A camera pans out on a hand pulling an unassuming book from a cabinet. The owner of the hand is revealed to be a man in a gold and silver lamé jacket, who looks at the book before he turns to speak to a woman – also clad in gold – who is working on an array of electronics at a bench opposite. ‘In a future with mass unemployment,’ he tells her, ‘young people are forced to sell blood.’ Over six minutes of garbled science-fiction ensue, as the opening frames of Sunspring give way to a short film loosely centred on a love triangle set on a spaceship, scripted entirely by an AI named Benjamin.
Sunspring was originally conceived by director Oscar Sharp and New York University AI Researcher Ross Goodwin for the 48-Hour Film Challenge at the annual Sci-Fi London festival in 2016. It is a slightly chaotic experiment made intriguing and enjoyable by its creator: a Long Short-Term Memory (LTSM) recurrent neural network. Benjamin, who named himself, is the same kind of AI that smartphones use for predictive text messaging. He was fed dozens of sci-fi screenplays and, using the ability to concoct phrases based on memory, generated his own work.
Beyond the futuristic Sunspring, algorithmic writing and idea generation is very much a current concept. And if two independent filmmakers were able to create and utilise a program like Benjamin, an omnipresent production company like Netflix would certainly be positioned to employ more sophisticated computer processing. The streaming giant already uses code to recommend films, by creating suggestions in the ‘Because you watched’ category based on each user’s previous viewing habits. It makes sense, then, that its original offerings are at least partially generated by algorithms: if it can create content that ticks as many cinematic boxes as possible, it can reach the maximum number of viewers.
Netflix has always used analytics when deciding whether to license certain shows or greenlight their production. For example, its critically acclaimed political thriller House of Cards was produced as a result of big data driving creative decision-making. In the New York Times, the late David Carr described the ‘Venn diagram’ behind the series: the success of the earlier British version; acclaimed director of The Social Network, David Fincher; and then-popular actor Kevin Spacey. Netflix had a direct line of communication with consumers based on their viewing preferences. In 2012, they crunched the numbers to produce House of Cards, the show that viewers wanted to watch before they even knew they did.
Netflix’s newer originals are less subtle. Two of their recent films – the would-be blockbuster, Bright, and low-budget holiday special, The Christmas Prince – are both overflowing with so many fail-safe motifs that they suggest some level of automatic manufacturing.
The former is a $90 million production starring Will Smith. Set in a fantastical world where humans, orcs and faeries coexist in an enduring class struggle, it centres on two cops, Daryl Ward (Smith) and Nick Jakoby (Joel Edgerton). They are begrudgingly partnered up and set out on a buddy-cop quest to find a secret weapon – which is a magic wand – to prevent a war, while learning about themselves and each other along the way. It draws on countless other films, from Alien Nation to Bad Boys to Lord of the Rings, without really offering anything new.
The Christmas Prince manages to cram in an even greater number of established tropes. Its heroine, Amber (Rose McIver), is an editor at a fashion magazine who gets sent as a writer to cover the coronation of the new prince of Aldovia. She is adorably clumsy, so the audience knows to relate to her, and sets off on a trite journey involving posing as a tutor to infiltrate a royal palace, get the scoop, and fall in love. It is eerily redolent of numerous romantic comedies and princess movies, from Never Been Kissed to The Prince and Me.
Both Bright and The Christmas Prince become such a mess of cinematic devices and previous plotlines that one almost hopes they were created by an AI, so as to be forgivable.
Benjamin’s Sunspring, on the other hand, is like the first tentative steps of a computerised Bambi on ice, endearing in its awkwardness because the viewer knows the truth of its creation.
Their questionable quality aside, are Netflix’s presumably algorithmic screenplays a sign that artificial intelligence is now capable of creativity to some degree? Is the creative writing industry at risk of automation? The use of AI to assist with all types of syntax is becoming more commonplace and it is not exclusive to film. Automated Insight’s Wordsmith program can churn through massive amounts of data to produce articles that can pass for being written by a human. Graduate business school professor, Philip M Parker, claimed in 2012 that he had developed an algorithm which allowed him to produce and sell over 800,000 reference books.
Yet the ability to mass-process information is not indicative of imagination. Even if lexes are computer generated, they can only be supplementary to creative thought. Wordsmith can produce statistically correct, data-heavy articles, but its prose is not inspired. Similarly, Parker’s algorithm casts a wide net over a range of data, producing mechanical reports on niche topics with titles including Soldering: Webster’s Quotations, Facts and Phrases, and The 2007 Report on Semi-Moist Dog Food: World Market Segmentation by City. Extremely efficient, but possibly a little dull.
Even supposedly creative AI rely on human collaboration. In screenwriting, as evidenced by the latest Netflix offerings, the capacity to comb through a backlog of popular cinema and combine the more crowd-pleasing elements is not a sign of creative ability. AI can only create from what has already been created, whereas the human mind searches through an alost infinite number of eclectic experiences, memories and thoughts, beyond a set catalogue of data.
Consider Sunspring. It is deliberately terrible, yes, but would have remained Benjamin’s unrealised vision without human collaboration. There is an allure in the madness of the written direction (he sits on the stars and his eyes are on the floor) and the dialogue is engaging in its abnormality (‘I don’t know, I’m going to talk to the skull’), but it is the cinematography and acting which make these elements so. Where Benjamin’s screenplay is fragmented, his human allies fill in the gaps. When he inexplicably dubbed two characters H – as he was unable to generate full names – Goodwin and Sharp named the second H2. When his script offers the line, ‘he takes his eyes from his mouth’, actor Thomas Middleditch interprets this as H regurgitating a plastic eyeball.
It seems that true creative ability remains a hurdle for AI, at least for the foreseeable future. Without human contribution Benjamin’s script would be little more than a messy generation of data, a random amalgam of preconceived ideas.
Which, when you think about it, is all it really is.
Image: Mario Klingemann/flickr