A recent study has shown a surprising finding: AI-generated comedy, specifically from OpenAI’s ChatGPT 3.5, is considered funny or funnier than human-written jokes, including those crafted by professional writers.
Writing comedy is a challenging task, as humour is highly subjective. So, how does AI, specifically ChatGPT 3.5, stack up as a comedy writer?
Through extensive research, researchers from the University of Southern California (USC) aimed to provide an answer to this question. “These studies provide evidence that you don’t need to feel the emotions of appreciating a good joke to tell a perfect joke yourself,” said Drew Gorenz, the study’s lead author and a doctoral candidate in social psychology at USC.
“Since ChatGPT can’t feel emotions, it tells novel jokes better than the average human.”
While AI’s ability to be humorous has been studied before, this is the first study to compare human and AI-generated comedy. Gorenz and Provost Professor of Psychology and Marketing Norbert Schwarz conducted two experiments to evaluate ChatGPT’s and humans’ humorous abilities.
In the first study, a group of US adults completed three comedy-writing tasks: creating humorous phrases for acronyms (e.g., ‘S.T.D.’, ‘C.L.A.P.’, ‘C.O.W.’), filling in the blanks for humorous scenarios, and constructing roast jokes. For instance, participants had to respond to a scenario where a friend with a terrible singing voice asked for feedback. ChatGPT 3.5 was given the same tasks, and another group of adults rated the funniness of the responses on a seven-point scale.
The outcomes were remarkable. Participants considered ChatGPT’s responses funnier than human ones, with 69.5% thinking the jokes produced by the AI were more amusing. The investigators stated, “Overall, ChatGPT 3.5 performed above 63% to 87% of human participants depending on the humour task.” Given ChatGPT’s aim of avoiding offensive speech, the AI surprised everyone by demonstrating exceptional strength in the roast joke task.
The second study compared ChatGPT’s ability to write satirical news headlines akin to those on The Onion. Without regular news updates, ChatGPT was tasked with generating headlines based on The Onion’s ‘Local’ news section. USC psychology students rated the funniness of these headlines, with the results showing a close competition: 48.8% preferred The Onion’s headlines, 36.9% favoured ChatGPT’s, and 14.3% had no preference.
“Participants, on average, rated the headlines as similarly funny, indicating that the average participant did not discern a difference in quality,” the researchers observed.
These findings are both intriguing and concerning. “That ChatGPT can produce written humour at a quality that exceeds laypeople’s abilities and equals some professional comedy writers has important implications for comedy fans and workers in the entertainment industry,” the researchers stated.
Moreover, they acknowledged that large language models like ChatGPT could pose a significant employment threat to professional comedy writers.
This study was published in the journal PLOS One.
Source: USC