Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:05):
Hey, welcome to Stuff to Blow Your Mind. My name
is Robert Lamb and I'm Joe McCormick, and it's Saturday.
Time to go into the vault. This time we're bringing
you part two of our exploration of the Uncanny Valley.
This episode was originally published April six. Uh, should we
jump right in, Robert, Let's do it. Welcome to Stuff
(00:28):
to Blow your Mind from how Stop works dot com. Hey,
welcome to Stuff to Blow your Mind. My name is
Robert Lamb and I'm Joe McCormick. In Today is going
to be the second part of a two part series
we're doing on the Uncanny Valley. Last time we ventured
into the Uncanny Valley, so if you haven't heard that episode,
(00:50):
you should go back and listen to that. First, we
lay a lot of the groundwork for what we're gonna
be talking about today, But we explored the origination of
the concept of the Uncanny Valley, what it means means
to to be in the Uncanny Valley, and some research
on whether this valley actually exists or not. Today, I
think we want to start off by looking at if,
if it does exist, what might be some explanations for it. Indeed. Yeah,
(01:14):
we're gonna we're gonna dive into it a bit more
and move as the title, uh suggests, beyond the Uncanny Valley.
But before we do that, I do want to talk
about RoboCop. Of course you want to talk about RoboCop,
because we talked about RoboCop pretty much every day. Yeah,
it's it's it's an important film. Important films, I'll say,
at least the at least the first two, arguably the
(01:35):
third one to throwing the TV series if you like.
But there was a TV series, Yeah, RoboCop TV series.
It was one of those that would I think it
would come on sci Fi or it came on various
cable channels. I only have a vague awareness of it
because it seemed to be a far lower key RoboCop
type show. Okay, So in many of the studies that
(01:58):
we talked about in the last episode, they were looking
at largely three categories of of robots and humans and androids.
So you had pure robots things that are just undeniably
machines and we're mostly okay with. Then you have humans
or or or perfect human replications. Okay, So you look
(02:18):
at it, it either is a human or it's such
a good representation of a human ideally that you would
not think that it was a robot. Right. The third
category here is where you're going to get into the
danger zone, right, the human like robots, where you look
at it and you say, I see what you're going
for there, but it's creeping me out. So I think
(02:39):
it's interesting to align this up with the Holy Trinity
of robocops. Okay, so this will mostly make sense if
you've seen the RoboCop films, but I feel like most
people know what we're talking about here. First of all,
you had the proto RoboCop and two oh nine. Oh,
this is maybe the greatest scene in the first film
is before we get a humanoid RoboCop, they just have
(03:00):
this big drone object that is supposed to enforce the
law and it ends up shooting somebody in the boardroom. Yeah,
it's a walking law enforcement tank with a robot commanding
robot voice, so that you look at it and there's
no to not know this is not a friendly device.
But it's not really humanoid at all, not really at all.
(03:20):
It hit just walks on big, big legs that can't
even navigate human stairs. But I would say, I have
great affinity for ed to oh nine, Yeah, I mean,
you know, without having to worry about it actually shooting me.
It's kind of cute in a way. Now, then we
have RoboCop itself himself, the classic RoboCop, the Peter Peter Weller,
(03:43):
and uh, he is a cyborg or perhaps an android,
depending on how you want to view the descriptions. So
he's he has a relatable, living human face which is
a fixed to to honor him in some some explanations,
or has to make him more comfortable not only as
a police killing machine but also a community law enforcement officer.
(04:05):
So RoboCop, you know, moves around with very robotic movements,
speaks in a very robotic voice, but his face is
a living human face. So in a way like that,
that seems like it might it just sort of you know,
to read perhaps more into the original film than was intended.
Perhaps this was a way to to get beyond the
Uncanny Valley. We can't replicate the human face, We'll just
(04:28):
get an actual human face from the dead cop and
just plaster it up there. Oh but I'd say RoboCop
with his mask on really does kind of get into
the uncanny Valley and and Weller does some good work
forcing us in there, Like I think they're the sort
of going for it. Yeah, okay, Well that brings us
to the next generation RoboCop two, which is not just
(04:50):
the name of the second RoboCop movie, but also the
model of RoboCop that replaced the original RoboCop. They think, Hey,
what would happen if we put tom noon in there?
