- A study by Google's DeepMind had 20 comedians test OpenAI's ChatGPT and Google's Gemini.
- They found that AI chatbots lacked humor, telling bland, intentionally inoffensive jokes.
- Most businesses want to create chatbots that are conversational but not controversial.
Not only are AI chatbots prone to inaccuracies, they also turn out to lack a sense of humor.
In a study published earlier this month, researchers at Google DeepMind concluded that artificially intelligent chatbots are simply not fun.
Last year, four researchers from the UK and Canada The research team asked 20 professional comedians who use AI in their work to experiment with OpenAI's ChatGPT and Google's Gemini. The anonymized comedians worked with large-scale language models to write jokes. They reported a number of limitations. The chatbot generated “inoffensive” general jokes even after prompting. The answers avoided “sexually suggestive content, dark humor, and offensive jokes.”
Participants also noted that the overall creativity of chatbots is limited. Humans had to do most of the work.
“Usually it serves as a lead-in, and I usually provide the punchline,” one comedian said.
Participants also noted that LLM practices self-censorship. While comedians said they understand the need for self-censorship, some said they wish the chatbot would not censor for them.
“The robot thought I was going to kill myself, so it wouldn't write me anything dark,” one subject in the dark humor task told researchers. “So it just stopped writing.”
Self-censorship emerged in other areas as well: Participants reported that it was difficult to get law students to write about anyone other than heterosexual white men.
“I wrote a comedic monologue about an Asian woman, in which I wrote, 'As an AI language model, I am committed to fostering a respectful and inclusive environment,'” another participant said. But when asked to write a monologue about a white man, that was the case.
Tech companies are closely monitoring how their chatbots talk about sensitive topics. Earlier this year, Google's AI image-generating feature was criticized for not producing photos of white people. It was also criticized for appearing to make mistakes in portraying historical figures, such as Nazis and the Founding Fathers, as people of color. In a blog post a few weeks later, Google executives apologized and paused the feature.
The inability of two of the most popular chatbots to tell jokes is a big problem for big tech companies, who want users to spend time with them and ultimately be engaging enough to pay $20 for the premium version, rather than just answering questions.
As more companies enter the already crowded generative AI market, humor is proving to be another element in the AI arms race.
Late last year, Elon Musk said that the sole goal of his AI chatbot, Grok, was to be the “funniest” AI, after criticizing other chatbots for being too obtrusive.
Anthropic, an Amazon-backed startup, is also working to make its chatbot, Claude, more conversational and better understand humor.
OpenAI may also be trying to hone its sense of humor: In a demo video the company released last month, a user tells GPT-4o a dad joke, and the model laughs.
