A Melbourne researcher is exploring whether AI can deliver genuinely funny performances on stage, not just produce quick one-liners. While robots often get a laugh from slapstick mistakes—think of clumsy stumbles and unintended collisions—the big question is whether machines can be funny on purpose. And this project tackles that head-on with a fresh approach.
If you ask a language model like ChatGPT for a joke, you’ll typically get something that feels like a Christmas cracker gag—short, punny, and not deeply memorable. Dr. Robert Walton, a dean’s research fellow in Melbourne’s Faculty of Fine Arts and Music, is charting a different path. With roughly half a million Australian dollars from an Australian Research Council grant, he plans to train a swarm of robots to perform stand-up. Importantly, the initial phase will minimize spoken language.
Robots tend to amuse audiences when they fail or collide with objects, which makes people chuckle at their unintentional mischief. Yet, when they try to be funny on purpose, the reaction often stalls because we don’t truly believe they can be funny. This project challenges that assumption.
The broader conversation around AI in humor is not just about clever wordplay. It’s about nonverbal communication—timing, sensing the room, and connecting with listeners through physical performance. Walton’s ensemble will consist of about ten robots that are not humanoid androids but ground vehicles ranging from 40 centimeters to two meters tall. They’ll work alongside humans to learn visual comedy first, prioritizing body language, movement, and audience cues.
The robots will be equipped with enhanced sensing capabilities—ears that listen not only for words but for rhythm, pauses, and shifts in energy. Think of them as infants in training: they’re learning how to interpret the world through more than just language, developing a more holistic sense of presence and interaction. This aligns with a broader aim in machine learning and AI: expanding the ways systems perceive and interpret human environments.
Stand-up is a natural place to study the human–machine bond because it is inherently feedback-rich. The team plans to introduce voices later if needed, depending on how the project evolves. Yet the central goal is not to produce festival-ready comedians or replace human performers. It’s to determine whether believable humor can be taught to machines, thereby deepening understanding of how humor and manipulation might function in human–robot interactions and what risks or benefits this dynamic could introduce.
A tension runs through the endeavor: humor can diffuse tension and disarm a situation, but it can also be used to pressure or deceive. The research envisions collaborations where comedians test timing with robots, while also considering practical applications—such as care robots learning to deliver comforting, well-timed remarks to support emotional well-being. Still, the team remains mindful of potential misuse and asks: what does the ability to make audiences believe in a robot’s humor really imply for autonomy, influence, and trust?
Public attitudes toward AI in creative fields remain divided. Some fear the encroachment of automation on live performance and the broader theft of creative content, while others see opportunities for new collaboration and tools that augment human artistry. Prominent voices in events like Melbourne’s comedy scene emphasize that human originality—the lived experiences, imperfections, and spontaneous connections—drives what audiences find funny. The question Walton’s project raises is provocative: could machines capture that spark, or will humor always be a distinctly human trait grounded in flawed, relatable humanity?
