The list of jobs that are being threatened by artificial intelligence is growing by the day, but could generative AI ever take center stage on Broadway?

In the past year, the unions representing Hollywood actors and writers won protections against the potential applications of AI in TV and film—just in time for OpenAI’s Sora to create its first short films. Scarlett Johansson has called out the company for allegedly copping her voice for ChatGPT 40 and musicians from Billie Eilish to Jon Bon Jovi have signed an open letter calling for more responsible use of AI, as AI-powered songs recreating artists’ voices crop up all over YouTube.

It’s hard to forget the viral consequences of using AI to write and market a live candy-lovers experience that turned into a Fyre Festival redux. But live theater’s reliance on flesh and feelings make the prospect of a total robot takeover on stage a little tougher to imagine. Even so, AI’s generative abilities have recently been explored on New York stages, in pieces that pondered, with some mixture of playfulness and alarm, where all of this might be headed.

Last fall, in “Prometheus Firebringer” at Theater for a New Audience, the playwright and performer Annie Dorsen used AI to create her co-stars, an eerie chorus of crude masks with a kiddie Mike Meyers quality. Dorsen used ChatGPT 3.5 to generate iterations, projected overhead, of a speculative passage from the myth of Prometheus, the Greek god who defied Zeus and gave humans fire — civilization’s foundational technology. 

The result was haunting but rote; AI’s workman-like variations on the story were mostly an exercise in semantics. But Dorsen’s portion of the piece, which she delivered seated behind a table with a screen overhead, appeared to demonstrate that AI is only doing what artists and writers have for centuries: building new work from existing ideas. Every line in her heady, philosophical exploration of innovation was footnoted on the screen as having been drawn from another text. It was proof that art and tech are more alike than not.

AI’s rudimentary ability to imitate an artist was the subject of “Artificial Flavors,” a piece recently staged by the Civilians at 59E59 Theaters. The company’s artistic director, Steve Cosson, acted as the narrator, offering audiences a basic primer on AI and then prompting ChatGPT to write the script and lyrics for a new musical on the spot. A game group of actors improvised the music, in tandem with music director Dan Lipton, resulting in a half human-, half machine-generated new musical.

Or at least the semblance of one. The point seemed to be what a poor substitute generative AI made for a musical theater writer, a project that ultimately seemed self-serving. The audience had to sit through the painful proof: a musical-theater-hater’s worst nightmare packed with inane metaphors and nonsensical turns of plot.

The most thrilling and unexpected recent use case leveraged AI onstage to allow for an artist’s own radical vulnerability. The playwright and performer Mona Pirnot, whose piece “I Love You So Much I Could Die” premiered in February at New York Theater Workshop, used a text-to-speech tool to narrate her story while she sat with her back to the audience. 

Pirnot first discovered the Microsoft tool when she had difficulty voicing her feelings even to her husband (the playwright Lucas Hnath, also the show’s director) in the aftermath of devastating personal events. We learned this not from Pirnot herself, but from the disembodied voice of the text-to-speech tool, which allowed her sufficient distance from her story to be comfortable sharing it with an audience. It was not unlike puppetry, in that inanimate objects can sometimes give life to a narrative that exceeds what performers can do on their own. 

The piece brought to mind AI’s potential use in mental health settings, to detect PTSD, for example, by listening for the speech patterns that correspond with lived trauma. In that case too, synergy lies in the dialogue between human impulse and machine capability.

AI’s ability to generate various aspects of IRL performance are very likely to improve. As another tool for pushing the boundaries of live theater, the possibilities it holds are in the hands of creators. But AI still comes with a degree of uncanny that’s tough to disguise — after all, our in-person BS detectors have been refined over millennia.