DISCOVERING THE OPTIMAL WAY TO USE AI IN THE BRAINSTORMING PROCESS
Imagine it was your job to plan an event to let students get their first taste of UW–Madison. You might propose a cheese carving contest. Or maybe they get to make homemade ice cream and have their first scoop looking out at Lake Mendota. The process by which you might generate these ideas is the focus of a new Communication Arts research project. By isolating different conditions, this study is discovering the optimal way to use artificial intelligence (AI) during the brainstorming process.
Ika Dai, a Communication Arts graduate student, described the study as a two-stage process exploring human-AI collaboration. “First, individuals generate creative ideas on their own or with the assistance of AI. Then, they work in pairs to continue generating and refining ideas.” Some of these pairs consist of two human participants, while other pairs consist of one person and an AI teammate.
Isolating specific conditions helps Dai understand whether receiving help from AI at the initial stage of individual brainstorming enhances creativity compared to when people generate ideas by themselves. After the initial idea generation, students engage with either a human partner or an AI teammate to build on and improve their ideas.
After each study session, the brainstormed ideas are graded based on the total number of generated ideas and their quality. “High-quality” ideas are those that score well for both originality (i.e., how innovative or novel the idea is) and feasibility (i.e., how practical it is to implement the idea.) Throughout the process, the research team can track when and how each idea emerged, whether it originated from the human participant or AI, and understand whether AI can function as an effective teammate, comparable to a human partner, by enhancing the overall quantity and quality of ideas.
After all the ideas are evaluated, the researchers can see which isolated condition produced the most high-quality ideas, and therefore, when it is most beneficial to use AI when brainstorming. So far, the research team has observed that people often generate more ideas initially without using AI assistance. This is consistent with other examples of “hybrid brainstorming,” according to Professor Lyn van Swol.
“If people brainstorm individually before getting into a group, they want to impress people, and this tends to motivate them to come up with better ideas to bring to the table,” van Swol said. Dai added, “On the other hand, when participants expect help from AI or know they’ll be working with an AI partner, they tend to put in less effort themselves and rely more on what the AI produces.”
Data is still being collected, so while the team cannot make concrete conclusions yet, they do note that the data shows humans tend to generate more ideas during the initial idea generation phases compared to those receiving AI support. However, AI overall appears to enhance the quality of the brainstorm by expanding on ideas, adding more specific details, determining logistics, or branching off to create even stronger ideas. Future results from this study are likely to have broad societal relevance as AI technologies continue to reshape everyday life.