I'm looking forward to people like Chris Cunningham exploring and exploiting the uncanny side of AI-generated audio/imagery/video, like the distorted faces at https://thispersondoesnotexist.com/. I'm sure they could do some unprecedented nightmarish things with it.
I think it is due to BD seeking precision from motors, whereas Wolfson has the robot tied to a pole so he does not need to compute stability, his robot can just wobble imprecisely for dancing. Which looks far more human because our muscles are not hyper precise and over/undershoot, and we either font care or adjust whole body plans to it in real time, instinctively.
He just sidestepped the precision (intermediate, artificial) goal :)
The economics of automation weigh heavily on what we create robots to do or represent. Something becomes cost prohibitive and we switch from getting it from people to getting it from machines. I used to argue what people called the thought experiment of the Turing Test, that determined whether people could be fooled by a machine - was really not about being fooled at all, but the representation of an Indifference Curve from economics, where what the machine produced was sufficient for the observer to be indifferent to whether they were being served by a person, or a machine, and they switch their choices accordingly.
The philosophical woo of Turing tests has been a multi-decade distraction from the clearer economic explanation that would have probably included more people in the discussion of what we should use machines to do, and the ethical responsibilities of the people who operate them.
The uncanny valley is just a threshold below the slope of indifference curve. It's interesting, but the preciousness of art often prevents us from looking at it as an expression of incentives.
The "Uncanny Valley" only exists in two dimensions.
In real life robots that look like "The Uncanny Valley" on the internet are fine.
Artists since they advertise in two dimensions use this 'trick'. But their work, like Ron Mueck's still looks cool IRL, it's just a different experience without the Uncanny Valley.
I'd like to know if anyone has actually made something that actually is the "Uncanny Valley" in 3 dimensions. I assume it's possible, but perhaps not. I'd be hard to advertise.
Perhaps you need to tap the part of the brain that extrapolates from 2 dimensions to 3 to get the horror.
> The "Uncanny Valley" only exists in two dimensions.
The original Uncanny Valley hypothesis came out of robotics back in 1970, its application to computer graphics came decades later.
I don't even think 2D vs 3D plays much of a role, a realistic Luke Skywalker wax figure will look just as uncanny as CGI Luke in Mandalorian.
I find deepfakes interesting in this regard, as while they look much more realistic than most CGI, they never go uncanny. They can go a bit glitchy, blurry and might not look like a perfect replica of the faked person, but they still end up looking like person, not like an uncanny CGI copy.
This feels more like “uncanny valley isn’t caused by what you think” rather than “it’s not a thing”. It’s simply easier to get proportions that we are most used to (IE average proportions of facial features across everyone we’ve met) when you go abstract, and when a lot of proportions look slightly off, it’s uncanny. And I bet it is close to the disgust reaction caused by a fear of disease as he states. I know he’s arguing uncanny valley isn’t a thing but my take away is that it’s more complex than how it’s usually stated
Maybe off-topic, but I can't help commenting on this sentence:
> it’s telling that no authorized English translation of the essay existed before 2012
I think the author was commenting on anglophones and a lack of interest, more criticizing, saying the anglophones are missing something. The word authorized changes that meaning, turning the criticism to the creator for not authorizing, saying the Japanese were missing something.
I can see how you could read it that way. My impression was a combination of two alternate readings, one that it was actually translated unofficially at least in part and/or two that the site never wastes an opportunity to describe how uncool they consider "piracy".
I think if you add a few more control points to animatronic faces with detailed texture coloring and hair and then if you can nail a human gaze behavior approximation, it will be convincing. Then the only creepy thing might be that you are looking at what is apparently a detached human head.
There was an experiment from Disney or something recently on simulating human gaze.
