PPIE Programme Plan: How AI-generated vaccine stories become believable
This is the design plan for a four-workshop public programme exploring how AI-generated vaccine imagery becomes believable. It is part of my UKRI Smart Data Research UK Fellowship at Manchester Metropolitan University. It is version 0.1, and it is provisional: the timings, activities and facilitator wording are all still open to change.
The question at its centre is simple to ask and hard to answer. What makes a misleading vaccine image feel credible, urgent, emotionally powerful or worth sharing? The programme will work with 10 to 15 members of the public in a community venue local to them, across four sessions of three and a half hours, with food from arrival. No specialist knowledge of AI or vaccines is needed. Participants are contributors shaping the research rather than subjects of it, and they are paid for their time.
Caption: The programme's four stages. Each workshop begins with familiar, everyday experiences of images online before moving towards vaccine-specific material.
The design leans on the DARE UK pilot public dialogue, which set out six principles for public conversations about complex data and AI: begin with the familiar and funnel inwards, build in time to reflect between sessions, meet people in their own community settings, favour hands-on activities over presentations, illustrate concepts with real examples, and test your language with the people who have to use it. To those I have added boundaries for a subject that can carry an additional personal sensitivity.
Caption: The session's emotional intensity curve. The most sensitive material appears only once trust and shared ways of working are established, and the final fifteen minutes bring the intensity back down.
Perhaps the clearest way to describe the programme is to say what it will not do. Nobody will be asked to disclose their vaccination status. Nobody will be asked to defend or change their views about vaccination, to classify every image correctly, to offer clinical interpretations, or to reach agreement about contested experiences. "I need more information" is treated as a valuable response, not a failure. This is not a detection test, and participants are not being scored.
The fellowship's tool, SDA Vision, sits deliberately far back in the design. It appears in Workshop 2, as one evidence source among several within the Check stage. It assesses how synthetic an image appears and explains the signals behind that reading. It does not decide whether a claim is true. That judgement needs source review, medical evidence and human interpretation, and it stays with people.
Caption: SDA Vision v0.1.0: Provenance-aware synthetic media analysis. Every provider is rated side by side, and the verdict is driven by agreement, not a single confident model.
Fellowship advisors will review the plan. An ethics decision is expected by mid-August, and the first workshop runs in the autumn. Whatever changes as a result will be recorded in a "You Said, We Changed" log and shared with participants at every session.
Read the full plan. This pilot plan is published while it can still be changed,. If you have a view on the scope, the language, the activities or the boundaries, please email: smartin@mmu.ac.uk