Interaction Design with Generative AI: An Empirical Study of Emerging Strategies Across the Four Phases of Design

TL;DR

It is demonstrated that Generative AI can successfully support the designer in all key phases, but the generated outcomes require manual quality assessments, and the successful prompting patterns used to create or evaluate outcomes of design activities require different structures depending on the phase of the design and the specific design activity.

Abstract

Generative Artificial Intelligence (Generative AI) holds significant promise in reshaping interactive systems design, yet its potential across the four key phases of human-centered design remains underexplored. This article addresses this gap by investigating how Generative AI contributes to requirements elicitation, conceptual design, physical design, and evaluation. Based on empirical findings from a comprehensive eight-week study, we provide detailed empirical accounts and comparisons of successful strategies for diverse design activities across all key phases, along with recurring prompting patterns and challenges faced. Our results demonstrate that Generative AI can successfully support the designer in all key phases, but the generated outcomes require manual quality assessments. Further, our analysis revealed that the successful prompting patterns used to create or evaluate outcomes of design activities require different structures depending on the phase of the design and the specific design activity. We derive implications for designers and future tools that support interaction design with Generative AI.

Figures (22)

• Figure 1: An overview of artifacts generated by the study participants. The artifacts cover selected examples and demonstrate how Generative AI could make helpful contributions in the four phases of user-centered design.
• Figure 2: P2 used a meta-prompt to generate a structured prompt that he could use to create scenarios.
• Figure 3: Prompts provided by P7 to identify components for a hardware prototype together with Generative AI. The key elements of this iterative prompting approach are annotated as explained in the text.
• Figure 4: Two different prompting patterns deployed by P10. (A) P10 deployed a persona and a flipped interaction pattern to design a controlled experiment with Generative AI. The flipped interaction pattern initiates a conversation with the Generative AI. (B) A single prompt in which P10 deployed a persona pattern to review a controlled experiment, immediately followed by the description of the experimental design. In this case, P10 favored a minimal interaction over an iterative prompting approach.
• Figure 5: Answers to the three items in the Likert questionnaire gathered in the second user study. The items evaluate the quality, level of detail and relevance of the generated artifacts. The x-axis displays the number of participants reporting positive responses (right of the central diverging point) versus neutral to negative responses (left). Overall, participants acknowledged the relevance of the artifacts to the provided design task, but were more critical of the quality and the adequacy of the provided level of detail.