ChatGPT’s Humor Challenges Professional Writers

Who’s funnier, AI or humans? A recent study shows that we often find ChatGPT’s comedy as funny or funnier than human-written jokes, even those by professional writers. This implies that AI could significantly impact the entertainment industry.

Creating comedy is challenging because humor is subjective; what one person finds hilarious, another might not. Comedy writers must skillfully balance timing, delivery, originality, and avoiding clichés.

So, how does OpenAI’s ChatGPT 3.5 perform as a comedy writer? Can it be genuinely funny? Researchers at the University of Southern California (USC) recently explored this question and shared their findings.

ChatGPT Excels in Crafting Jokes Without Emotions

“Although ChatGPT doesn’t experience emotions, it often crafts novel jokes better than the average person. This suggests you don’t need to feel the emotions of enjoying a joke to tell a great one,” said Drew Gorenz, a doctoral candidate in social psychology at USC Dornlife College, an amateur stand-up comedian, and the study’s lead and corresponding author.

Previous research has explored ChatGPT’s ability to generate humor, but not with a thorough evaluation and comparison to human-written comedy. Gorenz, along with Norbert Schwarz, a Provost Professor of Psychology and Marketing, aimed to address this by conducting two comprehensive studies.

However, In the first study, a group of US adults completed three different comedy-writing tasks. In the acronym task, participants created humorous phrases for acronyms like ‘S.T.D.’, ‘C.L.A.P.’, and ‘C.O.W.’.

In the fill-in-the-blank task, they filled in blanks for prompts such as, ‘A remarkable achievement you probably wouldn’t list on your resume: ________.’

Participants Tested on Crafting Roast Jokes

Lastly, the roast joke task required them to craft a funny response to a hypothetical scenario. For example, ‘Imagine a friend asks for your opinion on her singing. After a painful demonstration, she asks, “So how was it?” You decide to be honest and respond, “Listening to that was like ________.”’

ChatGPT 3.5 was then assigned the same tasks.

Another group of adults evaluated the humor of the responses using a seven-point scale, where zero meant “not funny at all” and six indicated “very funny.” ChatGPT’s responses were found to be funnier than those from humans, with 69.5% of participants favoring the AI-generated jokes (26.5% preferred the human-written ones, and 4.0% felt both were equally amusing).

“Overall, ChatGPT 3.5 outperformed 63% to 87% of human participants, varying by the type of humor task,” the researchers noted. “ChatGPT 3.5 excelled particularly in the roast joke task, which was surprising given its aggressive nature. Since ChatGPT is programmed to avoid offensive or hateful speech, we might have expected the opposite result.”

Assessing ChatGPT’s Satirical News Headline Generation

In the second study, the researchers assessed ChatGPT’s ability to create satirical news headlines similar to those from The Onion. Due to the AI’s lack of regular world news updates, they used the last 50 headlines from The Onion’s ‘Local’ news section before October 1, 2023, such as ‘Man Locks Down Marriage Proposal Just As Hair Loss Becomes Noticeable.’ ChatGPT was then asked to generate 20 new headlines.

USC psychology students rated the humor of these AI-generated satirical headlines on the same seven-point scale used previously. They also rated how much they engaged with comedy and satirical news. Students who reported a higher interest in comedy found the headlines funnier, regardless of whether they were AI-generated or written by professionals. According to average ratings, 48.8% preferred The Onion’s headlines, 36.9% favored ChatGPT’s, and 14.3% had no preference.

“On average, participants rated the headlines as equally funny, suggesting that they couldn’t distinguish between the quality of AI-generated and professionally written humor,” the researchers stated. “This is notable given the high caliber of the professional comedy writers used as a benchmark.”

This finding is not only intriguing but also concerning, as noted by the researchers.

“The fact that ChatGPT can create written humor that matches or surpasses the skills of the average person and is comparable to some professional comedy writers has significant implications for both comedy fans and industry professionals,” they said. “For professional comedy writers, our results indicate that LLMs [large language models like ChatGPT] could pose a substantial employment threat.”

Read the original article on: New Atlas

Read more: ChatGPT and The Dark Web, Yet, A Hushed Talk in The Tech World