Advertisements
<pre><pre>Hollywood quietly uses AI to decide which films to make

The film world is full of intriguing what-ifs. Will Smith famous rejected the role of Neo in The Matrix. Nicolas Cage was released as the lead actor Tim Burton & # 39; s Superman is alive, but he only had time to try the costume before the film was canned. Actors and directors are constantly looking at projects that were never created or created by someone else, and fans wonder what would have happened.

Advertisements

For the people who make money with movies, that is not good enough.

If Alicia Vikander instead of casting Gal Gadot is the difference between a flop and a hit, they want to know. As a movie that bombs the US, box office records would have been set across Europe, they want to know. And now artificial intelligence can tell them this.

The Los Angeles based start-up Cinelytic is one of the many players in this space, and promises that AI will be a sensible producer. Over the years, the company passes on historical data about film performances, then links it to information about the themes and the main talent of films, and uses machine learning to disrupt hidden patterns in the data. With the software, customers can play fantasy football with their film, enter a script and a cast, and then trade one actor for another to see how this affects the projected box office of a movie.

Imagine having a summery blockbuster with Emma Watson in the lead, says Cinelytic co-founder and CEO Tobias Queisser. You could use the Cinelytic software to see how changing her for Jennifer Lawrence could change the box office performance of the film.

"You can compare them separately, compare them in the package. Model both scenarios with Emma Watson and Jennifer Lawrence and see, for this specific film … what better implications for different areas," Queisser says The edge.

Cinelytic is not the only company that wants to apply KI to the film activity. A number of companies with promising similar insights have emerged in recent years. Belgium & # 39; s ScriptBook, founded in 2015, says the algorithms can predict the success of a movie by analyzing its script. The Israeli startup Vault, founded the same year, promises customers that it can predict which demographics will watch their films by (among other things) following how their trailers are received online. Another company called Pilot offers similar analyzes, promising that it can predict the revenue of the cash register up to 18 months before the launch of a film with "unparalleled accuracy".

Advertisements

The water is so hot, even established companies jump in. In November last year, 20th Century Fox explained how the AI ​​used to detect objects and scenes in a trailer and then predict which & # 39; micro-segment & # 39; would find the movie most attractive from an audience.

If we look at the research, the methods of 20th Century Fox seem to be missing a bit. (Analysis of the trailer for 2017 & # 39; s Logan, the company's AI software came up with the following, useless tags: "facial_hair", "car", "beard" and – the most popular category of all – "tree".) But Queisser says the introduction of this technology is too is late.

"On a movie that has now been set up, it's robots, it's drones, it's super high-tech, but the business side hasn't evolved in 20 years," he says. "People use Excel and Word, fairly simplistic business methods. The data is very quiet and there are hardly any analyzes."

That is why Cinelytic's most important talent comes from outside of Hollywood. Queisser was in finance, an industry that embraces machine learning for everything from fast trading to calculating credit risk. His co-founder and CTO, Dev Sen, stems from a technically difficult background: he built risk assessment models for NASA.

"Hundreds of billions of dollars in decisions were based on (Sen's work)," says Queisser. The implication: the film industry can certainly trust him.

But what are they right about? That is a more difficult question to answer. Cinelytic and other companies The edge spoke with refused to make predictions about the success of upcoming films, and scientific research on this subject is limited. But ScriptBook has shared the predictions it made for films released in 2017 and 2018, suggesting that the company's algorithms are doing well. In an example of 50 films, including hereditary, Ready Player One, and A quiet place, just under half made a profit, giving the industry an accuracy rate of 44 percent. For comparison, the ScriptBooks algorithms correctly advised whether a movie would make money 86 percent of the time. "So that is twice the degree of accuracy of what the industry has achieved," says data scientist Michiel Ruelens The edge.

Advertisements

An academic paper published on this subject in 2016 Similarly, it was claimed that reliable predictions about the profitability of a movie can be made using basic information such as the themes & stars of a movie. But Kang Zhao, who is co-author with his colleague Michael Lash, warns that this type of statistical approach has their flaws.

One is that the predictions of machines are often simply obvious. You don't need advanced and expensive AI software to tell you that a star like Leonardo DiCaprio or Tom Cruise increases the chance that your movie will be a hit, for example.

Algorithms are also fundamentally conservative. Because they learn by analyzing what has been done in the past, they cannot take into account cultural shifts or taste changes that will take place in the future. This is a challenge for the entire AI industry and can contribute to problems such as AI prejudices. (See, for example, Amazon's discarded AI recruitment tool that has penalized female candidates because it has learned to associate technical ability with the current male-dominated intake.)

Zhao offers a more benign example of algorithmic short-sightedness: the 2016 action fantasy film Warcraft, it was based on the MMORPG World of Warcraft. Because such game-to-film adjustments are rare, he says, it's hard to predict how such a movie would perform. The film did poorly in the US, taking only $ 24 million into its opening weekend. But it was a huge hit in China and became the best-scoring film in the foreign language in the country's history.

Who saw that coming? Not the algorithms.


