John Carter as an unmitigated box office flop - as predicted by AI. Photo / Supplied
It was easy to be optimistic about John Carter, Disney's grand 2012 movie adaptation of Edgar Rice Burroughs' Martian novels.
Despite a troubled gestation, it had a lauded director (Andrew Stanton of Finding Nemo), star writer (novelist Michael Chabon) and US$350m of Walt's money. But Black Swan, a UK artificialintelligence (AI) firm hired by Disney to predict successful films, thought John Carter looked like a flop – it was.
"But no one listened," says Steve King, Black Swan co-founder and chief executive. "People absolutely convince themselves... but the machines are always right – boringly always right."
Eight years on Hollywood is finally embracing AI.
Moguls were stung by the success of Netflix, which commissions, and cancels, shows based on data and which reports its 2019 earnings on Tuesday.
Disney and 20th Century Fox led the way, and now Warner Bros has signed a deal with Cinelytic, an AI company which claims to accurately estimate how well a film will perform with different combinations of actors, budgets, genres and themes.
"It's only starting to enter the mainstream," says Kathryn Arnold, a veteran Hollywood producer now consultant and expert witness.
"But as it becomes more user-friendly, as the price point comes down, any smart production and content creation company will have to look at data... Warner Bros will make it a lot more acceptable."
Data feeds the predictions and different firms have different tastes.
Black Swan is a trend-spotting company that ingests information from social networks such as Twitter and YouTube to see what people are talking about before the mainstream notices.
King says his system specialises in spotting outside chances which savants would not predict – such as 2013's Frozen, which "broke" merchandising supply chains with the scale of its success.
Meanwhile, in Belgium, start-up ScriptBook claims to be "democratising" storytelling by analysing scripts and predicting review scores and viewer satisfaction numbers as well as their box office returns.
"Hollywood is very much an elite... for a happy few [with] the right connections, and people are not getting chances," says its chief executive, Nadira Azermai, who claims an 86pc accuracy rate.
"We thought, 'Let's build an AI that actually will treat everybody equally'." She calls it a "misconception" that AI greenlights only the most formulaic films, saying systems often surprise experts by bypassing their prejudices.
Black Swan, which helps airlines decide which films to buy for in-flight entertainment, found that people watch more rom-coms when returning home than when leaving on trips.
Azermai says writers are shocked when software says how gender-unequal their script is. She is proud of having told Lionsgate unconventional musical La La Land would be a success, despite doubts in the studio.
Independent AI experts are sceptical. Lydia Nicholas, a researcher who has worked with the Government, the BBC and innovation think tank Nesta, warns that AI systems intended as "decision support tools" often end up being obeyed as decision-makers.
That would let studios "fob off responsibility" by attributing their risk aversion to a digital oracle.
Worse, she says box office returns represent "the ultimate polluted dataset" because of how biased and grubby Hollywood's decisions have often been.
AI trained crudely on such data might simply learn to reject any film that would have been sabotaged by Harvey Weinstein – a problem for the art of cinema but shareholders too, who stand to lose from blinkered decisions. "You'd never have another Marvel Cinematic Universe, because no one had done that before," says Nicholas.
Azermai argues things can hardly be worse: writers have an 0.003pc chance of having work picked up; scripts are often binned in minutes by agency interns on the basis of the first three pages.
She hopes to run an AI pitching service where writers would have scripts assessed for free (ScriptBook would take a cut of earnings) while studios and agents would search for highest-rated examples.
The movie business is a very slow adopter, says Azermai: "They need a good beating to wake up."
Given the intense secrecy in which Hollywood's use of AI is still veiled, and the arms race between studios and streaming services, it is likely the beating has begun.
Netflix's Biggest Hits: Five shows subscribers love
Stranger Things This 1980s-set sci-fi series reportedly pulled in almost 16 million viewers when its second season launched in October 2017. Its third came online last year.
The Crown The jewel in Netflix's crown, this royal drama reportedly drew three million US viewers for its season two premiere, and returned for season three, starring Olivia Colman, last year.
Fyre Released last January, this documentary charts the saga of the already social-media-infamous Fyre festival – sold (for a considerable price) as the luxurious 'biggest event in a decade', and in reality more of a nightmare scam.
Making a Murderer This true-crime documentary follows the story of Steven Avery, who spent 18 years in prison following a wrongful conviction, and was later convicted of a murder he also denies. Each episode is believed to have gained over 19 million viewers in 35 days.
Orange Is the New Black Debuted in 2013, this story of a middle-class woman sent to prison for drug smuggling has run for seven seasons, mixing themes of racism and brutality with dark humour.