Our process
How we designed the 5-step set of resources, created to help those who need it to make decisions around AI systems and Public Health
Our starting point
The Toolkit project grew out of the Human Behaviour-Change Project. The Human Behaviour-Change Project (HBCP) is a research project aiming to develop an artificial intelligence system to make predictions about how effective different behaviour change interventions will be in a given context, based on evidence from published research studies.
The HBCP AI system could be used in public health decision-making to help decide what behaviour change services or support to provide to citizens. However, a key issue raised by numerous people during the project was whether public health decision makers could judge the trustworthiness of this and other AI systems they might use, and whether the public would want decision-makers to use such AI systems to inform decisions.
These issues led the HBCP research team based at University College London to apply for “Research Enrichment – Public Engagement” funding to conduct a public engagement project around trust in artificial intelligence systems used for public health. The public engagement project aimed to bring together members of the public interested in health and wellbeing, research scientists, public health practitioners and a designer for a series of workshops to create a toolkit of resources around appropriate trust in AI applied to public health.
Building our public engagement team
The first key task was to bring together the people who’d form the public engagement team. The researchers were fortunate to link with a company that provides behaviour change support services to many local authorities. The company advertised the project to its service users and staff, a number of whom expressed interest. The research team also partnered with UCL’s Co-Production Collaborative, who advertised the opportunity to take part in the project to their network of people interested in participating in health-related public engagement projects. Finally, the research team recruited a graphic designer with extensive experience in designing materials in collaboration with public engagement groups.
The workshops
The original project plan was to run in-person workshops in central London. However, due to the coronavirus pandemic, we had to change course. Instead of meeting in-person, the workshops were run online, using Zoom. The workshops were held on Wednesday evenings, because a poll of potential participants found that this was the most convenient time. In total, there were six weekly 2-hour workshops, held between September and October 2021.
Running the workshops online did present some initial technical challenges, particularly for participants joining using their phones. However, a real positive of conducting online workshops was being able to involve people from a wide range of geographical areas and people with health challenges who would have had difficulty attending in-person workshops.
Each workshop was facilitated by the research scientists and/or the graphic designer. While the research team started the project with a loose plan for the workshop content, the content of every workshop after the first was informed by the discussions in the previous workshop and feedback from participants. A brief summary of each workshop is provided below
Workshop 1
Introductions to team members and to the project.
Introduced the idea of public health decision-making about what services could be available
Workshop 2
Introduced AI by talking about familiar services that use AI technology, such as product recommendations in online shopping or film suggestions from online entertainment services
Talking about how AI works, through the example of teaching an AI system what pizza toppings you like
Discussing possible issues with AI:
Using inappropriate training data
Use of personal data
Difficulties understanding what an AI system has learned (explainability)
Biased performance for members of minority groups
Workshop 3
Started with further discussion of the benefits and possible issues of using AI
Discussed trust and what influences trust in general terms
Then looked at what people wanted to know to decide whether to trust an AI system
This information was synthesised and became the origins of the “Not Sure What to Ask?” resource in the Toolkit
Workshop 4
Started with an introduction to graphic design – the idea of using colours, fonts and imagery to make an audience feel a certain way
Moved on to considering how participants want the resources to look and what they wanted the audience to feel
Workshop 5
The designer shared revised designs
The group reviewed these updated designs for the Toolkit and provided feedback
Workshop 6
Started with further discussion of the benefits and possible issues of using AI
Discussed trust and what influences trust in general terms
Then looked at what people wanted to know to decide whether to trust an AI system
This information was synthesised and became the origins of the “Not Sure What to Ask?” resource in the Toolkit
After workshop 6, the graphic designer refined the designs again, and they were circulated to the public engagement group via email. The Toolkit was finalised based on the feedback received at this stage.