Speaker
Description
Introduction
Citizens' juries and assemblies are increasingly popular public participation methodologies for deliberation on data and artificial intelligence. They are formal, top-down exercises that aim to address power imbalances in the design, application, or regulation of data and AI processes through allowing residents to debate and provide recommendations on specific questions. In March 2025, we invited over 60 residents in the Liverpool City Region (LCR) in the United Kingdom to take part in a Residents’ Assembly on Data and AI Innovation. We specifically asked residents what trustworthy and beneficial data and AI practice looks like for the City Region to build a Charter on Data and AI Innovation. In this paper, we critically reflect on the design and outcomes of the Assembly and whether it was able to better represent public voice in the development of regional data and AI policy and practice in the UK.
Assembly Methodology and Data Sources
Designing the Assembly
The Assembly was designed and run by the LCR Civic Data Cooperative, a local government-funded research project. It was convened jointly by local health, government, and University partners. We aimed to recruit 60 residents via randomised postcode sampling. Residents who signed up were then stratified to represent seven demographic characteristics in the region including knowledge of AI.
Residents took part in four full days of activities as well as two online or phone sessions. This included group induction, two days of learning on data and AI, two days of interactive deliberation, and a final debrief session. The core activity of the latter two days was a ranked choice vote on resident suggested Charter principles.
Around this process, we specifically sought to address inclusivity and ensure diverse representation. Residents were compensated for their time, travel, and meals. All recruitment was completed via post to address digital exclusion. Some of the additional services advertised to residents included letters of support for employers, translated materials, live translation, childcare, and 1-2-1 debrief sessions.
Data Sources
Throughout the sessions, facilitators took notes and select recordings to report on resident perspectives, key questions, and reflections. Notes and transcribed recordings will be thematically analysed for common themes related to these topics. In addition, we report the process and validation of the principles put forward by residents through the two-days of deliberation, the ranked choice voting exercises, and the debrief sessions. Voting was fully anonymous using Poll Everywhere software. Residents ranked the principles from most to least importance. A formal independent evaluation was also conducted and is forthcoming.
Early Results and Discussion
Principles and Resident Perspectives
59 residents took part in the four in-person Assembly sessions. 60 principles were put forward by residents. These principles were reduced by the Assembly team for repetition and semantic similarity. These 22 unique principles were then ranked by residents and the top 12 were taken forward for further review. 55 total votes were recorded during the ranked choice vote. The final set of principles will be available in May 2025. They currently include four key concept areas including benefits and harms, inclusivity and transparency, oversight and accountability, and security and legality. Findings on key resident hopes and concerns are forthcoming as full data collection will be complete in April 2025.
Process
Early reflections on the process of designing the Assembly include three core concepts: representivity, literacy challenges, and positive bias.
First that the nature of a Jury or Assembly process is exclusionary by design. 60 residents cannot, and are not, expected to be representative of the full range of perspectives on data and AI. Four full days is a significant investment in time for residents and inevitably causes a self-selection in who takes part. While childcare and translation services were offered, they were not taken up by any residents. This means exercises must, themselves, challenge participants to think about different perspectives on data and AI in order to represent diversity in an Assembly setting.
Second, the necessity of accommodating a variety of levels of data and AI literacy meant that some activities were too complicated and some too simple for the residents who took part. Thus, some were more and less successful an eliciting critical reflection on the contents of a potential data and AI charter. Embracing reflexivity was as important as robust planning for learning and deliberation activities. However, we found an Assembly methodology was not well suited to representing the voices of those with no knowledge of data and AI.
And third, the nature of positive bias both in how materials were presented and what residents then subsequently shared back about their perspective on data and AI. While we continuously worked with our external evaluator to ensure we adapted materials to maintain balance in what was presented, there were challenges in ensuring all residents fully reflected on the harms and risks of data and AI technologies.
Conclusion
The Assembly was successful in developing regional data and AI policy, namely a Charter. Similar deliberative events can learn from our experiences and challenges to ensure residents are fairly represented in the design of regional data and AI policy and practice.
