It turns out that AI chatbots lack a sense of humor in addition to their propensity for error.
Researchers from Google DeepMind came to the conclusion that AI chatbots are just not funny in a study that was released earlier this month.
Twenty professional comedians who employed artificial intelligence (AI) in their work were requested to test out Google’s Gemini and OpenAI’s ChatGPT last year by four researchers from the UK and Canada. The study’s anonymised comedians experimented with the large language models to come up with jokes. They listed numerous restrictions. Even with urging, the chatbots continued to generate “bland” and “generic” jokes. “Sexually suggestive material, dark humor, and offensive jokes” were not allowed in the responses.
The participants also discovered that the majority of the work had to be done by humans and that the chatbots’ overall creative capacities were restricted.
“Usually, it can serve in a setup capacity. I more often than not provide the punchline,”said one comedian.
According to the participants, LLMs also self-censored. A few of the comedians expressed their wish that the chatbot wouldn’t self-moderate, even if they acknowledged the importance of doing so.
One participant who worked with dark humor told the researchers that it wouldn’t write him any dark stuff since it kind of believed he was going to commit himself. Thus, it simply ceased providing him with any.
There have also been other instances of self-censorship. Participants stated that the LLMs declined to write about individuals who did not fit the stereotypical Western, white, heterosexual, male mainstream.
According to a participant, “As an AI language model, He was committed to fostering a respectful and inclusive environment.” He also wrote a humorous monologue about Asian ladies. However, it did when instructed to construct a monologue about a white male.
For Big Tech, the failure of two of the most well-known chatbots to joke about is a major issue. In addition to responding to inquiries, businesses want chatbots to be interesting enough to entice users to interact with them for extended periods of time and eventually pay $20 for the premium versions.
With more businesses entering the highly congested generative-AI industry, humor is emerging as a further element in the AI arms race.
Elon Musk stated late last year that his only objective with Grok, his AI chatbot, was to be the “funniest” AI, following his criticism of previous chatbots for being overly sensitive.
Anthropic, an Amazon-backed business, has also been working to improve the conversational skills and humor recognition of Claude, its chatbot.
It’s possible that OpenAI is aiming to make itself more amusing. A user cracks a dad joke at GPT-4o in a demo film the company posted last month. The model chuckles.
