Why machine-generated humor is the holy grail of AI

In “The Outrageous Okona”, the fourth episode of the second season of Star Trek: The Next Generation, the resident Android Data of the enterprise is trying to learn the one skill they could not have mastered before: humor. While visiting the ship’s holodeck, Data takes lessons from a holographic comedian to try to understand the business of making fun.

While the worlds of Star Trek and the real world can sometimes be far apart, this act applies to the machine intelligence here on earth. Put simply, it is extremely difficult to get an AI to understand humor and then generate its own jokes.

How hard Forget Go, Jeopardy !, chess and loads of other impressive demos: according to some experts, building artificial intelligence at the level of a top comedian can be the true measure of machine intelligence.

And although we are not there yet, it is safe to say that we may come a lot closer.

Witscript cracks the code

Joe Toplyn is someone who is not afraid of challenges. A trained engineer (with a huge career gap in terms of actual exercise), Toplyn had a successful career as a television writer. A four-time Emmy winner, he has been head writer for the likes of David Letterman and Jay Leno. A few years ago, Toplyn became interested in the question of whether there was an algorithm (i.e. a process or set of rules to follow) that would help you write really funny jokes.

“People think it’s magic,” he told Digital Trends. “I think some comedy writers or comedians try to portray their work as magic. Well, it’s like magic in the sense that a magic trick is constructed and designed and works in a way that leads one to believe that the magician has supernatural powers. But that really has a logic. “

That belief in steel logic in joke-telling – sharpened while Toplyn tried to teach aspiring comedians his “magic” – ultimately led him to seek an AI capable of generating spontaneous jokes that Regular conversations fit into the game. With the name Witscript, the results add up to an innovative AI system that makes improvised jokes. A chatbot that uses Witscript for ad-lib jokes could, Toplyn says, help create personable artificial companions to solve the “huge problem” of human loneliness. Think of it like PARO, the robot seal with punchlines.

“It’s contextual,” Toplyn said of Witscript, which was recently unveiled at the 12th International Conference on Computer Creativity (ICCC 2021). “This distinguishes it from other joke-generating systems that generate self-contained jokes that cannot easily be integrated into a conversation. When talking to a funny friend, chances are their jokes will be incorporated into a conversation in response to something you said. It’s much less likely that your friend will just start telling a standalone joke like, ‘A man goes to a bar with a duck on his head …’ “

The funny formula

This spontaneous quality comes from the joke-writing algorithms that Toplyn developed in-house.

“Basically, the basic algorithm for writing jokes works like this: It starts with choosing a topic for the joke, which can be a sentence someone says to you or the subject of a news story,” he said. “The next step is to pick what I call the two ‘topic handles’, the words or phrases in the topic that are most responsible for grabbing the audience’s attention. The third step is to generate associations between the two topic handles. Associations are what the audience is likely to think about when thinking about a particular topic. In the fourth step, a punch line emerges that surprisingly combines an association of one of the two topic handles with an association of the other. The final step is to create an angle between the subject and the punch line: a sentence or phrase that connects the subject to the punch line in a natural-sounding way. “

Francesco Prandoni / Redferns via Getty Images

If all of these holds and angles sound like hard work, the evidence – ultimately – lies in the pudding. With 13 input themes, Witscript generated a series of jokes that Toplyn then contrasted with his own efforts. For an evaluation panel, he outsourced the evaluation to Amazon Mechanical Turk workers, who rated each freshly minted joke on a scale from one (no joke) to four (a very good joke). One of Witscript’s best efforts received a rating of 2.87 (“That’s almost a joke,” Toplyn said) to his own 2.80 as a Beat Master student. The witscript joke? On a line about the 25th anniversary of the performance arts company Blue Man Group it quipped: “Welcome to the Bluebilee.”

While he may not be entirely ready to oust Dave Chappelle just yet, Toplyn believes Witscript proves that humor can be automated to some extent. Even if there is still a long way to go. “As machines do these algorithms better, the jokes they generate get better,” he said.

However, he also urged caution. “To generate [truly] elaborate jokes like a skilled human comedy writer, machines will need the common sense knowledge and reasoned ability of a typical human being. “

A pioneer in AI comedy

This, it turns out, can be the crux of the matter. Humor may seem frivolous, but for those who work in language, comedy, and artificial intelligence, it is anything but.

“We use humor in many different ways,” Kim Binsted, professor at the University of Hawaii Institute of Information and Computer Science, told Digital Trends. “We use it to build social relationships. We use it to define in-groups and out-groups. We use it to come up with ideas that we may not seriously want to express. Of course there is non-linguistic humor, however [linguistic humor] falls into a category of usage that is really powerful. It’s not just a stand-up on stage that uses it to get some laughs. It’s something we use all the time [within our society.]”

“It’s a huge sign of advanced intelligence because to be really funny an AI has to understand a lot about the world.”

When it comes to computer humor, Binsted is a pioneer. In the 1990s, she developed one of the (possibly) first AIs designed to generate jokes. Binsted’s JAPE (Joke Analysis and Production Engine) was developed with Professor Graeme Ritchie and was a joke-making bot that could create question-and-answer puns. An example could be: “Q) What do you call a strange market?” “A) A bizarre bazaar.”

“That was great because it allowed me to pick all the low-hanging fruit before anyone else,” she said humbly. “What I did with puns.”

An AI-complete problem

Since then, Binsted has developed various other computer humor bots – including one that can come up with variations on “Yo Mama” jokes. While Binsted’s work has since evolved to examine long-term exploration of human space, it still regards joke-telling AI as something of a holy grail for machine intelligence.

“It’s not in one of those things like chess that when AI started, people said, ‘Well, if a computer can ever really play chess, we’ll know it’s fully intelligent,'” she said. “That is obviously not the case. But I think humor is one of those things that flowing humor with a computer has to be really intelligent in other ways too. “

Microphone in a bar7713 photography

That’s why telling jokes is such an interesting challenge for machines. It’s not because cracking an AI is as useful to human beings as using machine intelligence to fight cancer, for example. But it’s a tremendous sign of advanced intelligence because to be really funny an AI has to understand a lot about the world.

“Humor depends on many different human abilities, such as world knowledge, linguistic abilities, argumentation, [and more]“Thomas Winters, a Computer Science Ph.D. Student researching artificial intelligence and computational humor versus digital trends. “Even if a machine has access to this type of information and capabilities, it still has to recognize the difficulty of the joke for itself. For something to be funny, a joke must not be too easy or too difficult for a person. A machine that makes jokes shouldn’t use knowledge that is too vague or knowledge that is too obvious with predictable punchlines. Because of this, computer humor is usually viewed as an AI-complete problem. [It means] We need an AI that has functionally similar components to a human brain in order to solve computational humor, as it depends on all of these abilities of the human brain. “

Think of it like a Turing test with a laugh trail. Coming to a super intelligence near you. Hopefully.

Editor’s recommendations

Comments are closed.