In the Battle Between Bots and Comedians, A.I. Is Killing

Last month, in the crowded back room of a bar in Williamsburg, Brooklyn, the fate of humanity hung in the balance.

Or at least that’s how Matt Maran, a bro-y comic from Queens, portrayed it. He was bidding for sympathy during what was billed as the first roast battle pitting artificial intelligence against a human comedian.

It didn’t work. Maran lost the crowd early with a joke that riffed on the idea that women aren’t funny. His opponent was a ChatGPT-powered version of Sarah Silverman, the comic who, as it happens, had sued the developer behind that chatbot for copyright infringement earlier in the week. On a screen nearby, her head shook back and forth. “Why did the human stare at the glass of orange juice?” it asked in a close approximation of her girlish voice. “They were trying to concentrate.” Then oddly, it proclaimed: “Roasted!”

Neither side was getting big laughs, but the A.I. was more unflappable, moving from quip to quip with the pace of a metronome. Some of its jokes were simple similes (“You’re as edgy as a butter knife”) and at least one didn’t make sense, but the bot drew blood with a punchline that used details from Maran’s jokes: “You’re from Long Island and you lost your virginity to a prostitute. Talk about starting from the bottom and staying there.” Maran looked defeated even before the image of Silverman was replaced by one of him. Then he got roasted by the digital version of himself. Humanity lost every round.

It’s tempting to conclude that this defeat was an existential sign of the looming robot takeover of comedy, but the face-off felt closer to your grandfather playing computer chess than to Garry Kasparov vs. Deep Blue. But the event did drive home an important point about one of the many current anxieties surrounding A.I.: If the human lock on humor is coming to an end, the blame will lie with our complacency as much as technological progress.

Until recently, comedy has been seen as so quintessentially human that it was assumed A.I. would kill humanity before it would at a club. But since the rise of large language models like ChatGPT less than a year ago, this common wisdom no longer applies.

A.I.’s potential displacement of humans has become a central issue in the labor disputes in Hollywood. Jimmy Kimmel has told jokes written by ChatGPT on his show in February, and once-cautious computer scientists are now predicting it will only be a matter of years before robots are regularly generating professional comedy. The accomplished comic writer Simon Rich appeared shook after using code-davinci-002, a bot not available to the public. When he asked it for a parody headline for the conflict started by Russia, it responded: “Experts Warn That War in Ukraine Could Become Even More Boring.” Rich said he didn’t think he could beat it and “and certainly not instantaneously, for free.”

There’s no question that A.I. can think faster than any comedian can and study the mechanics of a joke with more agility than ever before. And even if irony and tone can still be challenging, its sense of humor will only improve. It’s worth recalling that in the 1990s, supercomputers lost to chess grandmasters before they started winning. But comedy is not chess. And whether A.I. can intentionally generate truly funny art is as much a philosophical as a technological question.

THE MOST COMMON CASE AGAINST MACHINES is that by design, they don’t come up with new ideas so much as collect and synthesize old ones. And comedy, whether it’s Tim Robinson, “South Park” or Ali Wong, depends on novelty and surprise.

Tony Veale, a computer scientist who wrote a book on comedy and A.I., “Your Wit Is My Command,” is impressed with new large-language models’ ability to imitate genre and voice, analyze and generate metaphors, explain itself and even admit mistakes. He’s bullish on computers making professional-level jokes in five years and when asked about originality responded that A.I.’s process isn’t any different from that of young artists. “Many comedians, such as Eddie Murphy and Jerry Seinfeld, trained themselves by repeatedly listening to and repeating Bill Cosby’s early comedy albums,” he wrote in an email. “We all learn from those we aim to emulate and transcend.”

He’s got a point. Plus: Much comedy doesn’t get that far past the imitation stage. People like old jokes. Sitcoms and stand-up are often derivative. Topical comedy often leans on formulaic phrasing and predictable rhythms. It’s why the form most vulnerable to disruption (and displacement of jobs) might be late-night television, featuring hosts delivering sets, which comedians typically take months or years to develop.

