Hollywood is silently using AI
More specifically, AI plays the role of a person who helps you choose the cast and predicts how much you will earn. The movie world is full of unpredictable things. It is well-known that Will Smith rejected Neo's role in The Matrix or Nicholas Cage was cast as the main character Tim Burton in Superman Lives, but he only had time to try out the costume before the film was shot. Not only us but also directors and actors always wonder how the results will be when altering those options and going a different way. They wanted to know whether to choose Alicia Vikander over Gal Gadot would bring failure or breakthrough, or a US blockbuster movie will set a box office record across Europe. Now they have what it takes to help them answer this question, Artificial Intelligence (AI).
Los Angeles-based startup Cinelytic is among many companies promising that AI will become a wise movie producer. It synthesizes historical data on film screenings for many years, cross-references with information on film topics and key talents, then uses machine learning to work with hidden data. The software allows customers to play a fantasy football game with their movie, enter a cast, then switch one actor to another to see if this affects the box office in a movie project.
"Suppose you have a summer blockbuster with Emma Watson in the leading role," Cinelytic co-founder and CEO Tobias Queisser said. "You can use Cinelytic software to see how Jennifer Lawrence's character instead of Emma’s can change the box office performance."
Cinelytic is not the sole company that aspires to apply AI to the movie industry. In Belgium, Book ScriptBook was founded in 2015, saying that their algorithms can predict the success of the movie just by analyzing its scenario. Israel's Vault startup, founded in the same year, commits to customers that they can predict who will watch their movies by following the audience's response to an online trailer. Another company called Pilot also provided similar analysis, promising that they could forecast box office revenues of up to 18 months before releasing the film with impressive accuracy.
The market has gradually warmed up and even the other companies have joined. Last November, 20th Century Fox explained how to use AI to detect objects and scenes in a trailer and then predict which audience will be most attracted to the movie. However, when looking at the research we could see 20th Century Fox's methods seem quite “a hit or a miss” (when analyzing Logan trailer in 2017, the company's AI software pointed out these quite unhelpful tags: "Facial hair", "Car", "Beard", and the most popular of all - "Trees"). According to Queisser, the use of this technology is overdue.
"Now, on the set, there are robots, there are drones, there is super-tech, but the business side hasn’t been developed for 20 years. People use Excel and Word, which method is quite simple, business data is not good enough and there is almost no analysis at all", Queisser added.
That's why Cinelytic's power comes from outside Hollywood. Queisser has worked in finance, an industry that captures machine learning for everything from high-speed transactions to credit risk calculations. Co-founder and CTO Dev Sen also came from a similar technology background, he built risk assessment models for NASA.
But, are they correct? It is tough to answer. Cinelytic and other companies refuse to make any predictions about the success of upcoming movies and academic research on the subject is also scarce. But ScriptBook shared the company-made forecasts for movies released in 2017 and 2018, indicating that their algorithms are still doing very well. In a sample of 50 movies including Hereditary, Ready Player One, and A Quiet Place, just under half made a profit, giving the industry an accurate success rate of 44%. ScriptBook's algorithms, by comparison, correctly guessed whether a movie would make money in 86% of the time, double the exact rate that the industry achieved, according to data scientist Michiel Ruelens of ScriptBook.
An academic article published on this topic in 2016 similarly stated that the credible predictions of the profitability of the film can be made using basic information such as the film topic and the stars contributed. But Kang Zhao, co-author of the article with his colleague Michael Lash, also warned that these statistical methods could also make mistakes.
First, the predictions made by machines are often blindingly clear. You don't need sophisticated and expensive AI software to tell you that a star like Leonardo DiCaprio or Tom Cruise will improve your chances of making a blockbuster movie.
The algorithms are basically relatively "conservative" because they learn by analyzing what has been done in the past. Therefore, they are unlikely to take into account cultural factors or changes in tastes that will occur in the future. This is a challenge throughout the AI industry and can partly cause problems such as AI bias. (Amazon's AI recruitment tool has eliminated female candidates because it learned the skills from jobs that men currently dominate).
Zhao offers a more moderate example of the blind spot in algorithm: the fantasy action movie Warcraft 2016, based on the World of Warcraft MMORPG game. Because the transition from game to movie is so rare, it's really hard to predict how the movie would be making. The film did not receive a good response in the United States, grossing only $ 24 million at the weekend, but was a huge success in China, becoming the highest-grossing foreign-language film in the country's history. However, both AI and humans had not predicted that. There are similar stories that happened in ScriptBook's prediction for movies in 2017/2018. The company's software accurately predicted the success of the horror film Get Out but underestimated its popularity at the box office (revenue was estimated at only $56 million instead of $176 million in reality). The algorithms also rejected The Disaster Artist, a tragic story from Tommy Wiseau's classic The Room starring James Franco. ScriptBook says the movie will only earn $10 million but instead, it grossed $21 million. As Zhao asserts, "We only grasp what can be obtained by data. To take into account other factors (like the way The Disaster Artist benefits from The Room's reputation), you need a similar precedent. ”
Some are denying that Hollywood is fond of using AI to look at potential movies. Alan Xie, CEO of Pilot Film, a film industry analytics company says that he has never met an American film director who believes in AI script analysis, let alone integrating it into their decision-making process.
Xie said that studios may simply not want to talk about using such software, but he also said that script analysis is an inaccurate tool. The amount spent on marketing and advertising on social networks according to Xie is a much more reliable predictor of box office success. “Internally at Pilot, we have developed box office forecasting models based on the characteristics of the scenario and they show worse than models based on real-time social network data”, Xie said.
Although skeptical about the specificity of the application, big figures can change their minds about AI. Ruelens and investment manager Scarso said the only factor that convinced Hollywood to stop rejecting big data was Netflix. The online giant is always open about its data-based programming, monitoring the actions of millions of subscribers in great detail and perceiving a lot of information about them, and giving them good suggestions for users in the Choose Your Own Adventure-style bar. "We have a big global algorithm, which is extremely useful because it promotes the tastes of all consumers around the world," said Cameron Yellin, head of product innovation of Netflix in 2016.
Ruelens said that this transition was noticed by Hollywood. “When we started four years ago, we had meetings with big companies in Hollywood. They were extremely skeptical and said ‘We have had decades of expertise in the industry. How can machines tell us what to do? Now everything has changed. Companies that have done their own validation studies, they wait to see what the software has correctly predicted, and gradually learn to trust the algorithms. ”
“They said that they started accepting our technology. They just need time to see with their own eyes”.