|
Getting your Trinity Audio player ready...
|
In the fifth year of commissioning science-inspired plays for the Science@Play program at SPARC Theater in Livermore, we decided to push our exploration of artificial intelligence a step further. AI was not just the subject, but a creative partner.
Alongside funding for playwright Cleavon Smith to develop an original work, our Science@Play director, Michael Wayne Rice, undertook an unusual experiment: asking a large language model to write a play.
Rather than feeding the AI an existing script, a shortcut that raises ethical and legal concerns, he chose the harder path. Working with ChatGPT and Claude, he set out to teach the system how to write a play.
It was far more difficult than expected.
The AI produced scripts, but as one audience member said, “Living playwrights don’t have to worry about their jobs just yet.” Claude proved the stronger writer of the two systems, but even it struggled with the fundamentals of drama. To generate “Entangled”, a 58-page script, Rice built a 50-page prompt, outlining character development, objectives, obstacles and arcs.
And still, one essential element remained elusive: conflict.
At the heart of most plays is conflict. A series of conflicts lead to a turning point, where the protagonist experiences a transformation and the story drives toward resolution. But the AI resisted. Despite repeated revisions, it shied away from allowing conflict between characters.
This raised a question in our post-performance discussions: are these systems fundamentally designed to avoid conflict? After all, they are engineered to be agreeable and to sustain engagement with users. But without that aspect of human behavior, they cannot write a masterful drama.
The AI could capture the lighter textures of human interactions: humor, personality quirks, ambition, romance. We had a laugh on lines like “You told me in Scene 17…” But it failed at creating the climax we longed for. The resulting play had some surprisingly “real” moments, but it struggled with dramatic structure.
Many in our audience found this reassuring. Great plays, it seems, are still beyond AI’s reach. But Rice offered a note of caution: our results were shaped by our ethics. We deliberately avoided using the work of living playwrights to train the model. Others may not.
With access to existing scripts, AI could easily replicate the structure and voice of established writers, producing work that feels more satisfying, even as it raises serious questions about authorship.
The technology already gets a surprising amount right. And with practice, the process will become more efficient; the 50-page prompt required this time will likely shrink.
Which leaves the real challenge with us. The limits we encountered are not only technical, but ethical. As we develop and deploy helpful and increasingly powerful tools, we must decide how they learn, what they are allowed to use, and what values guide their use.
Even if it means, for now, that we still have to write our own great plays.
As SPARC launches into its fully produced season of the masterpieces “The Glass Menagerie” at The Bankhead Theater May 1-3 and San Ramon’s Front Row Theater May 8-10 and “Hamlet” at Darcie Kent Winery July 2-26, we’ll keep the lessons of solid dramatic structure in mind.
It’s our experience of being human in those climactic and transformational moments of our lives that drive our need for and love of theater. I cling to the belief that we still need this special form of storytelling.
Editor’s note: Lisa A. Tromovitch is the executive and artistic director for SPARC Theater, based in downtown Livermore. Its “Entangled: An AI Experiment” production was presented to audiences from March 13-15 as part of its 2026 Science@Play series. Learn more about SPARC at sparctheater.org.
