Artificial intelligence fails comedians' test

While artificial intelligence technology has been the most prominent development of recent years, chatbots are not yet perfect. They can provide inaccurate information from time to time and lack a sense of humor.

Artificial intelligence chatbots are becoming effective in many areas of our lives, but they still have a long way to go. Not only do they sometimes provide inaccurate information, but now it turns out they don't have much of a sense of humor.

In a study published earlier this month, Google DeepMind researchers concluded that AI-powered chatbots are not funny.

Last year, four researchers from the UK and Canada asked 20 professional comedians to try OpenAI's ChatGPT and Google's Gemini using AI in their work. The comedians, who were anonymized in the study, played with large language models to write jokes. They reported a number of limitations. The chatbots produced bland and generic jokes, even after prompting.

Participants also discovered that the overall creative abilities of chatbots are limited and that humans should do most of the work.

One participant who stood out for dark humor told the researchers, "He didn't write me dark things because he thought I was going to commit suicide, in a way."

It was also striking how strongly self-censorship manifested itself. Participants reported that it was difficult to get her to write about people other than heterosexual white men.

Tech companies are keeping a close eye on how chatbots talk about sensitive topics. Earlier this year, Google AI's image-generation feature came under fire for refusing to take photos of white people. It was also criticized for making the mistake of depicting historical figures like Nazis as people of different races. A few weeks later, in a blog post, Google management apologized and discontinued the feature.