A recent study by USC reveals that humans might need to sharpen their joke-telling skills compared to ChatGPT. Published on July 3 in the journal PLOS ONE, the study examined how people react to jokes crafted by ChatGPT 3.5 versus those written by humans. The goal was to determine if artificial intelligence can be funnier than humans.
“Since ChatGPT can’t feel emotions but tells novel jokes better than the average human, this study shows you don’t need to feel emotions to tell a good joke,” the researchers explain.
Quality of jokes
In the first study, participants rated the funniness of new jokes created by both people and ChatGPT without knowing who wrote them. Nearly 70% of participants found ChatGPT’s jokes funnier, while just over 25% preferred the human-crafted ones. Only about 5% rated jokes from both sources equally funny. These results were consistent across different demographics.
To see how ChatGPT compares to professional humorists, the researchers conducted a second study. They had ChatGPT create new headlines in the style of The Onion, then asked another 200 participants to rate the funniness of these headlines alongside original Onion headlines. The participants found ChatGPT’s headlines just as funny as the originals.
For the first study, both ChatGPT and 105 participants generated original jokes across three tasks. The first task involved creating humorous phrases for common acronyms. The second required generating funny answers to fill-in-the-blank phrases, like “A lesser known room in the White House: _____.” The third task was to write a roast joke based on an awkward fictional scenario, such as telling a friend how poorly she sings. This process resulted in 945 jokes from the human participants and 180 jokes from ChatGPT. A new group of participants rated the funniness of these jokes.
In the second study, ChatGPT was given original headlines from The Onion and asked to create new ones in the same style. Another group of 200 participants then rated the funniness of both the original and ChatGPT-generated headlines without knowing their sources.
The researchers caution that their findings have implications for the entertainment industry. “For those who want to add humor to their daily communications, this is positive news,” they conclude. “But for professional comedy writers, LLMs could pose a serious threat to employment.”
Whether AI will ultimately outdo humans in humor remains to be seen.
