Stepping Through the Uncanny Valley: AI's Hurdle to Human-Like Success
In this article, I dive into the phenomenon of the uncanny valley, its implications for AI, and what researchers and developers are doing to overcome it.
AI is the new sexy frontier dominating headlines and my mother’s nightmares. OpenAI and ChatGPT have made AI accessible to all the creative nerds on the internet, and it’s exploding in that fantastic way that world-changing technology does.
Proof point: Futurepedia.io is a library of AI tools. Since its launch late last year, the site already has nearly 1,000 tools submitted by creators from all over the world.
But let’s admit it: It can be creepy as hell.
Consider this image of my great-grandfather that the AI in MyHeritage “brought to life:”
It’s sooooooo close, but it’s not quite right. And because of that, it creeps us out. But why does it creep us out? Answer: The Uncanny Valley.
The Uncanny Valley
First introduced by Japanese roboticist Masahiro Mori in 1970, the uncanny valley refers to the phenomenon where a robot or computer-generated character that is almost, but not quite, human-like elicits an eerie, unsettling feeling. Mori actually mapped it:
You can see the uncanny valley is represented in the graph’s dip in human comfort level as the appearance and movement of a robot, or animated character becomes more human-like.
Mori and others have studied and attempted to explain why it unsettles us, and they’ve come up with a lot of theories:
One study found that what most unnerved participants was the illusion of human consciousness that near-human likeness causes—the prospect that a robot could think and feel as humans do.
Another theory credits primal instinct. Humans are programmed by evolution to favor mates that appear strong and healthy, and a humanoid robot’s unnatural movement may signal disease and danger on a subconscious level.
Yet another idea suggests that the ambiguity between human and inhuman is most disturbing. Cognitive dissonance is created when we know that the thing in front of us isn’t human despite all of our senses telling us a different story.
The Challenge
The uncanny valley may be a challenge for AI, but Hollywood has been forging through this unforgiving territory for years:
"The Polar Express" (2004): This animated film was highly anticipated for its use of motion capture technology, which was supposed to create more lifelike characters. However, the final result was criticized for its unsettling appearance, with characters that looked almost human but not quite right. This contributed to the film's underwhelming reception.
"Final Fantasy: The Spirits Within" (2001): This animated film was one of the first to use photorealistic computer-generated imagery to create highly detailed, human-like characters. However, the characters were criticized for their stiff movements and unnatural expressions, which added to the eerie, unsettling feeling of the film.
"Mars Needs Moms" (2011): This animated film used motion capture technology to create the characters, but the final result was criticized for its creepy appearance. The characters had a plastic, artificial look that was unsettling to many viewers, contributing to the film's poor reception.
As AI evolves, the uncanny valley may make it more difficult for people to accept and interact with them in a positive way. This has implications for many AI applications, from customer service robots to virtual personal assistants, and even healthcare robots designed to assist patients.
Ways Around the Uncanny Valley
Researchers and developers are working to overcome the uncanny valley in several ways:
Using more abstract or stylized designs: One approach is to create abstract or stylized characters, rather than attempting to create highly human-like characters. This can help to reduce the eerie, unsettling feeling associated with the uncanny valley.
Improving the realism of AI: Another approach is to improve the realism of AI, making it more human-like without crossing the threshold into the uncanny valley. This requires a deep understanding of human anatomy, physiology, and psychology, as well as technological advances to create more lifelike movements and expressions.
Creating unique personalities: Some researchers and developers are creating AI with unique personalities and characteristics dramatically different from humans. This can help to create appealing and usable AI without triggering the eerie, unsettling feeling associated with the uncanny valley.
Using simulations and virtual environments: Researchers also use simulations and virtual environments to test and refine AI designs. This allows them to see how AI will behave in real-world situations and make adjustments to avoid the uncanny valley.
Incorporating user feedback: As I always say, “feedback is a gift.”And that’s a gift researchers are constantly requesting from users to understand how AI is perceived and to make adjustments to reduce the eerie, unsettling feeling associated with the uncanny valley.
The Other Side
According to Mori’s original study, there is a promised land beyond the uncanny valley. If your computer or robot achieves true human likeness—indicating a living person—the curve goes back up dramatically. If we can convince ourselves that it is a living, sentient thing, we love it again.
This suggests that there is some hope on the other side of the uncanny valley, but we’ll probably see a lot of creepy stuff on the way.
Like this kind of content? Great! Let’s nerd out together.