Warcraft "data-upload-width =" 5120 "src =" https://cdn.vox-cdn.com/thumbor/IPWHdb9BZc04Eb--o_qI14CLSGU=/0x0:5120x2128/1200x0/filters:focal(0x0:5120x2128):no_upscale () /cdn.vox-cdn.com/uploads/chorus_asset/file/6615287/WarcraftMain.0.jpg

AI did not predict the success of & # 39; Warcraft & # 39 ;. (In all honesty, the people either.)

Advertisements

There are similar stories in the ScriptBook predictions for 2017/2018 films. The company's software correctly turned Jordan Peele's horror hit green Get out, but it underestimated how popular it would be at the checkout, predicting $ 56 million in revenue instead of the actual $ 176 million it made. The algorithms also rejected The Disaster Artist, the tragicomic story of Tommy Wiseau's cult classic The room, starring James Franco. ScriptBook said the film would only earn $ 10 million, but instead it recorded $ 21 million – a modest profit on a $ 10 million film.

As Zhao says: "We only catch something can are recorded by data. "To take into account other nuances (such as the way The Disaster Artist traded on the memeiness of The room), you have to keep an eye on people.

Andrea Scarso, a director of the Ingenious Group in the United Kingdom, agrees. His company uses Cinelytic software to guide investments in films, and Scarso says the software works best as an additional tool.

"Sometimes it validates our thinking, and sometimes it does the opposite: we suggest something that we didn't consider for a certain type of project," he says. The edge. Scarso says using AI to play with a movie's blueprint – swapping actors, increasing the budget, and seeing how it affects movie performance – & # 39; opens a conversation about different approaches & # 39; but it is never the ultimate arbitrator.

"I don't think it ever changed my mind," he says about the software. But it has many applications all the same. "You can see how sometimes only one or two different elements around the same project can have a huge impact on commercial performance. Having something like Cinelytic, together with our own analyzes, proves that (suggestions) that we make are not just our own crazy ideas. "

Advertisements

But if these tools are so useful, why aren't they used more widely? Scriptbook's Ruelens suggests that one non-Hollywood character is to blame: shyness. People are ashamed. In a sector where personal charisma, aesthetic taste and a sense of stomach are so important, getting used to the cold-blooded calculation of a machine seems like a cry for help or acknowledgment that your creativity is lacking and you don't care about artistic projects.

Ruelens says that ScriptBook customers are some of the & # 39; s largest Hollywood studio & # 39; s & # 39; but confidentiality agreements (NDA & # 39; s) prevent him from giving a name. "People don't want to be associated with these AIs yet because the general consensus is that AI is bad," says Ruelens. "Everyone wants to use it. They just don't want us to say they use it." Queisser says similar agreements prevent him from discussing customers, but current customers are "big indie companies."

Some in the company push back against the claim that Hollywood embraces AI to watch potential movies, at least when it comes to actually approving or rejecting a throw. Alan Xie, CEO of Pilot Movies, a company that offers machine learning analysis to the film industry, says The edge that he "has never spoken to an American studio manager who believes in (AI) script analysis, let alone (has) integrated it into their decision-making process."

Xie says it's possible that studios & # 39; s simply don't want to talk about using such software, but he says that script analysis, in particular, is an inaccurate tool. The amount of marketing spend and buzz of social media, he says, are a much more reliable predictor of the success of a kiosk. "Internally at Pilot, we have developed box office prediction models that rely on script functions, and have performed significantly worse than models that rely on real-time social media data," he says.

Despite skepticism about specific applications, the tide could turn. Ruelens and investment director Scarso say that a single factor has convinced Hollywood to stop rejecting big data: Netflix.

Advertisements

The streaming colossus always brags about its data-driven programming approach. It monitors the actions of millions of subscribers down to the smallest details and knows surprisingly a lot about them – which thumbnail best convinces someone click on a movie to the choices they make Choose your own adventure style stories like Black Mirror: Bandersnatch. "We have one large global algorithm that is super useful because it uses all tastes from all consumers around the world," said Telf Yellin, head of product innovation at Netflix. 2016.


http://www.theverge.com/ "data-upload-width =" 981 "src =" https://cdn.vox-cdn.com/thumbor/dzxHh3Ma09XXaL8IVRBV8llwt5M=/0x0:981x590/1200x0/filters:focal ( 0x0: 981x590): no_upscale () / cdn.vox-cdn.com/uploads/chorus_asset/file/15790982/Screen_Shot_2016-05-03_at_3.59.46_PM.0.0.1462305612.png

Netflix regularly changes the thumbnails on TV shows & movies to see what appeals to different viewers.

It is impossible to say whether Netflix's cancellation is justified, but the company claims it recommendation algorithm alone is worth $ 1 billion a year. (It certainly doesn't hurt that such a speech confuses the competition.) Combined with his huge investment in original content, it's enough to reach even the most die-hard Hollywood producer for a reinforcing algorithm.

Ruelens says the transformation was noticeable. "When we started four years ago, we had meetings with large companies in Hollywood. They were all super skeptical. They said: & # 39; We have (dozens of) industry expertise. How can this machine tell us what to do ? "" Now things have changed, he says. The companies did their own validation studies, they waited to see which predictions the software did well, and gradually they learned to trust the algorithms.

"They are starting to accept our technology," says Ruelens. "It just took time before they could see it."