“Sunspring” is a roughly nine-minute science-fiction short released online by Ars Technica on June 9, 2016, with a screenplay written entirely by an artificial neural network. Filmmaker Oscar Sharp and NYU AI researcher Ross Goodwin made it for the Sci-Fi London 48 Hour Film Challenge, building the script with a long short-term memory recurrent neural network the system named itself: Benjamin.
Goodwin trained Benjamin by feeding it, in Ars Technica’s account, “a corpus of dozens of sci-fi screenplays he found online - mostly movies from the 1980s and 90s.” Working at the level of individual letters, the network learned to predict sequences and eventually to produce text formatted like a screenplay, complete with stage directions and even a song. The resulting dialogue is surreal and frequently nonsensical, yet the human actors - including Thomas Middleditch - performed it straight, and a coherent-feeling story of romance and conflict emerged. Sharp recalled that at the first read-through, “everyone around the table was laughing their heads off with delight.”
The film is less a polished product than a vivid demonstration of what sequence models could and could not do in 2016.
Why business readers should care: Sunspring is a time capsule from just before the transformer era, showing both the promise and the obvious limits of neural text generation - the same trajectory that, a few years later, produced systems good enough to raise serious questions for screenwriters.