If you have not yet seen the short film Sunspring, you’re missing something fascinating, bizarre, and potentially historic. It is a sci-fi short script written entirely by an AI named Benjamin. Specifically, Benjamin is a type of neural net called “long short-term memory” that is most often used for high end speech, handwriting, and text recognition. In the case of Sunspring, it was fed a few dozen classic sci-fi scripts (full list shown in the movie’s titles) and told to write its own short, which the human creative team then attempted to faithfully produce.
The results are… interesting, to say the least. While the stream-of-consciousness style of the language has drawn comparisons to the “cut ups” of William Burroughs or even some of the works of James Joyce, there is also a fair amount of straight up gibberish as well. In fact, what makes the film so interesting is that the majority of the meaning cannot be attributed to the “intent” of the AI author but rather the creative interpretation of the actors and directors. Sunspring is a type of collaboration between performers, viewers, and an AI all trying to pull together a coherent narrative by “reading the tea leaves” of the patterns common to sci-fi stories.
In many cases, these patterns are essentially tropes. The fact that an AI recognized this and incorporated it into a script is worth examining, as this seems to speak volumes about the genre itself. For the purpose of this article, I am choosing to focus on the gender narrative and what it says about sci-fi culture and the role of gender in the geek zeitgeist.
Since the language is based on pattern recognition and predictive algorithms, it’s sometimes hard to determine exactly what the story actually is, but the script clearly revolves around a conflict between two people struggling to relate to each other and a third causing conflict in their relationship. While Benjamin could not get the hang of proper name usage (a common problem with this type of AI), it seemed to subtly imply that the people were two men and a woman. The human team certainly took it that way and interpreted this as a love triangle.
In the film, a man called H (played by Thomas Middleditch) and a woman called H2 (played by Elisabeth Gray) are having a tense exchange about the nature of their relationship. H implies he has seen H2 with another man called C (played by Humphrey Ker) who then walks in. H2 and C are clearly having an affair and H is noticeably upset. H later puts a gun to his own head in a visually bizarre scene that may either be a hallucination or physical reality. H is then seen standing over C’s lifeless body as a door closes behind him of its own force. The film ends with an emotional monologue from H2.
Now I know that sounds confusing… because it is. The key action scene involves H seeing himself sitting on the floor of a star-filled room, grabbing the camera’s POV, and falling into a small black hole; it’s abstract to say the least.The script is a mishmash of classic sci-fi condensed into a four page script written by a predictive text engine; to say it’s open to interpretation is a massive understatement. But the nature of the interpretation the production team chose is significant.
While Gray’s solid acting and the strong nonverbal communication between her and Middleditch are what really drive the film, the nature of how the story came to be in the first place also says a lot about gender in sci-fi.
But what does it actually say?
First, there is the fact that when fed a large volume of these scripts, the AI assumed that it should be writing about relationships. The director (Oscar Sharp) interpreted this as Benjamin focusing on people trying to understand each other as they make sense of their increasingly complex world. There are, for example, numerous uses of the phrases “I don’t know” and “what do you mean” often in situations where one character thinks the other is implying something or doesn’t understand the context. While the actors often hint at comedy in these moments (as they literally do not understand what they’re supposed to be talking about) the AI seems to use it to imply that the way we understand others is key to the way we understand our world. It also suggests that a lack of communication and incompatible expectations are the real issue in H and H2’s relationship.
Apparently the relationship between Mulder and Scully from the X-Files was something that Benjamin focused on heavily when writing. The sheer volume of those scripts in comparison to most of the other source material actually caused Benjamin to spit out dialogue between Mulder and Scully in early tests, which may well explain where some of the relationship dynamic in the final script comes from. Mulder relies on Scully for emotional support and expects her to help him tackle various conspiracies, while Scully usually just wants to solve the crime and deal with the reality of their situation. They develop genuine affection, but that doesn’t change the nature of who they are, nor does it always define their relationship in the events of X-Files. Similarly, H seems to suggest that H2 isn’t making sense because her actions don’t fit H’s preconceived notions about what he expects her to be concerned about: his own struggle. This hints in some ways at the manic pixie dream girl trope. While H2’s characterization doesn’t fit that mold, the idea that she’s supposed to help H simply by being with him rather than living her own life does fit, and Benjamin seems to have picked up both on the trope itself and how strong female characters don’t play into it. The Alien movies are also part of Benjamin’s source material, and I like to think Ripley’s refusal to let others (usually men) dictate her best course of action is a partial inspiration for this trait of H2.
Second, and perhaps more importantly, there is the way humans breathed life into the words on the page. The basic story they filmed was: “Woman falls out of love with a man and takes a new lover, man considers suicide and kills (or fantasizes about killing) his romantic rival, woman is left to pick up the pieces.” But, as I mentioned, Benjamin didn’t understand proper names, much less the implied gender they often convey. Multiple different names, seemingly referring to the same person, were used in the original script. The director shortened them to single initials for clarity, but also assigned the roles. The AI picked names with phonetics that had subtle gender implications but the humans were ultimately responsible for deciding that H and C were men and H2 was a woman.
