The following essay was produced as part of the 2019 Locarno Critics Academy, a workshop for aspiring film critics that took place during the 72nd edition of the Locarno Film Festival.
Artificial intelligence is everywhere: It can drive a car, chat with customers, or help patients with neuronal damage to recover their potential. But if data-assisted moviemaking can help predict a movie’s outcome, what room is there left for artistic freedom? At this year’s Locarno Film Festival, Sami Arpa, CEO and co-founder of Largo Films, a startup based in Lausanne, Switzerland, and creator of the LargoAI technology, shared his insight about the evolution of this maybe-not-so-unnatural union.
At Locarno last year to present sofy.tv, a VOD service for short films, Arpa recalled, “I was approached by industry professionals, mostly producers and distributors, who asked me if the AI developed for sofy could be used for their own purposes, to help them predict a movie’s outcome. A few directors also approached me, although they were much more skeptical at first.” Originally designed as a predictive analytics tool to help users pick the movies that would best suit their taste, Arpa set to working on an AI that’s use could be extended to support other needs of the film industry.
LargoAI relies on a database of some 30,000 movies. Following a top-down learning process, the software starts to recognize and understand repeating patterns both on a small and higher level. This means that not only are larger elements like editing, action, and music reviewed, but that a host of smaller detailed are also taken into account, such as the way characters move, their hair color, or even the objects they use. The program first understands general concepts before delving into much smaller details, at which point it becomes able to teach itself. “But one thing the AI cannot do,” Arpa pointed out, “is to explain the reasons behind the results. Defining the ‘why’ is our next challenge.”
Popular on IndieWire
As a country-specific tool, the AI is systematically adapted depending on the part of the world the film will be released in. As Arpa explained, “If a film is to be released in 20 to 30 countries, the AI can help define where the effort needs to be put.” Local cultural traits vary enormously, which is why defining a target with the customer first and foremost is crucial. For instance, in the U.S., if the “thriller” content of a movie is brought up by 10 percent, it will statistically increase its chances at the box office. In Italy, where comedy seems to be declining, adding some drama will help the film do better, as backed by statistics.
Just as important is the context in which a film will be released: Is it meant for TV viewing, the multiplex, or a festival? Whether the movie is intended to be an edgy work of art or appeal to the masses, AI can help.
Data-driven filmmaking has its limits, though, and no matter how technologically advanced it is, artificial intelligence will never be 100 percent accurate. Unlike a self-driven car that can recognize other cars, humans, or objects, “the problem with films is that people have a different definition of what genre or quality is,” Arpa said. Subjectivity, in a word, makes the concept of a “good movie” more abstract to grasp. Even though a common understanding can be defined in what makes a movie successful, predictions cannot be trusted blindly. Whereas AI is data-driven, cinema relies more on human creativity, and Arpa doubts the industry is ready to accept an all AI-directed movie anytime soon: for the moment, the inputs are limited in order to protect the director’s creativity.
If AI can be seen as a threat to diversity in the film industry, it can paradoxically help draw attention to the topic. After examining all the Oscar-winning features from the 2017 – 2018 season, the tool was presented to the Academy of Motion Picture Arts and Sciences in February 2019, where it was revealed that a majority of the movies, all categories combined, had an astonishing number of similar repeating patterns.
Hollywood, for example, has a strong liking for “hero’s journey” movies. Things are a bit less clear-cut in Europe, though, where diversity is more important and less data is available. The same problem occurs with new actors: Although the LargoAI database contains more than 200,000 names, predictions can be made only provided the person has a long enough film track record to analyse. The same can be said of highly experimental movies, and not only is the body of work insufficient to be analyzed, but chances are that critics’ opinions will be too disparate to be used as one large, coherent whole.
In a mostly profit-driven industry, is independent cinema doomed to disappear? Arpa, who has also directed a couple of his own short features, has a clear vision: for him, AI is best kept as an assistance tool. “I really like cinema as it is,” he said. “To me, the danger lies not in the use of AI, but rather in the fact that cinema is getting less and less democratic, with the biggest decisions left to a handful of powerful people.” AI might help even out the situation by providing predictions that show the movie, even a small indie, is likely to find its audience. For now at least, it seems technology hasn’t killed art. Yet.