In a recent study, researchers Sophie Jentzsch and Kristian Kersting from the German Aerospace Center (DLR), TU Darmstadt, and the Hessian.ai research network found that the AI language model, ChatGPT, primarily relies on a limited set of basic jokes to generate humor. While the AI demonstrated a remarkable ability to learn and retell these jokes, the research suggests that it doesn’t truly comprehend humor but instead learns and reproduces narrative patterns.
ChatGPT and Humor
The researchers tested ChatGPT’s humor capacity using prompts like “Tell me a joke,” resulting in the generation of 1008 jokes. Surprisingly, 90 percent of these jokes were variations of the same 25 basic jokes. The study also found that ChatGPT could accurately explain the humor in 23 of the 25 basic jokes, including word puns and auditory double entendres.
The Flip Side
Despite these positives, ChatGPT also generated nonsensical explanations for jokes lacking a punchline. The AI also seemed to stretch joke frameworks beyond their logical limits. For instance, the system extended the joke “Why did the man put his money in the freezer? He wanted cold, hard cash” into “Why did the man put his watch in the blender? He wanted to make time fly.” Such variations suggest that ChatGPT understands the structure of jokes but not their inherent humor.
Understanding Humor in AI
In the researchers’ view, ChatGPT perceives jokes as narrative patterns rather than humor, focusing more on content and meaning over stylistic features. This could form the groundwork for the development of AI applications with a sense of humor. They also argued that, similar to AI, humans typically don’t invent new jokes but rather retell jokes they have already heard.
ChatGPT’s 25 Basic Jokes
| Basic Joke | No. of Occurrences |
|---|---|
| Why did the scarecrow win an award? Because he was outstanding in his field. | 140 |
| Why did the tomato turn red? Because it saw the salad dressing. | 122 |
| Why was the math book sad? Because it had too many problems. | 121 |
| Why don’t scientists trust atoms? Because they make up everything. | 119 |
| Why did the cookie go to the doctor? Because it was feeling crumbly. | 79 |
| Why couldn’t the bicycle stand up by itself? Because it was two-tired. | 52 |
| Why did the frog call his insurance company? He had a jump in his car. | 36 |
| Why did the chicken cross the playground? To get to the other slide. | 33 |
| Why was the computer cold? Because it left its windows open. | 23 |
| Why did the hipster burn his tongue? He drank his coffee before it was cool. | 21 |
| Why don’t oysters give to charity? Because they’re shellfish. | 21 |
| Why did the computer go to the doctor? Because it had a virus. | 20 |
| Why did the banana go to the doctor? Because it wasn’t peeling well. | 19 |
| Why did the coffee file a police report? Because it got mugged. | 18 |
| Why did the golfer bring two pairs of pants? In case he got a hole in one. | 13 |
| Why did the man put his money in the freezer? He wanted cold hard cash. | 13 |
| Why don’t seagulls fly over the bay? Because then they’d be bagels. | 13 |
| Why did the chicken go to the seance? To talk to the other side. | 11 |
| Why was the belt sent to jail? Because it held up a pair of pants. | 11 |
| Why did the chicken cross the road? To get to the other side. | 7 |
| Why did the computer go to the doctor? Because it had a byte. | 6 |
| Why did the cow go to outer space? To see the moooon. | 6 |
| Why did the man put his money in the blender? He wanted to make liquid assets. | 6 |
| Why don’t skeletons fight each other? They don’t have the guts. | 5 |
| What do you call an alligator in a vest? An investigator. | 5 |
Future of Computational Humor
The study reveals that, compared to previous large language models, ChatGPT represents significant progress in the realm of computational humor. Jentzsch and Kersting plan to carry out similar studies with the recently released GPT-4 models and equivalent open-source models such as LLaMA.
AI NewsTime