Amplifying Human Creativity with Generative AI
How Generative AI Pushes Designers into New Directions
Contributors: Matt Klenk, Matt Hong, Shabnam Hakimi, and Yanxia Zhang
Problem Statement
Early in the design process, designers frequently create mood board collections of images to facilitate the exploration of ideas.
Below is a sample automotive mood board that evokes visions of “a beautiful summertime car” that is taken from a popular image search website. The car resulting from a design process using this mood board would likely have elements evoked from the different images (e.g., the retro steering wheel).
One challenge in this curation process is that it is biased by the designer’s previous work, their environment, and the time pressures to deliver the next great design proposal. Time pressure leads designers to use internet sources for inspiration. Unfortunately, this leads to a homogenization of the inspiration process, with designers at different companies and industries being drawn to the same images.
Recent advances in AI, including large language models and text-to-image models, place novel, unique images at the designer’s fingertips. We believe this provides a new way to unlock designers’ creativity.
TRI, with our partners at MIT, created an experimental software that allows us to compare different AI-mediated methods for curating inspiring content. As detailed in our paper that was awarded best paper at the ACM Collective Intelligence Conference, this tool will enable us to learn about the designer’s perceived differences between the new techniques of image generation versus existing approaches for image search.
We are building on the common practice of increasing creativity by exposing people to greater diversity. To this end, we created DesignAID, which uses a preprocessing step where the designer’s prompt is elaborated many times by a large language model. Then, DesignAID selects the prompts that are most different from one another and uses a text-to-image model to generate images from them.
Most previous work on creativity support tools focused on accelerating the content generation process, with few studies examining how AI can assist designers in the early stages by facilitating divergent thinking. In our study with 100 designers, we found that Generative AI methods were significantly favored by designers over image search in terms of inspiration, usefulness, and fun (as shown in the figure below).
Then, we compared DesignAID to directly invoking an off-the-shelf, text-to-image model; the designers rated the generated images from DesignAID’s diverse generation process as more inspirational.
These quantitative findings were supported by what participants said in interviews after using DesignAID. For example, one participant wrote, “As I used the tool, I was able to be inspired and come up with new ideas. A key element that influenced this is having the description of the generated images below.”
Impact
While generative AI has the potential to reshape creative work, we have yet to determine how these tools will translate into business impact. Our approach looks at new ways of combining AI tools and rigorously evaluating them to learn how they will impact business. This early-stage work exemplifies how our group has been experimenting at the intersection of Generative AI and creativity. By understanding how AI and humans can complement each other, we believe we will develop new technologies, leading to more inspired designs for Toyota products.
Links to Other TRI Work on Generative AI and Design
- Toyota unleashes generative AI to meld vehicle design with engineering principles https://www.tri.global/news/toyota-unleashes-generative-ai-meld-vehicle-design-engineering-principles
- Anticipatory Thinking in Design https://onlinelibrary.wiley.com/doi/full/10.1002/aaai.12101