That's right, So they have another essentially a walking tank
kind of like d to O nine, except it's powered
by the brain of a psychotic drug lord named Kane
played by Tom Noonan. He's fabulous in it. Uh, but
(05:13):
here's the here's the thing. It's walking around, it's killing
everything with a gatling gun, but then it can pop
a flat screen TV out of it out of the
front of its body, and on that screen is a
twisted uncanny lawnmower Man esque c g I face of
tom Noonon. Yeah, so that one really leans into the
Uncanny Valley. Well yeah, and this this does point out
(05:36):
another thing, which is that there have been plenty of
intentional realizations of the Uncanny Valley in film when when
people are trying to create an unsettling, unpleasant, humanoid for
story purposes, if it's supposed to be a villain, if
it's supposed to make people uncomfortable, because that's its role
in the plot. So one thing I kind of wish
we'd done. I hadn't even thought about this is too
(05:58):
if we could talk to somebody who has intentionally made
things in the Uncanny Valley? What did they do on
purpose to get it there? If your your job is
to make a humanoid robot or an animated humanoid figure
that intentionally pushes all the bad buttons and climbs as
far down into the valley as it can, what do
(06:18):
you do that would that would provide some really interesting
insight into what it actually takes to get there. Well,
you know, to come back to video games, something that
comes up a lot. You see these videos going viral
where it's like just a cut scene or a clip
from the video game with a humanoid character in a
more or less human situation, except something screwed up in
(06:39):
the face is missing, so it's just two floating eyeballs
and maybe a floating set of digital teeth. So the
context is key there, Like this thing is acting as
if it had a face, and it's in an environment
where we're unsupposed to just roll with it, but clearly
something's wrong. Yes, okay, Well, as we said in the
last episode, we talked about the origin of the idea,
We talked about evidence for and against the fact that
(07:01):
the Uncanny Valley actually exists, to to the point that
it actually does exist to some extent. Maybe not in
the naive original sense everybody would say, where it's just
related to the amount of realistic human nous in a figure,
but has other dimensions as well. What causes our Uncanny
(07:22):
Valley reaction? Obviously people do have this reaction they see
a humanoid robot or a humanoid animated character, the Scorpion King,
the Final Fantasy, the spirits within characters, whatever it is,
and we get so, what causes it? Why do our
brains react that way? Is it biological? Is it pure instinct?
(07:42):
Is it a learned psychological reaction? Is it part of
our culture? Is something coming from cognitive dissonance of some sort? Now,
to go back to the origins of the idea with
massive hero Mari in nineteen seventy and his original paper
on the Uncanny Valley, Morey speculates that the Uncanny Valley
might be a side effect of the self preservation instinct.
(08:04):
In other words, it's a biological adaptation that helps us
avoid disease and death. Uh. And he starts with the
observation that when a normal, healthy person becomes sick and
eventually dies, they basically tend to slide down from the
second peak of the uncanny valley, what we were calling
the realism peak, the reality peak, and slide down into
(08:27):
the uncanny valley. It's like they become more like these
upsetting puppets and robots. You know. They might uh suffer
some kind of change in their in their the the
appearance of their skin, of their eyes, of their facial expressions.
Things begin to look off about them. And he writes, quote,
the sense of eeriness is probably a form of instinct
(08:49):
that protects us from proximal rather than distal sources of danger.
Proximal sources of danger include corpses, members of different species,
and other entities we can closely approach. Distal sources of
danger include windstorms and floods. You know, this is interesting
that the mention of other species. Um. I don't know
(09:11):
how many films and documentaries I've seen of say, lions
running around in the natural habitat, and it's almost never creepy. No,
not not at all, and yet and then mostly a
lot of times when I'm in a zoo, it's not creepy.
But there have been times I take my son to
the zoo a lot here in Atlanta, and there are
times when we go down to the lion enclosure and
(09:31):
we're the only people there at the zoo because we've
arrived super early, and we just we're hanging out with
a lion on the other side of the glass. You know,
we're perfectly safe, but a deep uneasiness comes over me,
washes over me, and uh, and I'm just confronted by
the the the danger of this situation, Like there's a
(09:52):
danger that that goes beyond any reason because I am
in close proximity to a dangerous member of another species,
a carnivorous um predator, tory animal that would, in under
normal conditions, potentially kill me. Now there is I would
bet a very different kind of sensation going on inside
you when you're in proximity to a lion than when
(10:16):
you see a creepy looking humanoid robot or a creepy
looking animation human animation, right like it's it's probably exciting
the difference between the when when we talked about the
creepiness episode the creep the difference between the sort of
like uncomfortable threat ambiguity versus sensing that there truly is
a threat of some kind. Right, And you know, we
(10:38):
spoke to illness here, you know, like certainly someone we
talked about like what happens when a co worker walks
up to you and they're they're sniveling or more a
little bit pale, uh and they say, let's French. Well,
well hopefully not, but yeah, there's a there's an at
least an initial sense of, oh, what's wrong with this person? Um?
