June 10, 2019

Hollywood is quietly using AI to help decide which movies to make

Los Angeles-based startup Cinelytic licenses historical data about movie performances over time and cross-references it with information about films’ themes and key talent, using machine learning to tease out hidden patterns in the data. Its software then lets clients swap out parts such as lead actors to see how the movie would perform across different territories.

Why the industry is overdue for the AI treatment

Cinelytic’s key talent comes from outside Hollywood. Co-founder and CEO Tobias Queisser comes from the world of finance, where machine learning has been embraced for its utility in high-speed trading to calculating credit risk. Further, Cinelytic co-founder and CTO Dev Sen used to build risk assessment models for NASA, another industry where risk assessment is especially critical.

However, Queisser says that the business side of the film industry is 20 years behind (still relying on Excel and Word) the production side that uses all kinds of high tech to make movie magic.

We’ve been here before

Cinelytics isn’t the first company to attempt to harness AI to predict and improve box office returns:

  • ScriptBook: Belgian company founded in 2015, promised to use algorithms to predict a movie’s success by just analyzing the script.
  • Epagogix: Similarly, UK-based company founded in 2003, uses data from an archive of films to make box office estimates and recommend script changes.
  • Vault: Israeli startup founded the same year promised it could predict which demographics would watch a film by track metrics such as online reception to trailers.
  • 20th Century Fox: Uses a system called Merlin to analyze shots from trailers for content and duration (among other things) and see what other movies people will watch based on preferences.
  • Pilot: Centers its machine-learning process around audience analytics to make box office predictions.

Why this might be good and bad for film

Good: AI saves people the effort of doing some of the things they hate the most, such as sifting through scripts the way a manager would be sifting through a mountain of resumes. This effectively separates the wheat from the chaff.

Bad: For aspiring writers seeking to enter the big leagues, even talented ones, their work might never see the light of day much less a set of human eyes if their story, original as it may be, isn’t attractive to the algorithm evaluating them.

Good: If the prediction models become accurate enough, studios big and small can breathe easier and be more confident with their investments in films, ensuring higher returns on investments and confidence from shareholders, meaning they stay in business longer.

Bad: If an AI-assisted selection process becomes widespread, we’ll see a drop in the diversity of big-budget films being produced as studios seek to cater to the whims of as many demographics as possible, potentially complicating or watering down movies.

Good: Because box office numbers among other end metrics reflect audience choices, AI recommendations from cold hard data could override human prejudices that prevent certain stories from being told or certain people from being involved in a production.

Bad: If followed too closely, those recommendations might mean a studio will get an actor with the biggest box office market draw they can pay for without regard to whether or not the actor fits the story.

At the end of the day

Big studios have to regularly produce films to keep the money coming in year-round by pushing the right audience buttons at the right time of year from summer blockbusters to holiday movies. Movies that are released to coincide with certain attendance patterns (like say, a horror movie in time for Halloween) are usually designed to be enticing and entertaining if forgettable and for these, there are plenty of ways AI can help set big studios up for the best possible numbers they can.

These films types are formulaic enough that allow for this kind of “drag-and-drop” production. An example of this would be family-friendly comedies with a “big dude with a big heart” that have appeared regularly throughout the past twenty-plus years: Arnold Schwarzenegger (Jingle All the Way), Vin Diesel (The Pacifier), The Rock (The Game Plan and The Tooth Fairy) and most recently, John Cena (Playing with Fire).

That said, it remains to be seen if studios will ever let AI override decisions from accomplished writers, producers, and directors whose track record and reputation give them the authority to choose a given actor or other artist to work with. Further, AI can only make predictions based on what has already been made and not how tastes and popularity will shift in the future—the result of many factors outside of a strictly cinema-centric data set. That’s something that requires insight and instinct, something that humans are still valued for.

Play Pause