On the cutting edge of this work is Joe Toplyn, who studied engineering and edited The Harvard Lampoon before writing jokes for “Late Night With David Letterman,” where he helped come up with the concept of throwing watermelons off a five-story building (a very human idea). Toplyn has created a bot called Witscript that takes a headline or thought and spits out three jokes, then picks the best one.

In a demonstration at a conference last September, he presented two examples of what he considered “human-level” jokes that he claimed would make you chuckle if delivered by a friend. In one, he put in this prompt: “Some casinos are using fine-arts exhibitions to bring in new business” and Witscript responded: “I lost my Monet but gained a Jackson Pollock.” In another, Toplyn typed in “I’m thinking about growing a mustache” and the punchline was: “Tom Selleck called — he wants his look back.”

The only thing funny about these is their corniness. And yet, how confident are you that they aren’t a quick rewrite away from being ready for Jimmy Fallon? Toplyn was one of the head writers of “The Chevy Chase Show” (1993), one of the most infamous flops in the history of television, so he knows how much quality on late night can vary.

Comics have always been quicker than other artists to experiment with technology — Whitney Cummings had an eerie robot version of herself tell jokes on a Netflix special in 2019 — and most of the half dozen or so I talked to who did use some form of A.I. to make jokes seemed underwhelmed. Bill Oakley is a veteran of “The Simpsons” who, along with Josh Weinstein, produced many classic episodes, including “Who Shot Mr. Burns?” He asked a bot to write an episode in their style. The result, about a bees infestation, was “on the level of a seventh-grade fan,” he said.

Chris Onstad, who writes the comic Achewood, had more success, in part because he was involved with engineers in creating a bot that would issue an advice column in the voice of one of his characters. But he still sees A.I. as more of a tool than a threat. “It’s coming for fan fiction and pulp,” he told me during a video call, “but not high-quality literature.”

The problem is we live in an age that delights in fan fiction and pulp and too often overlooks literature. The internet replaced some of the power of human gatekeepers with algorithms — essentially a simpler form of A.I. — that recommend content for you on Spotify or YouTube. In media and culture today, a greater premium is placed on producing work quickly and in volume, an approach that already benefits computers more than humans. As swagger has moved from Hollywood to Silicon Valley, and English majors have become increasingly scarce, our culture has lost some confidence in the ineffable, essential qualities of human creativity.

FROM PLATO TO HOBBES TO FREUD, intellectuals have long offered theories of comedy. But over the past generation or two, there’s been a dramatic rise in the number of people treating art as a problem to be solved, a mystery to figure out and teach. An industry of gurus and critics and educators breaks jokes down into component parts. Upright Citizens Brigade was a hit for a time not only because its shows built on the success of their classes, but also because it created a system of comedy that leaned more on structure and rules than its improv predecessors did. Revealingly, in an interview, one of their co-founders, Ian Roberts, compared the U.C.B. approach to “the principals of physics.”

Every artistic field now seems to have its Robert McKee, whose popular screenwriting classes teach genre, plot and character with definitiveness. Toplyn was part of this wave, publishing a book on how to write late-night jokes. He broke down monologues, desk pieces and parodies with practical theories on how they work. And he used these conclusions to train his bot.

There is insight to be gleaned from Toplyn, McKee and the U.C.B. But at the heart of their entire enterprise is a lie, or at least a simplification. No theory of comedy that I have read captures it all because art can’t be boiled down to a recipe. And jokes, which include everything from puns and pratfalls to shaggy dog stories and satirical barbs, are especially varied and complex. As the cartoonist Saul Steinberg put it: “Trying to define humor is one of the definitions of humor.”

In the introduction to his new book of poems produced by A.I., “I Am Code,” Simon Rich describes showing the work to the poet Eileen Myles, who finds it badly derivative. There’s already a program producing formulaic poetry, Myles tells Rich, before quipping: “It’s called the M.F.A.”

