Last month, we saw what happened when a neural network was tasked with scripting a scifi short—a dazzlingly weird mess. Now, there’s Impossible Things, a horror movie that was written with equal parts human and machine input, with the goal of making the scariest film ever. And it is scary—but maybe not in the intended way.
Here’s a teaser from its Kickstarter campaign:
It’s an intriguing idea, and having an actual human on hand will ensure that the feature is more coherent than that sci-fi short. It also sounds like the software tool, created by Jack Zhang, won’t really be writing dialogue. The technology is more focused on generating story beats that will resonate most with audiences.
But the whole thing just sounds kind of sterile. According to Entertainment Weekly:
“A little over 85 percent of movies made today don’t make a profit in box office, which is the result of a mismatch between the movies being produced and audiences’ tastes,” Zhang said of the film, which his AI predicts will be most popular with women under the age of 25, via press release. “We used [AI] to generate the premise and the key plot points of the film. Before a single word was written, our AI told us that if we wanted to match audience taste, we needed to make a horror film that featured both ghost and family relationships, and that a piano scene and a bathtub scene would need to be used in the movie trailer to increase the likelihood that our target audience would like it.”
A bathtub scene. A piano. A title that evokes a certain hit Netflix series. So, basically, it will be the most formulaic, trope-filled horror movie ever. Is this the future of cinema: plots cranked out by computers who run the numbers, determining which stories please the audience the most and therefore rake in the most dough? (And is that really different from what studio execs do today?)
The answer is probably scarier than anything Impossible Things—a melodramatic tale of a young mother who starts losing her grip on reality after one of her children dies—will be able to generate.