In 2016, film director Oscar Sharp approached A.I. researcher, natural language processing expert, and fellow NYU grad Ross Goodwin, about having an A.I. write the screenplay for his upcoming sci-fi short film. Together, they developed ‘Jetson’, a recurring neural network that ingested dozens of sci-fi screenplays and teleplays, as well as 30,000 pop songs, to produce the script and music that became his short film, Sunspring (https://www.youtube.com/watch?v=LY7x2Ihqjmc&feature=youtu.be).
In line with deep learning methods, ‘Jetson’, which eerily renamed itself ‘Benjamin’, through many iterations first determined which letters are more likely to appear in which order, then which words. It began to formulate original sentences, learning screenplay format and character dialogue to ultimately form its own script complete with odd stage directions (‘He is standing in the stars and sitting on the floor’) and a musical interlude. The outcome is a bizarre existential storyline that alludes to a love affair, a dystopian future, and is hilarious when you consider a machine wrote it.
I believe this raises some very interesting concepts. First, is a story (or a painting for that matter), objectively beautiful regardless of the source it is generated by, or is the source that originates the piece of art as important to our understanding of the art as the final medium itself? Would Van Gogh be as popular if we didn’t know about his life of turmoil, including chopping off his ear? Or do we just love pretty pictures? What happens if we do not know and find out later? I believe our feelings towards the medium will change.
The ‘art’ of AI writing is based on the chosen inputs. The screenplays and music chosen by the director and researcher of the project to be fed into this algorithm represent their own biases and therefore so does the output. So, what is it we are selling? Is it the final screenplay? Or the process that ‘these inputs were hand-chosen by these filmmakers’?
I think there will be a market for AI-assisted writing, but when considering the implications of a machine outputting a story by itself for a mass-market based only on historical text-based inputs will only be valuable, for now, as farce. But the speed with which it produces these works could help improve the workflow substantially for human writers.
Additionally, I think an AI analysis of successful (or unsuccessful) screenplays will be valuable when evaluating things like zeitgeist; the themes we as society are collectively interested or passionate about, and could serve as a predictor of what market conditions lead to what might make a story popular. For example, I would love to see inputs that include macro-economic factors such as recessions, political climate, war, employment rate…etc. It would be fascinating to see these factors regressed against what stories were the most popular at that time and received public accolade and why. Were vampire and zombie stories correlated to a recession? Did the type of vampire story matter? Do Disney stories perform better when we are at war? I think AI will have an ability to open up creativity instead of outsourcing it. But in the meantime, I think we will see some pretty funny attempts (the filmmakers have since tried again, this time casting David Hasselhoff: https://arstechnica.com/gaming/2017/04/an-ai-wrote-all-of-david-hasselhoffs-lines-in-this-demented-short-film/ ).