Using Generative Artificial Intelligence (GenAI) to Co-Design Digital Health Technologies

Key Points GenAI tools are widely available and can be used to enhance the co-design of digital health technologies. Digital and AI literacy should be addressed so researchers and participants understand the benefits, limitations, and risks of GenAI tools as they may influence the co-design process and its outputs. Digital accessibility also needs consideration so that a range of participants can use GenAI tools throughout the participatory design process or bespoke GenAI tools may need to be created to cater for the specific needs of end users involved in creating a digital health technology. Researchers should examine how GenAI tools can be used across all stages of co-design to evaluate their utility and impact and how best to utilize their creative outputs in the creation of a digital health technology. Co-designing digital health technologies with end users in health care can help address some limitations such as poor design that makes them difficult to use1 and ensure the needs and preferences of patients, medical doctors, nurses, allied health professionals, and others are incorporated in the design of software and/or hardware.2,3 Many different methods, tools, theories, and mindsets (principles or standards) are used when involving different groups of people in designing digital health technologies. The Double Diamond framework for innovation outlines four stages of design,4 each of which can include multiple ways to engage stakeholders in the design process (Fig. 1). For instance, interviews and focus groups can be conducted with patients to discover issues affecting their care, followed by observation in patients homes to help define a clear problem a digital tool may address. Paper and digital prototypes could then be developed5 and undergo rounds of laboratory or “in the wild” testing with users to refine a digital health technology before implementing it in real-world settings.6Figure 1: Framework for innovation4 (CC BY 4.0 license). https://www.designcouncil.org.uk/our-resources/framework-for-innovation/Generative artificial intelligence (GenAI) tools are starting to transform how digital health technologies are co-designed, similar to how artificial intelligence (AI) is reshaping other areas of health research.7 ChatGPT and other text-based chatbots, built on large language models, were quickly followed by other multimodal GenAI tools that can process and produce not only text but also images, video, audio, and other digital media based on prompts from human users.8 Popular text-to-image models include Midjourney, DALL-E, and Stable Diffusion, whereas text-to-video models comprise Sora, Veo, and Synthesia among many others. Although there are no clear guidelines for using these to co-design any type of digital health technology, we used a number of GenAI tools in two research studies to explore their potential to improve the participatory design process. BACKGROUND In the first study, called VRUnwind, we integrated several GenAI tools into an existing design framework called experienced-based co-design and utilized them with patients with type 2 diabetes to stimulate their creativity when designing a virtual reality (VR) app for mindfulness to help them manage diabetes distress.9 We used think-aloud interviews to enable each participant to explore AI image, audio, music, and video generators (e.g., Picsart, Murf, Soundraw, and Pictory) to create a range of digital content that reflected virtual environments they wished to practice mindfulness in. Then, we used the AI outputs alongside other research data in a focus group to prioritize key design features to inform the development of a prototype VR mindfulness app.10 In the second study, called COLLATE, we conducted a mixed-methods systematic review of co-design in digital health and used the results to co-create an online methods toolkit.11 The toolkit showcases all the methods, tools, theories, and mindsets used to co-design any type of digital health technology and includes an implementation framework detailing the barriers and facilitators in this complex process (https://www.digitalhealthcodesign.co.uk/). To co-create the toolkit, we ran several in-person and online interviews and focus groups with participants, a mix of early-career researchers and patients from a Patient and Public Involvement and Engagement (PPIE) forum. We supported them to use several GenAI tools (e.g., Midjourney, Veo, Animaker) to generate digital content that we could use to inform the design of the toolkit to make it more engaging and easier for researchers to use. Although the GenAI tools yielded useful creative outputs that informed intervention design, we learned some valuable lessons about incorporating these new tools in participatory design research. GENAI FOR CO-DESIGN First, there is a growing range of GenAI tools that can create various forms of multimedia with AI animator tools like Animaker now available in addition to text, image, audio, and video generators. Existing design products such as Figma, Adobe, and many others have also incorporated GenAI in their software platforms to offer a wide range of graphic and other design ideas and prototypes, with agentic support. Therefore, as researchers it takes time to learn to use these tools to automate content creation and decide which are most useful for a research study depending on the participants, digital health intervention, timeframe, budget, and other needs. Some GenAI tools such as DeepSeek are available open access,10 whereas others have free trial periods to enable users to try out their core features. However, many require a monthly or annual subscription and from our experience, paying for a more advanced package yields benefits as the additional functionality allows participants to create more engaging content and richer research data can be gathered. The administration of multiple license fees and logins can be burdensome but integrated platforms such as POE (https://poe.com/) enable researchers to pay a single subscription and access hundreds of GenAI tools. This may support their use when co-designing a digital health technology with various end users. DIGITAL AND AI LITERACY Second, it is important to consider digital and AI literacy and educate the research team, including PPIE members, and participants about AI11 so that everyone involved in using GenAI tools and their outputs is aware of their benefits, limitations, and risks as they may influence the co-design process. The lead researcher on both our studies undertook an online training course to increase their knowledge and skills in GenAI and conducts research on AI in health care. They distilled key insights to the research team such as problems with algorithmic bias12 and copyright concerns,13 and a short presentation was delivered to participants in both studies to highlight the advantages and disadvantages of GenAI before using them. There are some massive open online courses on GenAI more generally and a specific training course on AI in design thinking from IDEO that could be useful for researchers wishing to improve their expertise in this area and integrate GenAI into