Why You Shouldn't Outsource Brainstorming to AI

Recently, I led a design thinking workshop where something telling happened. During a breakout session, one team member, lacking deep product knowledge, turned to AI (Perplexity and Google Gemini) to understand common customer complaints. The AI provided a broad overview of pain points scraped across the web, giving the team some discussion fodder.

But here's the thing – those AI-generated insights felt generic, lacking the rich context and nuanced understanding that comes from someone who works directly with the product and its customers daily.

Will AI Reshape Creative Processes?

This moment crystallized something I've been thinking about as AI increasingly infiltrates our creative processes: while AI can generate quantities of ideas and surface general patterns, it fundamentally cannot replace the depth of human expertise and collaborative creativity.

The fundamental question isn't whether AI will replace human creativity, but how it might reshape our creative processes. As someone who has spent years studying how ideas emerge and spread, I've become increasingly fascinated by this intersection of human and artificial intelligence in one of our most quintessentially human activities: brainstorming and ideation.

What Recent Research Tells Us

A series of recent studies have started to illuminate both the potential and limitations of AI-assisted ideation. In a detailed study examining AI-augmented brainwriting, Shaer et al. (2024) investigated how large language models (LLMs) affect both the divergence stage of idea generation and the convergence stage, where ideas are evaluated and selected. Their findings reveal a complex picture – while LLMs can enhance brainstorming productivity, they may also lead to more homogeneous thinking.

The research shows clear productivity gains when AI enters the picture. Shaer et al. found that when participants used ChatGPT during brainstorming sessions, they generated about 15% more ideas compared to traditional methods. More tellingly, they discovered that ideas hit about 27% more categories when using AI assistance, suggesting greater breadth in ideation. As one participant reflected:

"It helps to broaden the horizon and get inspired to think in different ways or areas than before. After I took the suggestions I was able to come up with a few additional ideas I didn't think of before"

However, the same study revealed concerning patterns. When analyzing the semantic similarity of ideas, researchers found that "ideas produced with the help of ChatGPT were significantly less divergent from the average embedding of all ideas generated for that task" compared to ideas generated using other methods. This suggests that while AI might help generate more ideas, it may inadvertently push everyone toward similar patterns of thinking.

The Hamburg study by Memmert and Tavanapour (2023) reinforces these findings. Their research uncovered a phenomenon they call "cognitive inertia," where participants reported that while AI suggestions "gave [them] a direction to think about," they also "closed other directions down." Some participants even noted that the AI made them "come up with ideas that were close to the ones the AI provided."

What We Risk Losing

This homogenization effect is particularly concerning because it may undermine one of the core benefits of group brainstorming: the diversity of perspectives. Shaer et al. found that when working with AI, 79% of participants preferred it to working alone, but many still expressed a preference for human collaboration, citing the "social aspect" and believing it would be "more fun" to work with other people.

Quantity vs. Quality

The quantitative data is equally revealing. In the Shaer study, while participants generated more ideas with AI assistance (mean=8.39 ideas with AI vs. 7.32 without), the semantic diversity of these ideas was significantly lower at the group level. Even more interesting, they found that about 35.2% of the ideas selected as "best ideas" by participants originated from the AI system, suggesting a concerning level of reliance on AI-generated content.

Another study by Anderson, Shah, and Kreminski (2024) found that while AI can help increase the number of ideas generated, it may lead to concerning levels of homogenization in the outputs. The researchers observed that:

"The ideas generated with assistance from ChatGPT were significantly less semantically diverse at the group level than ideas generated with assistance from the non-AI-based [creativity support tools]."

Protecting Human Creativity

While AI can serve as a tool for generating starting points or providing broad perspectives, we shouldn't mistake its ability to produce volume with its ability to produce value. Real innovation – the ideas that truly solve complex problems – happens when humans with sufficient context and experience spend focused time thinking together.

This isn't just intuition; it's backed by decades of research showing that cognitive diversity combined with relevant expertise creates the fertile ground from which breakthrough ideas spring. As one study participant noted, they would have "preferred to work with another human as compared to working with the AI, e.g., due to the 'social' aspect or because they believed it would be more fun" (Memmert & Tavanapour, 2023).

The challenge isn't about finding the right balance between human creativity and artificial intelligence. It's about protecting and nurturing the conditions that enable people to be creative and collaborative. We should be clear-eyed about AI's limitations: it can help generate more ideas, but it cannot replace the depth of understanding, intuition, and magic that happens when knowledgeable humans work together.

In the end, the best ideas don't come from AI, or even from AI-human collaboration. They come from groups of diverse humans with sufficient context and experience, spending focused time together to solve problems they deeply understand.

Studies Cited

Anderson, B. R., Shah, J. H., & Kreminski, M. (2024). Homogenization Effects of Large Language Models on Human Creative Ideation. Proceedings of C&C '24.

Memmert, L., & Tavanapour, N. (2023). Towards Human-AI-Collaboration in Brainstorming: Empirical Insights into the Perception of Working with a Generative AI. In Proceedings of the 31st European Conference on Information Systems.

Shaer, O., Cooper, A., Mokryn, O., Kun, A. L., & Ben Shoshan, H. (2024). AI-Augmented Brainwriting: Investigating the use of LLMs in group ideation. Proceedings of CHI '24.