I wonder if what they have as contagious, should they
(10:59):
even be at work? I hope they cover their mouth
when they sneeze, because we are going to be concerned
on some level with catching whatever they have. If it's
if it's it's transmittable, right and it I mean there
are some ways in which we know that seeing illness
and other people in sights a reaction in the disgust response, right. Yeah.
And there's been actually a lot of research on the
(11:20):
discussed the disgusted response. Darwin wrote about it discussed from
an evolutionary standpoint, is all about disease avoidance, you know,
like do not eat this, you know, stay away from Yeah,
something's not right here. You might hurt yourself or get sick.
So we're talking about sensuous disgust, So discuss that's tied
(11:41):
to the senses to sense information that we're absorbing. Uh.
This is deeply seated in the insula, the area of
the brain that that malfunctions in patients with obsessive compulsive disorder,
causing them to say, you know, wash and clean things
endlessly or vacuum unrelenting lye. So the malfunk of that
area gives us, you know, gives us a clue into
(12:02):
its functionality. Uh. One interesting fact about disgusting smells, however,
is that there's a drop off point for bad smell recognition,
but not for good smells. So I think we've all
encountered this where it's that like say, say you're in
your your office and and you share your office with
a cat box, and at some point the cat uh
(12:24):
causes quite a stink in there, and you register it first,
and like, geez, I should stop what I'm doing and
clean out that cat box. But you keep working, and
then after a while you don't smell it anymore. But
then maybe you step outside to check the mail, or
you go to the you know, the grocery store to
pick something up, or you know, your partner comes home
and when when you or they enter the room, you
(12:45):
go cheese. What happened in here? Did the cat do
something again? No, you're just re smelling with their Originally
you forgot about it, right, Yeah, the the brain kind
of decides, Look at this point, I assume you know
that she that the cheese is nasty and you're not
going to eat it um or or that yes, there
is animal poop around here. It's it's done, it's part
(13:06):
it's warned you. But the good smell will keep resonating
because the good smell is probably saying, hey, there's something
over here, delicious to eat with some fresh berries or whatnot,
And it'll keep saying, hey, the berries are still here,
why haven't you eaten them yet? This is there's not
a lot of sugar out here. You should get at
these berries while you have a chance. So the beautiful
(13:28):
remains beautiful, The sweet smelling remains sweet smelling. But something
that is disgusting even as you know, discussing as a
foul smell we can grow accustomed to. Okay, so what's
the analogy to the Uncanny Valley here? Well, I think
the the analogy here is that if you have a
disgusted response to the to visual sense information regarding an
(13:51):
individual's appearance, a robots appearance, of a computer animated characters appearance,
there could there could also be this disgust drop off
point as you grow accustomed to it. Oh, and that's
something that we have seen. We talked about a little
bit in the last episode, and that uh, some people
attest that when you spend time around these robots or uh,
(14:11):
well it was mainly for the robots. The robots that
at first seemed creepy, they stop bothering you become accustomed
to them, they're not creepy anymore. I don't so much
know if that's always going to be the case for
creepy looking animated humans. Um, but who knows. Now. Another
thing to keep in mind too, is that the discussed
response is going to depend on a number of different factors,
(14:32):
some of which are going to be tied to hormonal situations. So,
for instance, pregnant women are more sensitive to discussed and
this is linked to their elevated progesterone levels. So uh,
And of course there's gonna be other factors beyond that
for every individual. Yeah. So I think that there's certainly
could be some amount of biological instinctual response going on
(14:54):
in the uncanny value effect to the extent that it exists.
