According to the international research team from KTH Royal Institute of Technology in Sweden, LMU Munich in Germany, and TU Darmstadt in Germany, the meme captions generated by AI score higher on average creativity, humor, and shareability as compared to human-written ones. But this doesn't mean that human humor isn't unique because its skill scored higher in individual memes. One of the researchers said that AI has passed the meme Turing test which means that it has become harder to distinguish between AI-generated and human-created content now.
For the study, researchers tested meme captions in three categories: food, work, and sports to see the humor levels of AI and humans in these contexts. The results showed that memes related to work scored the highest for shareability and humor which shows that context can play a major role in meme effectiveness. The images in the study were not AI-generated; instead, existing meme templates were used, with captions created by humans and GPT-4o.
The study also found that even though AI-generated memes were rated the highest for creativity, humor, and shareability, the funniest of the individual memes still came from humans. When humans and AI collaborated, they generated the most shareable and creative content. This shows that where AI is good at generating broadly appealing memes at large scales, humans are good at generating expectational memes. The participants who used AI to generate meme ideas did so with less effort which made the whole process easier. But human-AI collaborative memes did not score any higher than the memes made by humans alone. AI also helps with the quantity of memes more than the quality and the participants who used AI to create memes felt less ownership over their work.
AI models use broadly appealing patterns from the internet to make humor that is appreciated by consumers. Human-generated memes are often created from personal experiences, while AI tends to focus on safe jokes so they can appeal to larger audiences. So this raises the question of whether AI is really funny or does it just know which audience to target. There were some limitations to the study as well like the meme creation sessions where short AI collaboration was underutilized. Another thing was using crowdsourced evaluators which could favor mainstream humor.
Image: DIW-Aigen
Read next: Chatbot Loneliness Crisis? New Study Exposes the Emotional Toll of AI Conversations
For the study, researchers tested meme captions in three categories: food, work, and sports to see the humor levels of AI and humans in these contexts. The results showed that memes related to work scored the highest for shareability and humor which shows that context can play a major role in meme effectiveness. The images in the study were not AI-generated; instead, existing meme templates were used, with captions created by humans and GPT-4o.
The study also found that even though AI-generated memes were rated the highest for creativity, humor, and shareability, the funniest of the individual memes still came from humans. When humans and AI collaborated, they generated the most shareable and creative content. This shows that where AI is good at generating broadly appealing memes at large scales, humans are good at generating expectational memes. The participants who used AI to generate meme ideas did so with less effort which made the whole process easier. But human-AI collaborative memes did not score any higher than the memes made by humans alone. AI also helps with the quantity of memes more than the quality and the participants who used AI to create memes felt less ownership over their work.
AI models use broadly appealing patterns from the internet to make humor that is appreciated by consumers. Human-generated memes are often created from personal experiences, while AI tends to focus on safe jokes so they can appeal to larger audiences. So this raises the question of whether AI is really funny or does it just know which audience to target. There were some limitations to the study as well like the meme creation sessions where short AI collaboration was underutilized. Another thing was using crowdsourced evaluators which could favor mainstream humor.
Read next: Chatbot Loneliness Crisis? New Study Exposes the Emotional Toll of AI Conversations
