Co-design and explainability prototyping

Residents compare alternative explanation designs of the same result; short tasks check interpretation.

Outputs: Bias and Fairness Monitoring Log, screen-level Findings Matrix

Sessions complete
05

Evaluation of understanding, reliance, and fairness

Test one combined, clickable prototype for correct interpretation, uncertainty, reliance, fairness, and usability.

Outputs: task-based usability and interpretation results

Before May 2027
06

Governance, knowledge transfer, and long-term explainability

Responsibilities, uncertainty over time, data rules, and onboarding so explanations stay accurate after handoff.

Outputs: handoff plan, uncertainty guide, bias and coverage checklist

Planned

AI literacy here means what residents need to understand about a specific AI tool to use it appropriately: what it does, how to read results, uncertainty and limits, when professional judgment is needed, and their data rights (adapted from Long and Magerko, 2020).

One concern, traced through all six stages

How a privacy worry became a design decision and a test

This is what a product team gains from PALEI: every requirement has a visible trail back to the people who raised it, and forward to how it will be checked.

Stage 1Scope stated

The System Flow Map shows what images the tool would use and what it would not collect.

Stage 2Concern measured

Privacy or data security was the top concern (64.3%); residents worried about insurer access.

Stage 2 → 3Requirement written

Value Map dimension “Privacy Boundaries” → principle: show data practices before any upload.

Stage 4Designed & probed

A Data & Privacy screen appears before media submission and is shown to every co-design participant.

Stage 5 · nextTested

Evaluation criterion: can residents correctly say who sees their images and when they are deleted?

Stage 6 · nextGoverned

Data-retention and sharing policy: no insurer sharing by default, documented for whoever deploys the tool.

Applied in two domains

Same method, different systems

The smart transit case helped establish the broader AI integration problem. The wildfire case is the primary application of the revised framework, from AI literacy at a town hall to co-design and usability testing.

Why it matters for product teams

A traceable path from evidence to every screen

Explanations designed before launch

Teams learn what users need explained while it's still cheap to change the model and the interface.

Evidence for every decision

The AI Value Map and Findings Matrix link each requirement to the participant evidence behind it.

Fairness checked throughout

A bias log runs during prototyping, not as an audit after release.

Reusable in other domains

Built for wildfire risk and transit; designed to transfer to health, finance, and civic AI.