The opening scene presents H2 as somewhat callous, all but flaunting her affair in front of a visibly distraught lover. This evokes the trope of a man in love with a cold-hearted woman who doesn’t value that love. The closing monologue, however, paints a different picture of H2. It shows her to be someone just trying to live her life and being torn apart by the emotional self destruction of her partner, who views her as a character in his story rather than the protagonist of her own. In many ways she subverts the trope here. She’s not a plot device, she’s the main character trying to navigate the emotional minefield of male neurosis and emotional dependency.
The human creative team also chose to have the closing monologue be delivered by H2, even though it is not entirely clear if the script called for it to be spoken by her or a fourth actor. They decided that the subtext of the dialogue made sense only if coming from H2 based on their interpretation of the context. The film initially seems to center around H and his romantic rivalry with C, but the longest, most cohesive, and emotionally strongest piece of dialogue is the closing monologue by H2. The humans picked up on what appeared to be a trope in the AI’s story and decided that it only worked if the final word was delivered by H2, providing a sense of closure for the audience by having her address us directly and ultimately subverting the trope.
While there is thematic subtext in the algorithmically generated text, the emotional subtext is mostly created by the acting. Tone of voice is absolutely essential in making sense of the dialogue. “I don’t know what you’re talking about” for example can be said in a way that implies either genuine confusion on the part of the listener or that the listener thinks the speaker is full of shit. The closing monologue is a perfect example of this. Reading it in the script, sizeable chunks do genuinely come off as predictive text patterns; a line of dialogue relating solely to the line before it. But when Elisabeth Gray reads it, looking at the camera and crying on cue, she clearly paints a picture of a woman struggling with the emotional weight of a man who needs her but can’t accept her agency, a man who she loves but no longer likes, and the violence born of that incongruence.
“He was like a baby and he was gone. … I know I don’t like him. I can go home and be so bad and I love him. … He looks at me and he throws me out of his eyes. Then he said he’d go to bed with me.”
The machine suggests meaning, the human makes it clear. This is where the collaborative nature of the production really emerges. While much of the script is vague or even nonsensical, the acting conveys a sense of clear emotional context. As a result, the viewer is able to infer meaning even in those cases where it is not necessarily implied. The human team knew the source material and they read the script based on their knowledge of those same patterns and tropes the AI writer seized upon.
Perhaps what’s most interesting in all this is that the story is not explicitly about a heterosexual couple or even a romantic relationship at all. In fact, the only clue to the intended gender of the characters comes, somewhat ironically, from humans recognizing patterns in the names the AI chose. But the notions of gender roles and romance as core components to much classic sci-fi and the tropes born of those notions are so ingrained that they seem almost inescapable, even (perhaps especially) when being presented in a story written by a computer which only understands patterns, not meaning. Benjamin picked up on a trope and used it in the script, but humans interpreted that trope and decided how they would use it to tell their story.
I found Sunspring to be a genuinely fascinating experience. It’s a film that employs tropes as a shorthand to present an “average version*” of our sci-fi worlds and an acting team that subverts some of those tropes with subtle humor and intense emotion. It’s a story that seems to be a somewhat nonsensical tale about an emotionally needy man pining over a cheating woman, but ends up being a film about a woman who’s as sick of that trope as most viewers are.
In a way, this film could be said to be a meta allegory about trope in sci-fi storytelling itself—there are patterns we’re so used to that we endlessly repeat them even when we seek to disrupt them. In the Ars Technica article that accompanied Sunspring’s online debut, Sharp spoke about how reading Benjamin’s script made him more aware of tropes and how they are so prevalent in most sci-fi, remarking how he would catch himself having “Benjamin-like moments” when writing his own stuff. But to me the point runs deeper. There are obviously patterns in all human fiction (and history for that matter) but just because we’re aware of these patterns doesn’t mean we need to be trapped by them—fiction writers are not predictive text AI. Just because we read Joseph Campbell doesn’t mean we can’t break the mold in shaping those archetypes to avoid tropes. Sometimes a trope may serve as an easy and useful way to advance a plot, but often it’s just lazy, it’s up to creators to tell the difference. The human team involved in Sunspring does just that. When processing all that sci-fi, the AI set up a tired old trope about a cheating woman driving a man to violence, but the human team dismantled that trope by presenting the woman as the one whose story and perspective is the one that really matters.
I can’t help but think about the ongoing pushback in some quarters (cough Hugo awards voting) to the attempt to diversify the creators of the sci-fi content we nerds and geeks love. The bulk of
our current stories were written by straight white men and this has created a set of patterns so pervasive that a computer can detect it, even choosing a male name for itself. Perhaps more than proving an AI can write an interesting story, Benjamin has shown us that uniformity in our human storytellers is locking our stories into patterns and closing us off to the very thing good sci-fi is supposed to be about: new ideas.
The Ars Technica piece goes much deeper into the technical and creative process involved in making Sunspring. If you’re interested in learning more about Benjamin, the nature of how its writing was interpreted, or the film in general, I’d highly recommend it.
*=The term filmmaker Oscar Sharp used to describe Benjamin’s script, the “average version” of all the sci-fi stories it was fed.