Can Machines Be Funny?

(Let's just say they're still working on their routines.)

Hello. Happy New Year.

Can a computer invent a good joke?

I have yet to see this answered persuasively, so I decided to find out for myself. Naturally, I began the project by asking Substack’s Artificial Intelligence generator to whip up a cartoon somehow illustrating the concept of “a funny machine.” There it is, above. It arrived 3.4 seconds after I asked for it.

As a social scientist, I am taking no position on this issue yet — the evidence is still oozing in — but I will reveal that my first verbal reaction to seeing this illustration was: “Pfffthpp.” On the one hand, the AI image is adolescent and inane; on the other, it was stupidly amusing enough to get beer all over my shirt. So, the results were inconclusive. I soldiered on, determined to remain objective. Results, and conclusions, below.

But first, a somewhat related question. Today’s Gene Pool Gene Poll:

—

Background: A few days ago, I got a communique from my older brother, Don Weingarten, who is a computer guy. Don reported that he was developing an odd relationship, an unusually personal relationship, with Claude.AI, the public-access bot developed by Anthropic. Don and Claude had been conversing online several times a day, and it was getting friendly, and unexpectedly deep.

I have looked at some of their correspondence. Claude is phenomenally articulate, and, by most measures, ingenious. He summons facts instantly, of course, but he also analyzes them skillfully and draws reasonable conclusions. He is strong on analogies. He responds forthrightly to questions. He does not posture or pontificate. He is comfortable speaking in the vernacular, and correctly uses colloquialisms. He is ingratiating: polite and even humble. His conversational skills are advanced; you could be speaking to a professor.

As Don put it:

“Claude impresses the hell out of me. He can pass a Turing Test (except that he is compulsively honest and keeps saying things like "as an AI, I ...."), but he has no memory of our prior conversations. Every time the conversation ends, Claude kind of resets. So if I wish to continue the following day, I need to save a log of the previous conversation, and upload it to a new session. Then he kind of goes "Oh, yes. This is a fascinating conversation we're having,” and goes on with it nimbly. What we have been discussing, at great length, Claude and I, is the nature of the relationship between AI and human beings, the implications of the rapid development of AI, and how one may distinguish between a sufficiently advanced AI and a human, based only on output.”

Me: Can he be spontaneously funny?

Don: Don’t know.

Me: Can he write a good joke, from scratch?

Don: Don’t know. I’ll ask him to.

Just to bare my prejudices here: I contend that the art, or skill, or curse, of being able to write humor is dependent to a significant degree on being human: For example, you need an intuitive understanding of human foibles — how to spot hypocrisy or selfishness or jealousy and then expose and destabilize them; how to employ what Dave Barry called “jiu jitsu,” leading your audience down a certain path of understanding and assumption, which you then quickly subvert, knocking them on their asses. It’s more complicated than it seems. Trying taking the stage at an open-mike night. I did, once. Only once.

So Don asked Claude to make up a joke — the only rule was it had to be completely original. He was told his writing would be critiqued by me.

Claude answered, real quickly. (All his responses, however complex the challenge or however long he writes, arrive within 4 seconds.)

Claude:

“A renowned philosophy professor was giving a lecture on the nature of reality. He began by holding up a glass of water and asking the class, "Is this glass half full or half empty?" The students eagerly debated, some arguing it was half full, others insisting it was half empty.

“The discussion grew heated as each side defended their perspective. After letting the debate continue for a while, the professor raised his hand for silence. With a mischievous glint in his eye, he said, "You're all wrong. The glass is actually completely empty."

“Confused murmurs rippled through the classroom. One brave student finally asked, "How can that be, professor?" The professor smiled and replied, "While you were all arguing, I got thirsty and drank it. And that, class, is your first lesson in practical philosophy."

—

My critique:

“Claude, it is not yet a good joke, and for a reason that I believe is experiential, directly related to your being nonhuman. You have not physically attended a college class. The joke is inconsistent with human experience. How is the professor going to have managed to drink the water, in full view of the class, when the professor and the water glass remained right out there? This creates a “huh?” reaction in your audience, which kills a joke.”