their co-design work. To address concerns around privacy and data security, we provided participants with access to a shared researcher-held account for each tool rather than requiring them to register individually. This approach ensured no personal data was collected by the GenAI platforms helping safeguard participant anonymity. We were also cognisant when analysing the creative outputs that AI bias may have affected some of the images, video, music, audio, or animations generated. Therefore, we are using coding clinics, member checking, and triangulation of creative data as ways to offset this risk and ensure the results represented participants views. DIGITAL ACCESSIBILITY Third, digital accessibility should always be considered in advance as some participants may not have a device such as a smartphone, laptop or tablet computer or access to a good quality Internet connection to enable them to get online and interact with GenAI tools.14 To improve digital equity, we offered some interviews and focus groups in person to enable a wide range of people to get involved. We also acknowledged that some participants might be hesitant or unwilling to directly engage with GenAI tools due to personal concerns or preferences. In such cases, participants were given the option to verbally articulate their prompts, which were then inputted into the GenAI tools by a researcher. This approach maintained participant agency in the co-design process while accommodating varying levels of comfort and trust in GenAI. In future studies, we plan to run co-design activities in the community in collaboration with local organizations that support minority groups15 to make sure they can take part in co-designing digital health technologies and learn how to use some GenAI tools as they are often under-represented in this area of research.2 Rodriguez et al16 implemented a digital access coordinator programme across a primary care network where multilingual ‘navigators’ were used to support a diverse range of people in accessing a patient portal, an approach worth exploring in co-design. A more diverse participant group would not only improve equality, diversity, and inclusion in digital health research17 but also yield richer research data on how GenAI tools can improve the co-design process, helping build the evidence base for these novel tools. BESPOKE GENAI FOR CO-DESIGN Nevertheless, current GenAI tools may not suit the needs of specific participants such as those who are neurodiverse or experience cognitive deficits.18 Hence, researchers can create their own bespoke GenAI tool for co-design when needed such as one that could help users with particular needs to create personas or storyboards to inform the design of a digital health technology. We plan to develop a customized chatbot to meet the needs of our participant groups as it could enable more accurate and contextually relevant responses, helping facilitate co-design by acting as a virtual design assistant that guides users through each stage of the creative process.19 Researchers could utilize existing large language models to develop a GenAI tool for co-design, use GenAI tools such as GitHub’s Copilot or Amazon’s CodeWhisperer to write programming code they need to create it, although this approach does have some downsides as the code may contain bugs and vulnerabilities creating a security risk.20 An emerging method is to leverage low-code/no-code AI platforms such as BotPress and Flowise AI enable users to build AI tools quickly while capturing the power of one or more large language models.21 This may speed up the software development process and possibly reduce errors when creating a new GenAI tool for co-design. Whichever approach it taken, researchers should consult with the intellectual property (IP) and commercialization department in their institution to ensure any potential IP when creating a new GenAI tool for co-design is deliberated and protected in advance. GENAI FOR DIGITAL HEALTH CO-DESIGN Fourth, digital health researchers should also examine GenAI tools for co-design in-depth especially how to utilize and publish the outputs of these creative tools given the copyright issues that affect them.22 At present, there are numerous legal cases in different countries where commercial companies that created GenAI tools are accused of breaching copyright law and using data inappropriately to develop their AI models and tools.23 Although we consulted with the copyright team in our university library about whether we could publish the GenAI outputs, their expertise did not extend to these digital tools as their legal implications are still unfolding and may vary across regions of the world. We also reviewed publishing guidelines from the International Committee of Medical Journal Editors Recommendations and guidelines on using GenAI in academic publishing from some of the top medical informatics journals. In the end, we erred on the side of caution and decided to use royalty-free digital images that were as similar as possible to what participants created through GenAI rather than publishing the originals. We also provided detailed descriptions of the original outputs when drafting manuscripts for publication. Researchers should also explore the ever-expanding range of GenAI tools to determine which are most useful for particular populations, interventions, outcomes, and settings when co-designing a digital health technology, alongside examining ethical issues such as autonomy, equity, privacy, and trust.24 Sharing these findings at conferences, in scientific journals, and through mass and social media will help us build and share knowledge in this emerging research area. The evidence base for these AI technologies is still in its infancy, and more research is needed across scientific disciplines to assess their impact on co-design in digital health. Human-centered artificial intelligence (HCAI) is already emerging in computer science as a way to utilize AI tools to amplify and augment rather than replace human software and hardware engineers,25 an approach that is also occurring in other design disciplines.26,27 Therefore, an interdisciplinary approach that involves designers from a range of backgrounds (e.g., graphic designers, illustrators, user experience designers) in combination with end users, GenAI tools, and existing co-design methods and tools may offer a more holistic way to develop digital health technologies. Mangal et al28 suggest these types of collaborations need careful planning with clear roles and responsibilities for each expert group, well designed and routinely used communication strategies, and adequate time for the design process to work across different disciplines. CONCLUSIONS We hope by sharing our experiences will the research community, GenAI tools will become more embedded in co-designing a digital health technology with patients, health and care professionals, and other users so that the evidence base on their use and impact will grow over time. As AI continues to advance, we may see more sophisticated digital tools that support all stages of co-design and help deliver more engaging, useable, and effective digital health interventions that improve health and other outcomes.