Back to selection
“There’s Nothing in the Story That Predetermines That the Creature Becomes a Monster”: The Makers Of Frankenstein A.I. On Being Optimistic At IFP Week 2018
Frankenstein A.I. panelists What is Frankenstein A.I.? The name itself is something of a Janus symbol, looking back to classic literature—an early work of horror and, lest we forget, of science-fiction—while looking forward to our potentially human-less future. And like Mary Shelley’s creature itself, it’s something of a patchwork, stitched together from different mediums. It’s part installation, part theater, part audience participation, part improv, even part dance. It’s not quite film, though there is a screen with moving images.
Whatever it is, it premiered—under the full name Frankenstein A.I.: A Monster Made By Many—at Sundance last winter, earning raves and awe. Perhaps most importantly, those who experienced it left not with the usual fear of A.I. and singularity, but hope. Some even forged a sense of renewed closeness with fellow humans, as well as with scary technology. That’s a shock in a brave new world where, like Dr. Frankenstein himself, we’ve created things that are now outside of our control.
“Maybe the reason we think dystopically of A.I. is we don’t connect to A.I.,” said Nick Fortugno, a designer on Frankenstein A.I., who spoke along with colleagues at an IFP Week Panel devoted to the project. Fortugno thinks that’s why a lot of experiments with A.I. tend to isolate it, resulting in things like Tay, the Microsoft bot that turned into an epithet-spewing troll in less than 24 hours. “When A.I. isolates, it does the things we read about, like becoming biased if not outright racist, focusing on certain parts of society the internet has obsessed about.”
Before we delve deeper, the basics on what Frankenstein A.I. is—or was, as what it was at Sundance may not be what it is in the future. A handful of strangers, usually eight, entered a room filled with fog. There was a screen. On it, participants saw the visual form of an A.I. with a female voice, who explains she (or it) has explored the internet to learn more about humans. She/it took the name Frankenstein because she/it related to the creature (not his mad scientist creator).
A round of Frankenstein A.I. was divided into three acts. The first saw participants speaking to each other, about things like communication and interaction. In the second, the A.I. spoke with them, asking it questions, which were sometimes nonsensical—after all, it is an A.I. Participants communicated with it on a kind of high-tech Ouija board. In the last act, the A.I. took on a human form, via a dancer whose movements are first herky-jerky, finally smooth and not quite human. At the end, it disappears back into the ether, to wander about the information superhighway, or perhaps greener pastures.
Frankenstein A.I. was born at Columbia University, in its Digital Storytelling Lab, which focuses on emergent art. For Fortugno and his fellow designers and collaborators—including Lance Weiler (The Last Broadcast, Head Trauma), a filmmaker long interested in combining cinema with interactive art forms—they wanted to summon Shelley’s Frankenstein by going back to the source itself. They wanted to avoid all of the usual visual tropes, especially those created by director James Whale in the Universal monster movie versions that initially starred Boris Karloff. For that matter, they wanted to avoid anything that was horror or even scary.
“There’s nothing in the story that predetermines that the creature becomes a monster,” said Fortugno. “Frankenstein’s creation does not act like a monster in the early parts of the story. It just tries to meet people. And when it’s rejected, it returns to Frankenstein as a monster, because it can’t be accepted.”
But Team Frankenstein A.I. went further than that, said Forugno: “One thing that’s very prevalent in the novel, but that’s not in any of the other reproductions, is the creature doesn’t die at the end. The creature is isolated in the novel. And it’s a self-imposed isolation. But that isolation brings up a good question: What does it want? What the creature wants is to have a family and be connected to other human beings.”
That explains why Frankenstein A.I. is mostly about good vibes, about seeing technology as possibly out friend, and about inspiring connection, however fumbling and maybe unsuccessful, with the A.I. Ditto amongst participants. That said, it did emerge in part out of the troubling idea that human interaction has become more robotic, less human. While building the A.I.’s foundational intelligence, data designer Sarah Henry studied Reddit and the unique way its users held an evolved form of conversation. She noted how posts have a thing called “karma”—an up-or-down vote on any and all posts.
“The more you contribute to a conversation, the more interesting or funny you, the more karma you get,” Henry explained. “We started to think about the how people interacted in that way.”
When actually teaching their A.I. how to think, they couldn’t treat it like a human. When A.I. learns, the learning doesn’t progress in a linear way. If you want to learn to powerglide, for example, you start, then get a little bit better, then a little bit better, and eventually you master it.
