Why isn’t AI funny?
Oh, the humanity!
Humor is arguably the most nuanced form of human communication. It’s about finding common ground between individuals and cleverly elevating it to the point of provoking an exceptional response — laughter. While a well-crafted message could play on common tropes to incite anger or fear, prompting laughter is a more delicate art. It’s a complex mix of timing, context, shared understanding, and surprise which, ironically, creates a binary problem — a joke either lands or it doesn’t.
Humans have vast mental libraries of cultural references and linguistic nuances to draw upon when hearing or telling a joke.
Artificial Intelligence Can Now Craft Original Jokes — And That’s No Laughing Matter, TIME Magazine
Currently, we might find AI’s attempts at humor genuinely funny, but it’s often because of the sheer absurdity of what the AI comes out with rather than an explicit attempt at humor. We could, of course, prompt the AI to deliver low-brow humor, filled to the brim with the equivalent of fart gags. But — can AI master the timing? Or know when a joke has been overused? Or even, knowing that it’s been overused, push it even further for comedic effect?
Does funny have a formula?
One of my favorite British comedians, who I’ve watched perform in London, is Stewart Lee. What struck me about his stand-up routine was how analytical and self-aware it was. He’ll often describe to the audience the structure of the joke he’s about to tell, tell it, and yet it remains hilarious. Lee, Rowan Atkinson and Jerry Seinfeld are all comedians who seem obsessively analytical about comedy as an art form, but that doesn’t prevent them from being incredibly funny — which seems counter-intuitive.
Given there has been so much academic research into the analysis of comedy, you might think that you could write down the ‘rules of comedy’ and it would be a perfect recipe for an AI to follow. In fact, late-night TV writer Joe Toplyn thinks he’s done exactly that with his patented system Witscript.
And yet — I think we remain skeptical. Comedy, like any other art form, risks becoming derivative without originality, and a formulaic approach does not seem conducive to originality. This brings us to a potential key limitation of AI in comedy: the lack of self-reflection. The process of self-reflection, essential for originality, is something that AI currently lacks. It gives us the best approximation of what we’ve asked for based on all of the existing material it has been given.
Could an AI be taught self-reflection? Could it carefully decide and consider its influences, develop a unique sense of humor and then refine its own brand of humor given feedback from its public ‘audience’?
That remains to be seen, but again I believe this is an artificial general intelligence (AGI) problem. If it can come up with a unique voice that isn’t just random and arbitrary, but follows a consistent internal logic & reason to produce something that is emotionally resonant — is that not the same thing as true self-awareness?
AI claps back
I like to give ChatGPT a chance to dispute my posts, which I usually generate as a talking head video, but I’ve run out of credits for this month — so instead I asked it to fact-check my work and its response included this…
The idea that AI ran out of time to do its homework? — now that’s funny!
