Nadira Azermai has created an artificial intelligence called ScriptBook that uses an algorithm that can analyze film screenplays and predict whether or not they will be hits or flops. To test this theory, ScriptBook was fed the scripts for 32 Sony movies that lost money between 2015 and 2017 and was able to pick out 22 of those that it “predicted” would be failures. According to Azermai, “If Sony had used our system they could have eliminated 22 movies that failed financially.”
As she described ScriptBook’s mission:
“Our mission is to revolutionize the business of storytelling by using AI to help producers, distributors, sales agents and financiers assess their risk.”
ScriptBook could greatly impact the financial structure of the film industry. For instance, during the 2015-17 period that ScriptBook analyzed, the studio released 62 movies, with 32 of them losing money. That’s over half, and while ScriptBook only flagged 22 of those failures, that’s still MILLIONS of dollars that could have been saved. Furthermore, this could eliminate the need for human script readers and the costs of unreliable market research, test screenings and focus groups.
ScriptBook data scientist Michiel Ruelens stated:
“When we show this to customers, the first question is: How is it even possible to give a script to a computer and somehow it can come up with all these outputs?”
Screenplays are uploaded to ScriotBook in PDF format and in just five minutes, the program spits out a “detailed analysis” that “among other things: predicts the MPAA rating, analyzes its characters, detecting the protagonists and antagonists; assesses the emotions of each character; predicts the target audience, including gender and race; and, most importantly, makes box office predictions.” Along those same lines, ScriptBook can be of use in the modern era, when more attention is paid to equal representation in films. “The software can detect, for example, whether a film passes the test of including at least two female characters having a conversation that is not about men. Also, it can measure how many dialogs are between two men, how many between two women, and how many between males and females.”
ScriptBook has been “trained” by uploading 6,500 existing scripts and it has taken that information, provided by a human, and then continuing to “learn” from there. It will only “green light” a project if its analysis scores 84% or higher. According to Azermai, this is three times as reliable as human findings.
But it’s not flawless. As stated, it only picked out 22 out of 32 flops from Sony’s output. And on the flip side, it pegged ‘La La Land’ as only making $59 million. (It did green light it despite that because it detected that it would only require a small budget.) In reality, it made over $100 million thanks to awards season buzz and subsequent wins. (It’s really NOT that good though, am I right?)
ScriptBook isn’t exactly cheap. It charges $5,000 per script but does offer discounts for bulk analyses. But compared to other methods that studios already use– which include reshoots and re-editing– it may be a bargain.
Does this sound like a good idea? Or does this sound too clinical and devoid of a human touch?