According to a study, ChatGPT’s humor is not quite human yet
4 mins read

According to a study, ChatGPT’s humor is not quite human yet

ChatGPT is known for its ability to generate texts that reflect human conversations, making it a widely used tool for various industries including digital marketing.

However, a recent study questions the model’s ability to generate and understand humor, a key component of engaging and connecting with audiences.

The research, conducted by German researchers Sophie Jentzsch and Kristian Kersting, suggests that while ChatGPT excels in some areas, it has significant limitations when it comes to generating original humor.

Recycled Laughter: The Question of Originality

A study conducted by Cornell University seeks to answer the question, “How does an artificial intelligence model deal with humor?”

Researchers are examining the originality of AI-generated humor, ChatGPT’s ability to understand and explain jokes, and its ability to detect humor.

The research team states in their report:

“We more than figured that out 90% of the samples produced were the the same 25 jokes. This repetition suggests that these jokes were not originally generated, but were explicitly learned and stored from model training.”

The researchers concluded that these responses were likely learned and stored during training of the AI ​​model, suggesting a limited ability of the model to generate original humor.

In a report, the researchers describe the ten most commonly generated jokes, including classics like “Why did the scarecrow win an award?” Because he excelled in his field.”

The study not only reveals the complex ways AI handles humor, but also serves as a warning to people who want to use ChatGPT to create content with a humorous twist.

The implications for digital marketers who rely on AI for content generation are clear: while AI models like ChatGPT can reproduce pre-learned patterns to create jokes, they lack originality.

Despite the repetition, a small portion of the responses generated were unique. However, these were largely created by combining elements of various well-known jokes and didn’t always make sense.

Explaining the joke: Beyond superficial humor

The study also examined ChatGPT’s ability to explain humor, which required a deeper understanding of the joke’s structure and implications.

While the model could deconstruct and explain stylistic elements such as personification and puns, it showed its limitations when confronted with more unconventional jokes.

The team observed:

“ChatGPT has trouble explaining sequences that don’t fit into learned patterns. It doesn’t indicate when something isn’t funny or there’s no valid explanation.”

In cases where ChatGPT couldn’t recognize unfunny jokes, it created plausible-sounding explanations.

For marketers looking to engage their audience with nuanced humor, using AI alone may not produce the desired results.

Joke Recognition: Deciphering the Punchline

Beyond generating and explaining jokes, the research team tested ChatGPT’s ability to recognize humor.

They found that while the model could correctly identify jokes based on structure, wordplay, and subject matter, it failed to classify a sentence as a joke if it shared even one of these characteristics.

This underscores the model’s reliance on learned patterns and lack of a fuller understanding of humor.

What does this mean for marketers?

While ChatGPT has revolutionized the field of AI-generated content, this study advises caution when relying on the model for humor generation.

The study concludes:

“Although ChatGPT’s jokes are not regenerated, this does not necessarily detract from the capabilities of the system… However, whether an artificial agent is capable of understanding what it has learned is an extremely difficult question.”

As digital marketers look to AI to diversify and expand their content offerings, it’s important to understand the limitations and strengths of the model. In the realm of humor, human creativity is irreplaceable, at least for the moment.

The research team plans to conduct similar investigations on newly released AI models such as LLaMa and GPT-NeoX, promising further insights into the world of computer humor.


Featured image created by the author using Midjourney.