Community in the loop development 1

The first HUMBLE Project PPIE workshop took place on 4th November at UCL and explored whether AI-generated summaries of human experience resonate with lived human experience. The group consisted of six diverse participants. Participants were asked:

Can the way computers analyse people’s experiences still feel meaningful to humans?

After a short introduction to HUMBLE’s aims, Dr Mel Ramasawmy provided an overview of qualitative research methods, showing how themes are constructed from real quotes. Participants then conducted their own thematic analysis using excerpts from people’s experience of health and wellbeing during Covid-19, collaboratively identifying key topics and emotions.
These human-led insights were then compared with AI-generated labels from Microsoft Copilot. Participants reflected on overlaps, gaps, and what was lost or gained in the machine analysis. The workshop concluded with an animation explaining how large language models work, supporting discussion about where public contributors add value in AI-assisted research.

Participants were sceptical of the AI-based analysis. A central question from the discussion: can we interrogate what is not there in the data? In other words, what might be missing, or misrepresented, when data is interpreted without human context? Participants reflected on whether AI or automated systems can simulate the work of PPIE contributors—and where in the research process humans are essential.

Participants strongly felt that humans are needed to identify what isn’t captured in the data. While AI can generalise, it doesn’t offer the contextual grounding, emotional nuance, or experiential resonance that PPIE groups provide. Humans can triangulate perspectives, drawing on lived experience and mutual dialogue. One participant noted that research, when done well, invites people to step outside their comfort zones and into others’ shoes. PPIE work makes this bridge possible in a way that AI systems may not be able to.

There was a sense that AI lacks the lived experience “in the room.” It may offer neat summaries, but participants felt it often misses deeper meaning and representation—especially of groups marginalised in mainstream research. For example, people with limited literacy, BME communities, or non-native English speakers are often underrepresented. 

Another discussion revolved around the idea that AI may not see what’s important—because it can only work with what it’s given. This emphasises the importance of PPIE at various research stages, including during survey or dataset design.

THEMATIC ANALYSIS

As part of the session, participants worked in pairs to thematically analyse the data. They organised quotes into the following themes. 

1. Social Connection

• People valued family and friendship, but also described fear of others during Covid-19.
• Isolation, lack of contact, and longing for normal connection.
• Empathy was seen as both important and sometimes absent.

2. Living Conditions

• Those with access to green space, fresh air, and secure housing may have had a better experience of Covid-19.
• Overcrowding, damp, mould, or poor housing made things harder.
• Being able to “carry on as normal” was protective for some.

3. Family & Wellbeing

• Concerns about children’s futures and loved ones’ health came up frequently.
• Some expressed gratitude for good health and access to food or exercise.
• Financial security was often tied to family wellbeing.

4. Finances & Employment

• Some experienced job loss or financial instability.
• Remote work and volunteering were positives.
• Many voiced uncertainty about the future.

Meta-reflections of the group

• Several participants noted the dataset lacked representation of different cultures, disabilities, or age groups.
• Questions were raised about how inclusive or representative the data really was.
• One person said the dataset didn’t reflect their community’s experiences at all.
• Some of this may have been a misunderstanding of the exercise (expecting the data to be exhaustive), but the critique itself is useful: PPIE contributors want to see themselves in the data.