ChatGPT has capacity for humor but struggles with originality says study

(Image Courtesy: Nurphoto via Getty Images)

A new study conducted by researchers Sophie Jentzsch and Kristian Kersting suggests that ChatGPT struggles with generating humor.

The study conducted for Cornell University said that while OpenAI's generative AI chatbot shows some promise, it has problems with originality, and often repeats the same joke.

Also Read | Why are India unable to win ICC tournaments? ChatGPT gives the reasons

“We discovered that more than 90% of the generated samples were the same 25 jokes," the study says. "This recurrence suggests that these jokes are not originally generated but are explicitly learned and memorized from the model training.”

Moreover, while the system seems to understand and explain jokes, it also has a tendency to make up fictional explanations. Typical characteristics of a joke found in some sentences also led the AI to classify them incorrectly as humor.

Also Read | What India needs to build an OpenAI rival?

The study concluded that the responses were likely learned during training but the AI system itself made no attempt to create original jokes using the format. It was essentially regurgitating pre-learned patterns found in jokes but lacked the creativity to come up with something original.

Humor is an essential part of human communication, and while generative AI systems can mimic human emotions, it doesn't seem to understand what makes it special just yet. This also has implications in fields like digital marketing or stand-up comedy where AI might fall short.