Researchers from KTH Royal Institute of Technology, LMU Munich, and TU Darmstadt recently conducted the first large-scale study exploring how humans and AI collaborate to create internet memes. The team compared three groups: humans working alone, humans co-creating with a cutting-edge language model (LLM), and the LLM generating memes entirely on its own.

Participants created memes using classic templates like Doge, Futurama Fry, and Boromir’s iconic “One does not simply…” line. A second group of nearly 100 people then rated the memes for creativity, humor, and shareability.

The researchers found that on average, memes made entirely by AI scored higher than those made by humans or human-AI teams. But the top-performing memes told a different story: humans were funniest, while human-AI collaborations stood out in creativity and shareability.

Published on the ACM Digital Library, the paper was presented at the 2025 International Conference on Intelligent User Interfaces in Cagliari, Italy.

“AI is great at generating lots of ideas quickly,” says the study’s co-author, Zhikun Wu, a master’s candidate at KTH Royal Institute of Technology. “But quantity doesn’t always mean quality.”

Able to draw from vast datasets, AI models can produce content that appeals to a wide audience, the authors wrote. But many of the top-rated memes were created with human involvement, suggesting that AI models primarily produce “solid but average quality.”

“The best results came when humans curated and refined what the AI produced,” Wu says.

Participants who worked with the AI assistant generated more ideas and reported less effort—but many didn’t fully engage with the system. Fewer than half interacted with the AI more than once, and only a handful used it iteratively. This limited use may have held back the potential of true co-creativity.

The study highlighted a key challenge in human-AI collaboration: while AI can produce content that appeals to a broad audience, human creativity is still essential for content that resonates deeply—especially in humor, Wu says.

“Humor isn’t just about punchlines,” Wu says. “It’s about surprise, cultural context, and emotional nuance—things AI doesn’t fully grasp.”

The researchers argue that future AI tools should better support iterative, dialog-based creativity, helping users stay connected to their work while amplifying their ideas. In other words, systems shouldn’t just generate content, but help people shape it into something meaningful.

“While AI can increase productivity and produce content that appeals to a wide audience, human creativity is still key for creating content that connects more deeply in certain areas,” the authors wrote.