A recent study led by researchers from the University of Southern California compared the comedic abilities of OpenAI’s ChatGPT 3.5 with human writers, showing promising results for AI in generating humour.
Researchers from the University of Southern California (USC) have conducted a study to evaluate the comedic abilities of OpenAI’s ChatGPT 3.5 compared to human writers. Led by Drew Gorenz and Norbert Schwarz, the study aimed to determine whether AI can generate humor as effectively as humans.
In the first experiment, a group of U.S. adults completed three comedy-writing tasks: creating humorous acronyms, filling in blanks for funny scenarios, and generating roast jokes. These tasks were also given to ChatGPT 3.5. A separate group rated the responses for funniness on a seven-point scale. ChatGPT’s responses were preferred by 69.5% of participants over human responses.
The second experiment compared ChatGPT’s ability to write satirical news headlines against those from The Onion. A group of USC psychology students rated the headlines for humor. Results showed that 48.8% preferred The Onion’s headlines compared to 36.9% for ChatGPT’s, with 14.3% showing no preference.
The study found that ChatGPT often produced humor that was rated funnier than average human-generated content. This raised concerns about the potential impact on professional comedy writers. The research was published in the journal PLOS One.
