A new study conducted by researchers from KTH Royal Institute of Technology, LMU Munich, and TU Darmstadt has provided fresh insight into the role of artificial intelligence in humor creation. Tasked with generating internet memes using popular templates like Doge and Futurama Fry, groups of humans, artificial intelligence models, and human-artificial intelligence teams competed to see who could make the funniest, most creative, and shareable content. The results reveal that artificial intelligence can indeed produce memes that rate as funny and broadly appealing as those of humans, particularly when considering average scores across multiple examples.
Despite artificial intelligence-generated memes performing best on average, the sharpest and funniest memes still emerged from human creators. A key finding from the study is that while artificial intelligence excels at delivering high volumes of ideas with ´solid but average´ content, it struggles to achieve the emotional nuance, cultural context, and surprise factor that allows a joke to truly resonate. Human-artificial intelligence collaboration resulted in the most creatively diverse and widely shared memes, but these collaborations did not consistently yield the top scores in humor. The researchers observed that the best outcomes arose when humans actively curated and refined the ideas proposed by artificial intelligence, blending efficiency with human intuition.
Interestingly, participants in the collaboration group often underutilized artificial intelligence, with fewer than half engaging with the system more than once, highlighting the need for better iterative and dialog-driven creative tools. The authors suggest that future artificial intelligence should aim to facilitate deeper, more interactive co-creation processes rather than simply outputting large numbers of options. While artificial intelligence is poised to boost productivity and expand the reach of creative content, the human element remains vital for crafting humor that forges genuine connections and emotional impact. These findings were presented at the 2025 International Conference on Intelligent User Interfaces and published in the ACM Digital Library.