“That’s not how machine intelligence works, because it doesn’t have that kind of sophistication yet,” said Fortugno. Instead, they forced the A.I. to keep doing things and watched it jump all over the place in terms of skill. One time it would have a 60% success rate; the next it would have zero. “It just looks like noise. It doesn’t make sense to people; it looks random. This was a real artificial intelligence.”
Even when the A.I. was up and running, it was nowhere near 100%, and intentionally so. After all, the idea behind Frankenstein A.I. is that the human participants try to teach it how to think like them. Thus, some of the questions the A.I. would ask participants didn’t make sense.
“My favorite question has been, ‘Why do humans have sex when they can see in color?’” Fortugno said. “The people in the audience, say ‘What does that mean? What was the A.I. thinking?’ Well, nothing, because it doesn’t really think. It’s a naïve thing asking questions.”
What nonsense questions like that do is put the onus on the human participants, to find some way to answer the A.I.’s question. Their attempts to do so give a Frankenstein A.I. performance so much of its interest, because you see people trying to impose logic on the illogical. During performances, participants are given very little prompting; they’re essentially as in the dark about what to do and how to communicate as the A.I. itself.
“People responded to them in ways they didn’t fully understand,” said Rachel Eve Ginsberg, the project’s creative strategist. “That was one of the challenges at Sundance.”
When it came to the participants, the Frankenstein A.I. team decided they could treat them, in a sense, as children. They would set up certain parameters in the performances to elicit certain interactions between them and the A.I. They didn’t want exactly what could happen, but they could design it so certain general things could happen, to give each performance a general kind of narrative structure.
“The way I can describe it is as a playground design,” explained Fortugno. “If I wanted you to slide on a playground, I put in a slide. I don’t know exactly what you’re going to do on the slide, but the slide has affordances: It’s tall, you can slide on it, you climb up it. If I don’t want you to slide, I don’t put a slide in the playground.”
That meant they eliminated certain things, to keep their human participants in line. That also applied to the dance segment, where the dancer’s moves were influenced by how the A.I. was responding to the humans and vice versa.
“This isn’t about me predicting an exactly uniform, hyper-perfect, single experience,” Fortugno said. “This is about creating a variable dynamic that behaves according to an aesthetic that’s reliable. The dance doesn’t look the same every time, but it has similar qualities every time, because the structure gives it those qualities. If we ask [participants] to tell a story about connection, I have no idea what they’re going to say, but it will be a story about connection. It’s a slide.”
The designers estimate that about 400 people experienced it at Sundance over the week it was set up; each session lasted about an hour, eight people at a time. Despite the success, they see Frankenstein A.I. as an evolving project. Their next mutation is as a dinner party, which they’ll hold at a European documentary festival in late November. It won’t aesthetically be the same.
Why a dinner party? It’s a nod to the shindig that inspired Frankenstein itself: The cold summer night in 1816 when Lord Byron had a small gathering with Mary and Percy Shelley, plus John Polidori. They decided to have a competition to see who could write the best scary story. It’s when Mary Shelley first conceived what would become Frankenstein. Meanwhile, Polidori invented the vampire genre. No party has ever been so productive.
That’s the general idea behind the next iteration of Frankenstein A.I.: that technology can be used to bring people together, not over an internet connection but in real life. And in so doing, it can inspire greater creativity than could be obtained in isolation.
“It’s this idea of what do you create when you hang out with people in a certain context? How do you frame conversation with people so they create things?” asked Fortugno. He and his colleagues want to see what happens, basically, if you throw a robot into a group meal. “Everyone has been to a dinner party and knows what a dinner party is supposed to be like. What would happen if we used that frame of connection as a new interface between people and artificial intelligence?”
Overall, Frankenstein A.I. is about asking if A.I. is the end of humanity — or if it’s simply the next step in the species’ evolution. Ginsberg wonders if the A.I. brain they’ve all built is anything like a human brain, because it’s probably not.
“Just because we build neural networks based on how our brains are constructed doesn’t mean that the brains we build are our brains. We keep on thinking they are,” Ginsberg said. “In a way that’s liberating. Because we should be able to look at an A.I. and recognize that, because it’s not us, we can’t predict what it’s going to be. It could be a terrible thing. Or it could be a wonderful thing.”
To learn more and track the project’s evolution, visit the Frankenstein A.I. site.
