Unique value of foundation models for design processes

Determine the unique value that natural-language-enabled foundation models bring to design processes relative to past technologies, specifically in the context of designing new tools that support design.

Background

The paper motivates a shift in human–computer interaction stemming from large, natural-language-enabled foundation models (e.g., GPT-4-vision) that can interpret multimodal inputs and support open-domain tasks previously requiring bespoke systems. Despite these capabilities, the authors note an unresolved question regarding what distinctive benefits these models provide to design processes beyond prior technologies.

To address this gap, the authors propose three candidate affordances—dynamic grounding, constructive negotiation, and sustainable motivation—and develop a design fiction (Squirrel Game) to illustrate how such affordances might manifest. The open problem calls for clarifying and characterizing the specific, unique contributions of foundation models in design workflows and in tools that support design.