Tory Shepherd 

A robot walks into a bar: can a Melbourne researcher get AI to do comedy?

Machines can be funny when they mistakenly bump into things – but standup is a tough gig even for humans
  
  

Stage set up for standup comedy. A Melbourne researcher is setting out to see if AI can be trained to be genuinely funny.
An ensemble of about 10 robots – which will not be androids but ground vehicles between 40cm and 2 metres tall – will work with humans to learn how to be funny. Photograph: Getty Images

Robots can make humans laugh – mostly when they fall over – but a new research project is looking at whether robots using AI could ever be genuinely funny.

If you ask ChatGPT for a funny joke, it will serve you up something that belongs in a Christmas cracker: “Why don’t skeletons fight each other? Because they don’t have the guts.”

The University of Melbourne’s Dr Robert Walton, a dean’s research fellow in the Faculty of Fine Arts and Music, is taking a different approach to working out whether robots can do comedy.

Thanks to an Australian Research Council grant of about $500,000, he will train a swarm of robots in standup. And, at least in the beginning, they won’t use words.

“Robots are good at making people laugh … they are humorous because they break and they bump into things, and so we’re laughing at them,” Walton says.

“However, when they try to do something funny on purpose, it ain’t so funny any more. We don’t laugh at them because we really, deep down, don’t believe that they can be funny.”

Saturday Night Live’s Tina Fey said exactly that at this year’s Edinburgh comedy festival. AI is “unable to be funny”, she said.

But what Walton is looking at is not AI based on text or large language models.

He is going to start with non-verbal communication, something that has to be performed rather than written. The fundamentals of comedy, he says, are timing, reading the room, the connection with the audience, along with physical comedy such as clowning.

So his ensemble of about 10 robots – which will not be androids but ground vehicles between 40cm and 2 metres tall – will work with humans to learn how to be funny visually in the first instance.

They’ll sense movement, the way a head tilts or when someone laughs.

“We’re giving these systems more senses, like human senses … giving them ears, not just listening for words but for things like the gaps in between words, the rhythms of things,” he says.

He likens them to babies who don’t yet know how to make sense of the inputs.

“That’s partly what we’re trying to do with machine learning and AI – giving it more ways to sense and more ways to build a more holistic understanding of what it means to be in the world,” he says.

“It is in standup comedy, really, that the connection between the robot and the audience is so clear, and there’s so much feedback going on.”

Asked if eventually they will add voices, Walton says “potentially”. “Depends how we go,” he adds.

There is a tension here, as the performance industry is just one of those where jobs are threatened by AI, and AI steals creative content.

Walton’s project is not about creating robots that will take over comedy festivals, though, but about investigating whether believable comedy is something robots can be taught, to better understand how machines might use both humour and manipulation, and to better understand human-robot interactions and their risks and benefits.

A paradox at the heart of his work, Walton says, is that humour can be used to disarm a situation but can also be used coercively.

He says it might be interesting for comedians to work with robots with comedic timing, but the same techniques could be used, for example, by care robots that can learn to say the right thing at the right time to cheer people up.

“But while I’m looking into this work of building belief in comedy performance by machines, I’ve got this other eye on what does it mean, and how might this be used coercively?” he says.

Many doubt whether that first step, making robots funny, is possible.

At this year’s G’Day USA arts gala, Australian comedian and polymath Tim Minchin told the crowd that humans are interested in “the agency of their fellow human behind the art, struggling, striving, making choices and errors”.

“AI might come for the perfectible stuff but never for our flaws,” he says.

“Our flaws are our humanity.”

The director of the Melbourne comedy festival, Susan Provan, says what makes comedy enjoyable is “the authentic human originality”.

“A performer is bringing something only they can bring, because they are bringing their individual lived experience to the material,” she says.

“What’s funny is something that comes from a moment, a magic moment, a pause, an interaction with an audience member, an idea that connects or doesn’t connect.

“You’d be laughing at the robot stuffing up. That’s what would be funny.”

 

Leave a Comment

Required fields are marked *

*

*