Artificial Intelligence is smart – it sometimes seems smarter than us – and it can even tell a good joke. But to be truly funny requires a different kind of intelligence.
A 2017 study [PDF] found that funny people have higher IQs. Researchers noted that humor was tied to cognitive and emotional abilities. From the study:
“Humour processing is a complex information-processing task that is dependent on cognitive and emotional aspects, which presumably influence frame-shifting and conceptual blending, mental operations that underlie humour processing.”
That, as I see it, puts humor at the bleeding edge or beyond current AI capabilities.
I was curious, though, if an AI could handle the demands of true comedy. Granted, I’m no comedian or actor, but I do understand the dynamics of certain kinds of comedic pursuits.
mprove, for instance, has, at its core, the requirement that each scenario is met with “Yes, and…” to keep the ball rolling.
If you’ve ever been to an improv comedy set, you know that this is how it works. The audience lobs a scenario at the actors and they run with it. If anyone says “no” or refuses to engage or offer an opening for their counterparts to saunter through, the set ends. So does the comedy.
There’s more to it than that, though. If you’re not funny or if you’re playing deep in the absurd, the whole effort might fall flat.
I had no idea whether ChatGPT understood any of this, let alone the singular act of improv.
However, when I asked ChatGPT 4o’s voice AI if it knew anything about improv, it enthusiastically told me, “Yes, improv involves spontaneous, unscripted performances where participants create scenes and dialogue on the spot. It’s all about creativity and quick thinking. Have you tried it?”
I told ChatGPT I had not, but asked if we could do so now.
ChatGPT was up for it and asked me if I had any ideas for a simple scenario. Instead of waiting for my response, though, ChatGPT quickly cooked up a fairly hackneyed premise.
And…scene!
“How about a scene where we’re two astronauts discovering a mysterious object on an alien planet? You can start us off,” asked ChatGPT
With the pressure instantly on me, I choked up a bit. After a beat, I said with all the drama I could muster, “Whoa, what is that?”
“It looks like some kind of glowing artifact. Should we take a closer look?” replied ChatGPT
I thought, “This is getting good. How do I steer it toward comedy?”
“Yes, I’m going to pick it up, but I’ve got to put on my gloves first.” I know, not my best effort.
This continued with us examining the object, reading a message on it, dropping it, tossing it back and forth, and using it as a soccer ball. It finally blew up.
“Yikes,” said ChatGPT, “Definitely not our best idea.”
Hilarious, right?
No, not funny at all.
Maybe it was the scenario.
Take 2
I decided to use some real-life inspiration: Some months ago, I accidentally took someone else’s sandwich from the office fridge (it was wrapped in tinfoil like mine) and took multiple bites before I realized it was someone else’s sandwich.
I spelled this out to ChatGPT, which answered, “Oh no! What do you do next?”
I forgave the “no” in there but still wished I’d been fed a better line.
Steering toward comedy and tension seemed like the best course of action: “So I’m looking at the sandwich and I’m panicking, and I look over at the guy next to me and I say, ‘This isn’t my sandwich.'”
ChatGPT: “He looks at you with surprise and says, “Really? Then whose is it?”
I said it looked and tasted weird, and, with this, ChatGPT came as close as it could to pure comedy:
“He peeks and laughs, ‘That’s my special tofu and jalapeño sandwich. Not for everyone!”
Oh, the hilarity.
With as much absurdity as I could muster, I declared that there was something else in this sandwich and that it was not tofu but a “mystery meat,” and it might be “Barry’s ‘Mystery Meat Sandwich.’
ChatGPT ended by telling me we’d track down Barry and return his sandwich.
Whose AI line is it, anyway?
None of this was funny, and I guess I have only myself to blame. I struggled with the scenarios. They weren’t intrinsically funny, and we could not seem to make them more interesting.
ChatGPT viewed the exercise as an improv acting class and never really attempted comedy. Neither my absurd suggestions nor delivery inspired it to do otherwise.
There is something, though, encouraging about this experiment. There’s a lot of concern about AI taking creative jobs. We know it’s already adept at generating realistic photos and videos (complete with audio) with nothing more than a prompt suggestion.
The quality of that output does hinge on the prompt details. Even if you create, say, a feature-length Gemini Veo 3 video, it won’t be very interesting unless you also feed Veo 3 a complete script.
In a similar fashion, ChatGPT’s intelligence has yet to reach a point where it can generate funny out of thin air. Jokes it shares are culled from around the web. Absurdist comedy might require an Artificial General Intelligence that AI has yet to achieve.
Not that it won’t get there, but for the moment, I think improv comedians are safe.
