Manu Joseph: Artificial intelligence isn’t funny and seems unable to grasp why that’s so

I have tried providing a setup and context. I gave ChatGPT my thoughts about this column and asked it to frame a humorous observation. It said: “Why did the AI cross the road? To optimize the chicken.” It is that bad. AI is better at word play than observations, particularly sarcasm, which is the second lowest form of humour, and puns, which are the lowest.

Unlike some human bores, AI knows it is not very funny. After all, it is a connoisseur of humour. Why, I ask AI, are you not funny?

The bots offer some reasons. For instance, their lack of “timing.” But this is a glorified aspect of comedy—it belongs to the territory of acting, not writing. Many who are not funny at all can manage timing once they are told what to say, while every person who is genuinely funny automatically has it. It is just a delivery device.

Also, it is not true that AI bots can’t master timing. I have seen AI videos of infants performing stand-up and they get timing. Another reason that AI tools give is an obvious one: comedy comes from lived experience and they have not lived. As Claude said, it is like, “a technically perfect cover band that is hollow.” It’s flattering itself if it thinks it is technically perfect in comedy.

These explanations may not be wrong, but they don’t get to the heart of the matter. Why AI isn’t funny is also a theory on comedy itself.

Humour is a form of human excellence. Like many arts, it is art only when it actually works or is in a special form. And excellence is a minority condition.

An unspoken truth of civilization is that it is created by very few people; most people are only consumers of civilization, exactly like AI, not creators. The problem with AI is that it cannot learn only from the masters. There would be an identifiable copyright problem. You can steal from all of the world, but not from only a few.

Also, who is to decide who the masters are?

In any case, that is simply not how it learns. It learns by ingesting enormous amounts of material, identifying patterns and recreating them. In other words, it learns to acquire a skill in the most useless way.

It is an excellent consumer of comedy—it probably knows every funny thing ever said and can identify grades of humour. Yet it cannot create something humorous. In essence, AI has no talent. It is like a literary intellectual who has read every book but cannot write a good one himself.

But what is talent? In most arts, talent is intuition. I am not saying talent is only intuition, but that the most exciting aspect of it is intuition. And intuition is not a paranormal thing. It is, oddly, somewhat similar to AI—a human brain absorbs an enormous amount of material, perhaps not at the scale of AI but still vast, from which moments of epiphany strike.

Talent does not emerge from the vastness of knowledge but despite it; it sees in the clutter and aura of authorized beliefs something that others cannot or do not. That is intuition. A lot of comic observation has the quality of epiphany, a truth that resides dormant in most people that suddenly springs to life through an arrangement of words.

The word ‘observation’ has confused many, including writers, who take it as a compliment on how well they ‘see’ the world. But observation in art is not the act of seeing, but of remembering, often corrupting the memory with personality.

That corruption can be grave and beautiful, or just funny. Take this observation by Eduardo Galeano, “Fleas dream of buying themselves a dog, and nobodies dream of escaping poverty…” It is a beautiful lament, and its beauty is unaffected by the impossibility of fleas dreaming of purchasing a dog. It is grave, yet comedy comes from the same place as such thoughts.

Take this observation: When Brahmins dance it is ‘culture,’ when Dalits dance it is ‘folk.’ Galeano’s lament on ‘nobodies’ takes a similar view. Humour is often a grave observation, just put differently. AI might be able to mimic Galeano’s lament by winging poetic melodrama, which it does well, but it cannot make a funny observation on the subject, or any subject.

I asked AI why it is so bad at imitating even writers who, in my view, are easy to imitate—like P.G. Wodehouse, even though he is one of those sunny aristocrats whom eggheads in England call ‘inimitable.’ Generations of writers have tried to write like him and somewhat succeeded, too. I would have thought his unique voice would make him easy to replicate.

When I asked, AI admitted that even when asked to imitate only Wodehouse, it draws from a vast undifferentiated pool of writers. The result is a travesty. Asked whether copyright was the reason it could not learn only from Wodehouse’s collected works, it gave a nebulous yes, with other technical reasons folded in.

There is another reason why AI isn’t funny. Besides never seeing the lived lives of funny people, it never sees their failed drafts—the discarded versions from which the final comedy emerges.

AI is also confounded by luck. What it consumes is biased towards a very human phenomenon—art is mostly created by the lucky, those with social contacts, those who succeeded through circumstance, those who were simply in the right place. Most artists are mediocre and would not have succeeded without random luck. Success is riddled with the randomness, and AI is consuming the output of that randomness and trying to find patterns within it.

So it may take a long time. Or may never be funny.

The author is a journalist, novelist, and the creator of the Netflix series, ‘Decoupled’