But I think also, based on what we've seen so far,
probably does not account for all of it. Uh. And
I think another thing to consider would be going to
more more complex sort of cognitive psychology, such as cognitive dissonance. Now,
if you were to just ask me what I thought
(15:15):
was the most likely answer before I got into the research,
I would intuitively tend to think that the best answer
for what causes the Uncanny value effect primarily is our
inherent discomfort with category confusions. Uh. This is something that
I think about a lot, in like the creation of
monster mythology and stuff like that. We we don't like
(15:36):
the feeling produced by by things that don't fit into
our normal taxonomys for objects in the world and seem
to violate our tax taxonomic ordering system. And this is
why I think monsters are so often hybrids of existing
things a bull's head on a man's body, things that
defy our intuitive classification rules. They make us uncomfortable and
(16:00):
cause a sense of unease, leading to this uncanny feeling.
So that that's what I would have intuitively said, Yeah,
that makes sense. Is it wolf or is it man?
Or right? But negative affinity resulting from this difficulty and
assigning the entity to a category is it robot or human?
Despite my intuitive favor for this explanation, I think it
(16:23):
looks like experimental evidence for this is not strong, and
in fact, in some ways some of the studies we've
looked at have somewhat invalidated this. For example, in the
last episode at the end, I was talking about that
study by matherin Rifling, and it did not find evidence
of a strong correlation between here. Here's what you notice
here the time it took people to rate the mechano
(16:46):
humanoid qualities of a robot, and that robot pictures likability.
So if you were talking it up to category confusion,
you would probably think the robots that people took the
longest time to figure out how to ray on the
mechano humanoid scale, and those would be the least likable, right,
Because they're the ones causing the most category confusion, right.
(17:09):
But I mean you could also say that, well, you're
not maybe having not a you're not having a visceral
gut reaction to them either, you having to think it
out and try and you know, analyze how you feel
about it. Oh yeah, I'd say that's exactly what we
don't like. I mean, we like to be able to
viscerally categorize things. But but then the uncanny Valley is often,
at least in terms of the way you're going to
find it invoked by the average person. It's often discussed
(17:33):
if it's as if it's a visceral reaction, the sort
of i'll kill it with fire reaction that people might have,
where I hate that saying yes, I'm not a fan
as well, especially when it is applied to outside of
of a fictional connotations. Right. But so they point out
the authors here point out that they did not find this. Uh.
They point out that it's just not a fact that
(17:54):
the things that took longer to look at and decide
where the least likable. Though while this is not statistically important,
just as a point of curiosity, the single face that
took the longest to rate on the mechanical versus humanoid quality.
Was also just about the most disliked face in their
whole collection of faces, but that was just like one outlier. Overall,
(18:15):
this did not present as a general effect. Other studies
have also looked into this and have failed to find
solid evidence for category confusion as the primary driver of
the negative affinity at the bottom of the uncanny valley.
So it looks like my intuitions here I think are wrong.
But there's something that's kind of related as an idea
(18:36):
that's been explored, and that is the idea of perceptual mismatch.
So several authors have advocated the idea that this perceptual
mismatch could be the primary cause of what we don't
like about things that we would intuitively say fall into
the uncanny valley. So one piece of research I want
to mention is a sort of review by uh Cat
(18:58):
Siri that is a review of empirical evidence on different
uncanny valley hypotheses support for perceptual mismatch as one road
to the valley of eerieness that gott to give it
a different name, apparently in Frontiers in Psychology. And so
in this study, the authors review present research and claim
that experimental research attempting to show the Uncanny Valley has
(19:20):
been inconsistent. They don't exactly say that the uncanny valley
doesn't exist, but that it's not as simple as often
believed to be. Something we've been saying for a while now.
It's it's not that any manipulation of the variable of
human likeness leads to Uncanny Valley effects. So, in other words,
that the horizontal axis on the graph is more complicated
(19:41):
than just the question of how realistically human is it.
I've seen this come up enough now that I'm pretty
convinced that that is not necessarily the only or even
the primary factor here. But they still recognize that there
is some kind of effect here. So they claim that
there's evidence against the category confusion basis that we were
just talking about. But they claim that there has been
good evidence in support of the perceptual mismatch hypothesis. And
(20:05):
I want to read what they say. They say quote
taken together, the present review suggested that although not any
kind of human likeness manipulation leads to the Uncanny Valley,
the uncanny valley could be caused by more specific perceptual
mismatch conditions. Such conditions could originate at least from inconsistent
realism levels between individual features, and the examples they give
(20:28):
would be like artificial eyes on a humanlike face, or
the presence of atypical features such as a typically large
eyes on an otherwise humanlike character. So what they're saying
there is not necessarily that you can't tell whether it's
a robot or a human, but that there have been
multiple experiments that seem to show people are unsettled and
(20:51):
made unhappy by things where the features on the face
or the features of the figure as a whole are
inconsistently realistic. Like, we're more okay with a robot that's
consistently realistic at a certain level it maybe say, seventy
percent than something that has eyes at nine d percent
(21:11):
and skin at This reminds me, this is just coming
off the top of my head. So I don't have
the artist's name here, but there's there's an artist who's
worth made the rounds where they took cartoon characters and
they depicted them realistically. So it's Homer Simpson. Yes, Homer
Simpson with like pores on his skin, you know, horrible. Yeah,
(21:32):
So that comes to mind is a kind of a
possible example of this. Yeah. I think that's a good explanation.
