The uncanny valley is a paradoxically attractive mythos to technologists. Despite being a theory that dates back to the steam engine era, we cite it regularly as a grounding ideology, accepting without question that there must be a future point at which robots will look so almost human that we’ll be sickened by them.
In reality, modern CGI and robotics have already pushed us well past the uncanny valley. No, that doesn’t mean we’ve crafted the perfect plastic humans yet. What’s changed is our reaction to these too-good-to-be-true fake humans; our visual culture has evolved past the point of disgust over eerily convincing human synthetics.
But in our obsession with the aesthetic uncanny valley, we’ve overlooked a different, deeper uncanny valley waiting around the corner. It’s not the way robots look that will bother us, but how these systems act. The new uncanny valley is the world of artificial intelligence.
In 1970, the Japanese roboticist Masahiro Mori proposed that robots could reach a point where they were just short of lifelike, thereby eliciting an unsettling, even shocking, response from actual humans. This was the “uncanny valley.” To work around it, Mori had some brilliant ideas. Want to craft anthropomorphic robot eyes that weren’t repulsive? Consider modeling them after eye augmentations we already used, like glasses. Want to create an articulating robot hand? Forget the faux skin and show its strings: Carve the hand out of wood.
The fundamental philosophy goes back even further. Mori himself was actually building off an idea of the “uncanny” from a 1906 essay by German psychologist Ernst Jentsch. Interestingly enough, while Jentsch didn’t specifically talk about cyborgs, it’s almost as if he understood how society would respond to their presence better than Mori. Here’s how Jentsch explained the concept back then:
Without a doubt, this word appears to express someone to whom something ‘uncanny’ happens is not quite ‘at home’ or ‘at ease’ in the situation concerned, that the thing is or at least seems to be foreign to him. In brief, the word suggests a lack of orientation is bound up with the impression of the uncanniness of the thing or incident.
Or put differently, familiarity is the antithesis to the uncanny.
Jentsch’s explanation is particularly important, because if you believe in this whole concept to begin with, you’ll see where Mori misinterpreted the role of the uncanny in society. To Jentsch, the uncanny was not necessarily perpetual. It was merely the unfamiliar, a real but malleable feeling he believed led to “mistrust” and even “misoneism.” That word "misoneism" is a particular tell. It means "hatred of change or innovation."
What Mori might not have understood in the year 1970 is that culture would soon shift, and misoneism would become an endangered species. We’d flower into a tech-loving society that would not only embrace the new but constantly expect it. The role of video games, born with PONG just two years after Mori’s famous paper, is particularly notable. In 1972, PONG was the first shot in a graphics arms race. By the mid-'80s we had side-scrolling adventures starring colorful, 2-D cartoon humans. During the '90s, these characters were rendered in 3-D. And into the new millennium, motion capture, high-resolution textures, and advancements in the physics of rendered light have many of us staring at unbelievably lifelike human facsimiles for hours every day. Yet with each new generation of console, we’ve only become more excited about the improvements in believable, virtual people.
There are moments in games like the PS3's Heavy Rain that truly are a spitting image for humanity, until suddenly the illusion breaks down. Are we repulsed when this happens? Do we throw the controller across the room and vomit? No. Heavy Rain sold over 2 million copies because we’re amazed. We recognize something inhuman is trying to look human (precisely what Mori warns us against), and we simply don’t care. Are our graphics good enough to regularly cross the valley without us even noticing? I don’t believe so. I think the cavernous drop of the uncanny valley was filled, from the mid '70s on, by the ever-changing tastes of a tech-rabid society.
While we’ve been so caught up on Mori’s aesthetic arguments, however, we’ve lost sight of a more fundamental paradigm shift: artificial intelligence. In many cases, like Siri’s "natural" speech cadence and propensity for humor, today’s AI is overtly mimicking our own tendencies. It shows that the goal of AI won’t be to shy away from the valley but to embrace it head-on.
To reach the valley in artificial intelligence, just as we’ve seen in robots and CGI, software’s first step to being human is honing all of its corporeal traits (does Siri’s voice actually sound like a person?). But this surface-level polish is just the beginning. Because once software knows how to sound and look alive, it also needs to master the casual logic of human communication.
Currently, AI’s greatest shortcoming is that it doesn’t recognize when it’s being extraordinarily dumb. Consider when Apple Maps directs us to a Starbucks that’s really just an empty field, or when Siri can’t comprehend the most basic request, partly because she is unable to recognize her own error. It’s an idea I might summarize as "can a machine understand questions even when it doesn’t have the answer?" And it’s an incredible challenge to overcome—one that IBM’s Watson never quite figured out, despite wiping the floor with Ken Jennings in Jeopardy.
With common sense logic in place, only then will we witness AI’s inevitable uncanny valley--social interaction. Should Google Now speak to us like a naive child, an omniscient god or our obedient dog? Should Siri ever interrupt? Ironically, it’s a problem that programming expertise may only make worse, compounding as companies become better and better at leveraging big data to discern our individual tendencies before understanding how to properly address them. It’s Target knowing when its shoppers are pregnant before they do, or Google discerning flu patterns based on the way sick people search. Imagine if software scheduled your mammogram after calculating the risk factors involved in your last month of behaviors (cross referenced with your genetic makeup, of course)--how would Siri handle that conversation?
We’re reaching a cliff of AI, where the height of human knowledge falls off into a wasteland of poorly automated social grace. And where we may theoretically get used to idiotic, omniscient AI--much like we did the PS3's glassy-eyed depictions of the human figure--there are numerous social concerns that software (let alone society) has only begun to address. For instance, when should you give someone bad news? How should your intonation change based upon context? Or, maybe most challenging for the Googles, Apples and Facebooks: Do you still tell someone the truth when you know they won’t want to hear it? Or when they simply can’t do anything about it?
In other words, Mori was right about the uncanny valley. He was just wrong about where it might affect us most. We can become habituated to the quasi-human figures in video games. In fact, we already have. But a piece of software that’s simultaneously smarter than us and entirely clueless? Is it infuriating or horrifying? Take your pick.
Thanks to Cliff Kuang for contributing many of the core ideas in this piece.
[Image: Robot, Roberto Ferrari via Flickr]