Rapid Iteration: Leveraging Generative AI in Participatory Design Processes

In our paper written for the course: IN3250, we investigated the following question:

How can generative AI be used to strengthen the role of older adults as co-designers in a participatory design process?

Older adults are frequently underrepresented in the development of new digital technologies despite their increasing demographic importance. While Participatory Design (PD) emphasizes collaboration among diverse stakeholders, users with lower technological competence can find their active participation limited by challenges in imagining future technological solutions. Recent research, however, suggests that Artificial Intelligence (AI) could support active user involvement by lowering co-creation barriers, establishing a shared language, and bridging the conceptual gap between users and designers.

Our Approach: Methodology Highlights

To address our research question, we adopted an iterative process, relying on continuous feedback from experts before conducting a final user test with older adults. Our journey involved:

Developing "Rapid Iteration" Tool:

After an initial prototype (Prototype 1) proved more suitable for designers than workshop users, we developed Prototype 2, which became the "Rapid Iteration" tool. This dialog-based tool uses generative AI to produce images for critique and discussion, allowing for effective, multi-iteration image generation. Its purpose was twofold: as a warm-up activity to broaden mental models of generative AI, and more crucially, to explore solutions to real problems within a concrete design phase, aiming to empower participants as co-designers. The tool displayed generated images, a selected image view, a notes section for designers, and a prompt area for modifications based on user feedback.

Expert Evaluation Set-up:

We conducted an expert evaluation with nine experts in PD, HCI, and AI, many of whom have extensive experience working with older adults and have authored relevant literature. This evaluation focused on how the tool could enhance active participation among older adults in a workshop setting. We structured it with a presentation, a workshop walkthrough, and a shared digital document for experts to provide feedback on specific questions related to AI's ability to clarify options, recognize contributions, impact co-designer roles, and support decision-making.

User Test Set-up:

The final user test involved five older adult participants (aged 70-80) at Sagene Senior Center in Oslo, all familiar with AI tools and remote controls. We facilitated a 90-minute workshop using the Rapid Iteration Tool to explore representations of a remote control, supported by a persona and scenario to make the task relatable. Participants provided verbal input and feedback, leading to the generation of 24 remote control images.

Key Findings

Our study revealed several crucial insights:

Early Feedback Insights:

We learned the value of providing participants with summaries after sessions to highlight their contributions, the importance of facilitators managing diverse participant needs, and prioritizing active user participation, especially in generating prompts.

Expert Evaluation Outcomes:

Experts identified significant advantages and considerations.

• AI can expand the design space by providing new ideas that go beyond the user input.
• Rapid visualization of numerous potential solutions can lower the threshold for user critique, as participants clarify that the solutions presented are not ones that the designers have spent significant time developing.

• The introduction of AI in the design process requires increased focus on balancing the interaction between AI and the participants.
• There is an increased focus needed on ensuring that participants feel seen and heard, to enhance their perception of their role as co-designers.
• In the case of AI autonomy and interpretations, user ownership may not be facilitated in a sufficient way according to the PD objective of collaborative decision-making.
• AI's output may not reflect the user's input sufficiently, thereby constraining the design space and the PD process.
• It was suggested that participants might want to bring their own images or scenarios as input to the process, which could make the experience feel more real and authentic.
• Facilitators need to provide clear communication to foster an understanding of the relationship between the physical discussions and the AI-generated outputs, as older adults might struggle to understand their influence.

• Clarifying User Preferences: The Rapid Iteration Tool was highly effective in quickly identifying participants' preferences, particularly their dislikes, and facilitating discussions that extended beyond aesthetics to include functionality and materiality.
• Challenges with AI Behavior: We observed the AI occasionally fixating on irrelevant details, requiring significant time and prompts to adjust, which detracted from the core design focus. Participants expressed dissatisfaction with the AI's unresponsiveness and perceived that it did not "listen" to their feedback, generating "strange suggestions".
• Bridging Expectation Gaps: Participants did not fully grasp their role or how their contributions fit into the overall activity. There was a mismatch between our goal of stimulating discussions about various solutions and their expectation of agreeing on a final design.
• AI as the Third Participant: The AI unexpectedly took on a significant, often dominant, decision-making role. While it introduced new ideas, its uncooperative nature at times created a "clear sense of opposition" between the human participants and the AI, paradoxically strengthening the group dynamic and fostering a shared effort.