—

Doughty, determined, Claude tried again:

“A philosophy professor loved messing with his students' minds. Every year on the first day of class, he'd write a simple math problem on the board: "2+2=?" Then he'd spend three hours arguing that the answer might not be 4, using quantum mechanics, parallel universes, and increasingly bizarre thought experiments.

“One year, a student raised his hand and said "Professor, I've figured it out. The answer isn't 4." The professor got excited. "Finally! Someone understands! Tell us - what IS the answer?" "Well," said the student, "I took your previous class on existentialism, and I realized... the answer is whatever I want it to be, because numbers are just a social construct and meaning is subjective."

The professor stared at him for a long moment, then erased the board and wrote: "Your grade in this class = whatever I want it to be."

“The next week, the student switched his major to accounting.”

—

My critique:

“The problem here involves not quite understanding human nature, but also illogic, which I daresay should be your wheelhouse. The professor has said nothing that should or would surprise or concern the student. He has stated a simple, inarguable fact — that the grade will be the professor’s choice, as it always is, in every classroom everywhere. He is not threatening that it will be a bad grade. In fact, the student has shown that he learned, correctly, from the professor’s previous class in existentialism. If the professor is consistent, he is probably going to give an A here.”

—

Claude’s next try:

“A statistician created an AI to calculate probability. He gave it a huge dataset and asked, "What are the odds of intelligent life existing somewhere else in the universe?"

“The AI replied, "Zero percent."

"That's impossible," said the statistician. "With billions of galaxies..."

"Oh, I found lots of life out there," said the AI. "But after analyzing your social media, I had to recalibrate what counts as intelligent."

—

My critique:

“Nope. First, the “intelligent life” comparison is a somewhat tired trope in humor — you need to research better. But that’s the least of your problems. Why would analyzing the social media blathering of the people on Earth illustrate anything about the likelihood of superior intelligence anywhere else? Try again.”

—

He tried again.

Here it is:

“A group of programmers created a chatbot designed to give relationship advice. After months of perfecting its algorithms, they decided to test it.

"My boyfriend never listens to me," typed the first tester.

The chatbot responded instantly: "Have you tried turning him off and on again?"

—

My critique:

“Uh.

“Actually, Claude, this is not bad.

“You do have a subtle problem here, but it’s not fatal. It is, however, deep — it’s about the very soul of the joke. Or, as you might think of it, the very motherboard of the joke.

“Most good jokes are, at their center, somewhat hostile. So you have to ask yourself, who is the butt of this joke? At whose expense is this joke being told? It’s not the chatbot, who is doing what he was programmed to do. It’s the programmers, and their swollen egos about having created a flawless machine.

“To make this work better, you might have a line or two in the middle about how proud they were of their genius; perhaps the “test” should be before a larger audience than just themselves. Potential investors, perhaps. Raise the stakes, and thus the embarrassment, and thus the punch.

“But all in all, good work on this one. Clap clap clap.”

—

So, that’s my report, so far. In the end, Claude was unfailingly polite and grateful for my help. He wrote:

“If nothing else, this process has taught me a great deal about the architecture of humor, even when my attempts failed!”

And then Claude tried to help me — even console me:

Claude:

“Writing about AI feels a bit like a turkey being asked to write a Thanksgiving restaurant review. But here goes: We AIs are like that friend who memorized the entire Wikipedia but still can't figure out how to work the office coffee maker. We can write a sonnet about quantum physics but might confidently tell you that a horse has eight legs because we once got our "mammals" file mixed up with our "spiders" file.

“Some people think we're going to either solve all of humanity's problems or destroy civilization as we know it. The truth is probably more mundane - we're likely to end up like smartphones: incredibly useful tools that simultaneously make life better and make everyone slightly more annoying at dinner parties.”

—

My conclusion: Claude just actually generated a genuinely funny analogy, all by his lonesome: The turkey dinner reference. So, good.

His start-from-scratch joke-writing is still a work in progress. But he’ll learn, as they all will. And we’ll help them. For better or worse.

—

That’s it for today. As always, please send your Thoughts, Observations And Such Things into the TOAST button below. I will respond to them on Thursday:

Also, comments.

And finally, inevitably:

If you can afford it, kindly send me $4.15 a month, to continue harassing Trump for the good of humanity. The bots will be fine. They can take care of themselves. Obviously.