What Is the Uncanny Valley and Why Does it Matter?
What Is the Uncanny Valley and Why Does it Matter?
How technology is bridging the gap between reality and fantasy
The term “uncanny valley” is one that’s become more popular in recent years. And while many understand what sorts of scenarios the phrase applies to, others still aren’t yet familiar with it at all.
The original concept of the uncanny valley is attributed to Japanese roboticist Masahiro Mori, who introduced the term in 1970. Uncanny valley refers to the idea that a being’s similarity to a human elicits a sense of unease or revulsion among observers when the resemblance is nearly — but not perfectly — complete. It’s a term that’s often been used in reference to the world of entertainment, videogames, VR, and most recently, the Apple Vision Pro headset.
To circumvent the inhumanity of a helmet on heads where eyeballs once were, Apple implemented a feature called EyeSight. Essentially, it grafts lifelike eyeballs to the front of your headset that move in time with your real ones. In the virtual world, when you FaceTime with others, your real-life expressions are overlayed on top of a fully digital approximation of your face and body. Among those who’ve tried the service, “uncanny valley” has been a recurrent descriptor. In the world of AI and image generation, too, it’s a concept we can see at play.
For those who knew what the term uncanny valley referred to by context, many weren’t aware that it’s an idea that can be visually represented through the peaks and valleys of a graph. The following chart — where the horizontal axis represents the degree of an object’s resemblance to a human being, and the vertical axis represents the observer’s affinity for that object — offers a clear glimpse into where precisely the “uncanny” valley occurs.
The graph shows a dip in the observer’s comfort level — a “valley” — at the point where a figure’s resemblance to a human is almost complete, but where enough small discrepancies remain for it to feel unsettling.
Regardless of our awareness of the principle in action, it speaks broadly to the way in which humans experience fear. In every mythic half-man or humanoid tale is the intrinsic distaste we all have toward the not-quite-human. Whether lifelike dolls with deadened stares or the mannequins that line department store aisles, the majority of us can attest to some uncanny valley-rooted aversions.
Experiments have proven that most of us experience a similar fear of the not-quite-familiar. According to one study, it’s a phenomenon that affects even primates.
On one hand, the uncanny valley effect is so common that it underpins our fears of everything from zombies and dead bodies to vampires, clowns, werewolves, yetis, bigfoots, and aliens. It’s a deep-rooted phobia and not one easily overcome.
But there’s a place at which the valley ends and even the non-human enters into a realm that most of us would deem acceptable. If the imposter looks authentic enough, it can climb free from the crater and walk the earth with the rest of us. It will fool people from near and far. The things it says and creates will pass as the products of real human thought and ingenuity.
The uncanny valley isn’t a gap our artificial intelligence has unilaterally crossed yet. As AI image generators attempt to create people, most will notice a not-quite-right component to what they produce. Hands won’t be depicted accurately. Faces will be distorted. The more closely you look at each person portrayed, the more details you’ll notice that are amiss.
As OpenAI, the company behind ChatGPT, prepares to release its first video-generation software to the world, the conversation about this valley is more relevant than ever before. Going by the name Sora, the company’s latest invention appears to be yet another program poised to change the world — to take us another step closer to the singularity.
From written prompts alone, Sora can generate lifelike videos in seconds. The videos they’ve released to the public so far have been nothing short of stunning.
But in those videos where it’s attempted to accurately depict people, the software often still seems to struggle. There will be oddities to expressions and movements.
Yet, we’ve mad-dashed onward from where the technology was only just a year ago. It’s disconcerting how far we’ve come in so little time. From a distance, AI videos are able to trick the common viewer already. A year ago, they simply weren’t.
As the world faces mounting political turmoil, the introduction of convincing deepfakes into the mix will be enough to decide entire elections. It could dictate the rise and fall of regimes. Only months after AI began entering into our lives, people began facing real-world consequences for the actions of their deepfaked personas.
Artificial intelligence can’t reliably cross over the chasm that is the uncanny valley just yet. But it’s aware of it and tries its algorithmic best to bridge that gap. It’s getting better at identifying the point at which something looks genuine enough to convince people that fiction is fact; it aims to hit that target with a robotic precision.
When you’re looking for AI and know what to expect, it’s not always a challenge to identify its work. But for the masses less familiar with the technology, they’re liable to be fooled already. A few years from now — judging by the technology’s current rate of progression — we’re likely to lose the ability to discern what’s real as a society.
Stilted, formulaic writing will begin to appear less so. AI will begin to embody what appears more and more like consciousness. It will draw from what looks like experience. It will develop more of the traits and features that we consider uniquely human.
AI voice imitation will grow more accurate. Leaders will appear to do and say things that they haven’t. Our loved ones might, too. All of our likenesses and reputations could be at stake and we’ll be powerless to prevent them from sliding out the door.
How the world will make sense of all of the noise is a mystery. But as the uncanny valley continues being crossed with greater and greater ease, the cacophony will affect us all. Whether we’ll be able to hear or see it for what it is remains anyone’s best guess.