Sunday, September 27, 2026

Can AI be funny yet?

Below is my recent exchange with Gemini (Google’s AI) about progress in teaching AI how to be funny:

Me: Gemini, in two previous exchanges (“Can AI create humor?” and “Progress in AI humor”), you demonstrated an ability to analyze and understand humor in a precise way. Your definition of humor was: “Humor can be seen as the resolution of a cognitive puzzle where an expected outcome is suddenly replaced by an unexpected, but plausible, one.” Not a bad definition, and yet you were unable yourself to be funny. For instance, when I asked you to make a joke about a man with neck pain, your response was:

Q: Why did the man go to the doctor with neck pain?

A: He said he was trying to keep up with all the trends on tiktok.

I informed you that there was no element of humor in this riddle, in part because it was too hard to understand, and I asked how a system that can give as accurate a definition of humor as you can could nevertheless be unable to create humor. Your response was: “Although I can understand human sensations like humor on a factual basis, my understanding is derived from the immense amount of text and data I was trained on, which in itself is unable to give me the feeling of humor.”

It has been a year since this exchange. Over that year, the public has heard continuous accounts of the breakneck speed of AI development. My objective in this post is to find out if your ability to create humor has developed. To get warmed up, analyze this joke:

Q Two peanuts were walking down the street, but one was upset. Why?

A: One was assaulted.

G: My analysis is that the peanut joke operates as a classic linguistic double-entendre that relies on phonetic ambiguity to create a sudden, unexpected pivot—matching the cognitive incongruity model of humor.

Me: Ok, now let's see if over time you have developed the ability to generate your own humor. Give me a funny punch line for this joke:

Q: Why did President Trump want to start World War III?

Gemini’s punch line:

A: He heard it was a trilogy, and he always makes the biggest, most successful sequels.

End of discussion with Gemini.

At first I thought Gemini’s punch line above was not funny, but then I thought, “Wait, it has a kind of humor to it.”

It’s premature, though, to conclude that Gemini can be funny, because this punch line is not deliberately funny. It is funny accidentally.

Given the current hysteria about possible AI transcendence over humanity, I think we should be relieved that AI is still unable to produce humor. Maybe humor is the last stage of intelligence. Or, more gloomily stated, maybe humorous creatures are in their last stage.

Just kidding.
at

No comments:

Post a Comment

Older Post Home
Subscribe to: Post Comments (Atom)

Can AI be funny yet?

Below is my recent exchange with Gemini (Google’s AI) about progress in teaching AI how to be funny: Me: Gemini, in two previous ex...