There are elements of math to comedy and poetry (iambic pentameter requires an ability to count), but we must not reduce them to that. The subconscious, the source of much creativity, cannot be mapped so neatly. And the closer you look at great art, the more you realize that parts of it cannot be entirely explained or predicted. This is especially true for comedy. The funniest aspects are often those that resist logic. A.I. will figure out sarcasm before silliness.

To use the most popular blockbuster of the summer, Weird Barbie is the funniest one. That said, Stereotypical Barbie gets a different kind of laugh, one that might help explain that rout at the roast battle. Rigid characters trying and failing to escape their mechanistic situation is classically funny. (Think of Charlie Chaplin trapped in the gears or Lucille Ball at the chocolate factory.) Henri Bergson, one of the first great modern philosophers of humor, who was wise enough to reject “imprisoning the comic spirit in a definition,” saw comedy as a corrective to Industrial Age automatism. He believed that we laughed as a response to people acting like machines.

Bergson saw comedy as distinctly human. That may be in part because he understood that it’s deeply, inescapably social. We not only laugh more in groups, but also what we find funny depends on who is telling the joke. A punchline about a car crash will be hilarious to one person and offensive to another. Artificial intelligence can come up with jokes, but it can require emotional intelligence to make them work. What makes people crack up is not just the joke but also the connection with a human consciousness telling it.

Maybe audiences could make a similar connection with a sophisticated computer of the future, but the Catch-22 for A.I. humor is the further it gets from seeming human, the more unfunny it seems, but the closer the imitation, the creepier it becomes as it falls deeper into the uncanny valley. After the robot revolution, what may save human comedy from irrelevance is our own deep-seated tribal biases. We like to laugh at our own.

COMEDY BY ROBOTS is more likely to succeed as a tool behind the scenes — or in disguise. One of the more successful recent attempts at integrating A.I. into comedy suggests as much.

Curious to see if A.I. could be funny, Jason Woliner, best known for directing “Borat Subsequent Moviefilm,” built a crude cowboy robot named Dale using speech-synthesis software and ChatGPT. He gave it a grizzled character, training it to speak in monologues filled with folksy metaphors and old-timey locutions like “I reckon.” Woliner tried him out at a party at his house where guests could ask Dale questions.

“It was kind of captivating,” he said of the conversations. “The A.I. was filtered through this character, so it didn’t have that creepiness. It was like a parlor trick from the 1800s or something.”

This was a warm-up for a July show at a Los Angeles theater based on the conceit that Woliner was presenting his own career retrospective. Dale delivered a scripted introduction and then interviewed his creator, receiving some prompts backstage from Eric Notarnicola, a collaborator on this project. About half of what Dale said came from the A.I., half was written. But importantly, Woliner said, he didn’t tell the audience he was using A.I.

“That would be a cheat,” he explained, leaving the crowd to wonder if the voice came from a person or was prerecorded. Dale got laughs. Some from the rambling Old West oddness of its style, but not only that. After listening to Woliner talk about working with Jim Belushi, Dale killed when he quipped, “Did you ever think the wrong Belushi died?”

Woliner sounded almost uneasy confessing that this experiment made him hopeful about A.I., pointing out that losing jobs is a serious concern and merits the striking writers’ fight for a new deal. Human creativity, he believes, is not replaceable, but it is able to integrate this new technology in exciting, creative ways. With Dale, he found a “a new tone” to play with and plans to continue to work with it.

The conversation about A.I. today gravitates toward doomsday scenarios, but consider the utopian outcome, that bots don’t replace comics but become useful tools. That feels modestly realistic. The ideal, it seems to me, is that artificial intelligence gives comedians incentive to abandon the more generic forms of their work, the predictable and hack, and inspires them to be more eccentric, personal and original. Competition from increasingly clever computer programs will force artists to not only rely more on intuition than imitation, but also to think harder about what makes them, and their work, distinctly human.

Our advantage in the comedic battle with machines is that human flaws are always funnier than any kind of perfection. What if A.I. pushed comedy to become more ambitious? A critic can dream.

Jason Zinoman is a critic at large for The Times. As the paper’s first comedy critic, he has written the On Comedy column since 2011. More about Jason Zinoman

Source: Read Full Article