Hollywood has been pumping out movies about AI for years, from Fritz Lang’s Metropolis in 1927 to Alex Garland’s Ex Machina in 2015. Present throughout the majority of these is an undercurrent of fear and paranoia around the potential for AI to tear apart the fabric of society, a fear that many dismiss as bloody minded sensationalism. However, just because you’re paranoid, doesn’t mean they’re not after you. The negative impacts of AI that filmmakers have presented on screen likely seemed all too distant to them, taking the jobs of factory workers, not the creative brains that machines could never hope to emulate. Those that thought as much may find their complacency misplaced. AI is now being used to write scripts, and it seems that even the movie industry may not be as insulated from the coming AI tornado set to tear through the world of work after all.
Sunspring is a short sci-fi movie from the minds of director Oscar Sharp and AI researcher at New York University, Ross Goodwin, made for the annual film festival, Sci-Fi London. Although, to say it’s from their minds may be stretching it somewhat, as it was actually written by a machine called Benjamin, an LSTM (long short-term memory) recurrent neural network.
An LSTM recurrent neural network is a form of AI used most commonly for text recognition. Goodwin decided to use an LSTM algorithm because it has the ability to sample much longer strings of letters and can therefore predict whole paragraphs as opposed to sentence fragments. Most importantly, it can generate original sentences as opposed to merely copying and pasting.
In order to train Benjamin to create the science fiction masterpiece they were looking for, Goodwin fed a number of classic sci-fi screenplays from the 1980s and 90s into the network. Benjamin then analyzed these, looking for patterns so it could teach itself which letters and words tended to occur together so it could imitate the structure of a screenplay, replete with stage directions, prompts, and dialogue.
Sunspring stars Thomas Middleditch (Silicon Valley, Wolf of Wall Street) as a character called H, alongside Elisabeth Gray, who plays a character called H2. Humphrey Ker plays C, who seems to be the third wheel in a love triangle, although frankly the dialogue is so nonsensical that it’s impossible to tell. I would love to describe the plot, but it’s essentially nonsense. It starts strongly, with three characters - two men and one women - in what appears to be a hipster office. H, opens with the lines, ‘In a future with mass unemployment, young people are forced to sell blood,’ before pulling a book out of a drawer and thumbing through it. ‘It’s something I could do,’ he continues.
This is a strong declarative opening that sets up what could be an interesting premise. Maybe if they’d stopped using AI there, they could have seen it through, but unfortunately they persisted. One of the men spits out an eye ball at one stage and continues about his business as if nothing has happened, but other than that it’s essentially three people in a room spouting words seemingly at random. The chaotic nature of the dialogue stretches the use of intelligence in artificial intelligence to its limit. Sharp and Goodwin would probably have had more success if they’d just pulled words out of a hat. Some reviewers have called it dreamlike, but really it’s only interesting in that it has been created by AI. If it had been created by a human, the script would have been laughed at. One of the judges of the Sci-Fi London contest, Pat Cadigan, put it best, saying: ‘I’ll give them top marks if they promise never to do this again.’
Despite its flaws, it was still an extraordinary achievement, although there was still a significant amount of human input involved, from feeding in the scripts, to the actors interpreting the lines. To say Benjamin created a truly original screenplay alone would also be wrong, as it is still based on what other people have written. Theoretically, moving on from this, the project could be taken far further and likely create a far better script were the AI to learn from real world conversations and incorporate things it picks up in real life, as well as the formalities of script structure. Any device with a microphone that picks up speech based commands can become what is essentially an intelligent discovery system that can be used to understand the workings of the human mind, so there is huge potential for AI to learn a sufficient amount about humans to create a script indistinguishable to that written by a person.
In his novel Galápagos, in which he wrote about a fictional financial crisis which he blamed on the human brain, Kurt Vonnegut wrote about ‘that mystifying enthusiasm a million years ago for turning over as many human activities as possible to machinery: What could that have been but yet another acknowledgement by people that their brains were no damn good?’ Could a machine gives us a coherent film that gives us an insight into humanity and can touch the audience in its searing beauty? Machines’ ability to do everything better than us suggests that at one point in time, they likely will be. At the moment, however, the technology is still in its infancy, and it is likely to be many years before we see AI capable of producing an original screenplay that can really compete with one created by a person. As Benjamin himself noted of its own application of AI in an interview at the film festival: ‘It's a bit sudden. I was thinking of the spirit of the men who found me and the children who were all manipulated and full of children. I was worried about my command. I was the scientist of the Holy Ghost.’