Comedian Karen Hobbs was feeling more nervous than usual before going on stage one evening in June.
Hobbs, one of Britain’s best-known travelling comedians, is all too familiar with the grind of stand-up.
The term “travelling comedian” refers to people who perform stand-up comedy in bars, clubs, theatres and festivals. This type of show is notorious for its unpredictability and “forced laughter”.
Hobbs is a comedian who has fought this battle in some of the meanest venues in England.
On this special evening above Covent Garden Social Club in central London, Hobbs took to the stage not with his usual material but with a stand-up set written for him by artificial intelligence platform ChatGPT.
He also addressed the audience as follows:
“If you watch all this with a smile, we will all be unemployed!”
ChatGPT, the artificial intelligence robot developed by OpenAI, has been the first vehicle to bring artificial intelligence to the masses for two years.
However, when it comes to art, it is highly debatable whether artificial intelligence can be creative.
Large language models (LLMs) like ChatGPT process billions of lines of text collected from the Internet and other sources, extracting patterns and relationships between words and sentences, and so on.
Artificial intelligence tools also use this data to produce the most likely answers to a particular question.
As can be seen from here, artificial intelligence tools can only replicate information that exists in one form.
However, the result may also be a unique combination of ideas. Whether this counts as creativity is a philosophical question, and unfortunately there is currently no satisfactory answer.
Can a robot be funny?
“One way AI can make jokes is to do what any five-year-old child does, which is to repeat a successful joke they’ve heard or try to do an obvious variation on it,” says Les Carr, a professor of internet science at the University of Southampton who does stand-up comedy in his spare time.
“So comedians who have spent the last five years putting more content on TikTok, Instagram, and YouTube to gain followers should be concerned that OpenAI, Google, and Facebook are stealing their work. Just like writers and other artists.”
“Jokes are something that people love to share on the internet and on social media. So it’s very hard to tell where an AI-generated joke comes from. The question is: [yapay zeka] Did he make it up or just repeat it?
When Hobbs asked the AI to write jokes for his scene, he encountered another unexpected problem. ChatGPT started writing the jokes in the voice of a male comedian, and the jokes were about a shopping-obsessed girlfriend.
Hobbs asked the AI to rewrite the jokes from a woman’s perspective, this time telling the story of the shopping-obsessed girlfriend in first person.
The jokes the AI generated for Hobbs often turned into crude, lazy stereotypes about millennial women:
One of the jokes the AI produced was, “My social life is exploding. If by exploding you mean my best friend is a potted plant named Wilson…”
“A good stand-up tells a story”
It’s no surprise that AI models have trouble coming up with satisfying plots and punchline-pushing final jokes, says Michael Ryan, a graduate student and artificial intelligence expert at Stanford University.
“A good stand-up can take the audience through a funny story and into a funny joke. The comedian knows exactly where the joke is going and takes the audience there.”
Ryan reminds us that modern LLMs don’t work that way, which suggests that AI, unlike a comedian, can’t adapt in real time – at least not yet.
However, this may change.
Research continues to enable AI to better understand the world around it.
“Researchers are currently working to perfect the voice-related capabilities,” Ryan adds.
Experts are working on AI models that can “understand social factors and adapt to an audience, as well as have comedic timing.”
It could be in a few years
Ryan was part of a large analytics project working to test the limits of AI-generated jokes.
His research has shed light on current limitations on jokes, but Ryan believes that the growth of AI technology could lead to LLMs that can produce “really funny AI jokes in the next few years.”
It may be a while before AI can mimic the pitch of master comedians, but when it comes to writing a single funny joke, researchers have already made progress.
Screenwriter Simon Rich used a developed but unreleased OpenAI model called “code-davinci-002” for creative tasks and wrote about his experiences for Time magazine in 2023.
Rich collaborated with two other writers on a book of poetry written by an AI (and later read aloud by Werner Herzog), but not before asking the AI to tell some jokes.
The results were so good that Rich said it gave him nightmares.
Rich fed the AI a bunch of headlines from the satire site The Onion, and asked it to produce similarly sarcastic headlines.
“The budget of the new Batman movie increased to $200 million when the director insisted on using the real Batman,” wrote the artificial intelligence in one of them.
Humor is subjective, but the AI robot’s jokes far exceeded Rich’s expectations.
Drew Gorenz, a doctoral student at the University of Southern California who studies the psychology of what makes jokes funny, believes that AI-generated comedy is only as good as the commands given to the model.
Gorenz says the following on this very subject:
“Most people, including comedians, don’t do well when asked by a stranger to ‘say something funny.’ The more specific the request, the better the response.”
Some AI models may produce better results compared to others.
ChatGPT and Google’s Gemini, for example, were developed for general-purpose applications. An AI tool built solely on humor is likely to produce better jokes than other models.
Investing in people
The quest to digitize comedy is impractical, unethical, and futile, according to Alison Powell, an associate professor of communications at the London School of Economics who studies the impact of artificial intelligence on media.
He notes that developing generative AI models requires enormous amounts of energy and expense.
“Instead of investing in artificial intelligence for this job, investing in young comedians and cultural production will probably be cheaper, more interesting and bring much more surprising results.
Hobbs, who returned to London this summer, warned his audience that they and they were “in the hands of new cyber masters”. The positive response to his AI-written jokes could be a watershed moment for stand-up comedy.
Fortunately, it was just fitful laughter.
“I once gave my wife glue instead of lipstick. She still won’t talk to me,” Hobbs read to the audience.
Another joke was: “Dating is like shopping. You go out to buy what you want and end up buying something you don’t need.”
“I’ve never felt so stupid in my life,” Hobbs said on stage immediately afterward.
However, we are faced with a deeper truth. An AI model can perform the movements of a joke. It can even pick up on some of the nuances.
But only a human comedian can surpass those explosive moments in front of the audience.
AI models haven’t figured out the secret sauce yet. It looks like comedians won’t have to dust off their CVs just yet, so they can breathe a sigh of relief.
#funnier #humans
2024-08-25 03:38:22