So I want to get into my main takeaways from
looking at the Uncanny Valley research so far. Maybe you
can let me know what you think about this. I'd say,
first of all, I think the Uncanny Valley is a
real thing, but it's not as simple as Maury's original
hypothesis would lead you to believe. Uh. First of all,
(21:53):
people definitely do get creeped out by lots of almost
human looking things, but it's not a necessarily just that
the near failed human realism is what makes them unsettling.
There are other things that appear to be making them
unsettling though, that the near humanness plays some kind of role.
(22:13):
And the other big thing is that there appear to
be multiple dimensions to explain the phenomenon. Right, So, synthetic
humanoid images, whether robotic or animated, offer multiple dimensions of
attraction and revulsion. I think it's possible that there are
some biologically triggered effects the appearance of health or disease,
the appearance of life or death. But then I think
(22:34):
there are possibly other things triggered by psychological cognitive dissonance,
probably not category confusion, um, but but some good evidence
for this idea of the perceptual mismatch being the cause.
And then the final thing is that the Uncanny Valley
effect is context dependent. How long have you been exposed
(22:54):
to the image in what's setting? Is it part of
a narrative or some other context in which you're being
asked to suspend your disbelief or otherwise put yourself in
a state of openness. Robert, what do you think about
all this so far? You've got any disagreement? No, I
mean I I feel like my view on it closely
lines up with with with yours here basically that it's
(23:15):
just that there is an effect going on, but it's
far it's far more nuanced than simply oh these are
these are the factors that make something fall into the
Uncanny Valley, not just how realistically human. There's there's other
stuff going on, right, and not just a mere hybridization.
It was instantly thinking about um, the borg and UH
(23:36):
and occasionally the sort of sexy boards that show up
in the Star Trek universe. No, like, there's there's clearly
category confusion going on there, but they're they're not depicted
as particularly uncanny, Like was the board queen? Was she uncanny?
Kind of? I don't know, I mean, even when I'm
not a big fan of the way the board look well,
(24:01):
but the outside of the board. You can also think of,
you know, various sort of hybrid human creatures depicted in
fantasy and fiction that that are created in such a
way to be the alluring, uh, like they managed to
fetishize the inhuman qualities of them the category of confusion. Yeah,
uh yeah. It makes me think that there are certain
(24:23):
qualities of the human appearance that, if altered, are much
more significant in terms of our affects response than others.
So it could be and I'm just making this up.
I don't know if this is true, but that like, uh,
getting the getting the size of the eyes wrong could
quite easily lead to a disgust response and revulsion. But
(24:45):
getting the size of the nose wrong wouldn't. Does that
make sense? Yeah? Yeah? Or just thinking of eyes, like
definitely making the eyes inappropriately large leads a creepy factor,
and this is often employed. I instantly think of the
Van Piate movie. What twenty thirty Days of Night? Is
that the name of it? Oh? Something like that. Yeah,
(25:05):
where they did some sort of digital effect to make ority.
I don't remember the number, some number of days of PPD,
days of night. Um. I didn't actually see if the
trailers were we're certainly interesting. But likewise, if you just
take an individual and have them wear blackout contact lenses
like that is often sometimes that's played up for creepiness,
but a lot of times it's played up for to
(25:26):
be alluring. You'll have uh, male or female characters that
are are otherwise dressed in some alluring fashion, but they
have blacked out eyes and it's who It's kind of
like supernatural, sexy, cool, as opposed to just like, oh
my god, why why are your eyes pits of darkness? Yeah,
but maybe that's just me. Okay, well, I think we
should take a quick break and when we come back,
(25:48):
we will go beyond the Uncanny Valley. All right, we're back, Okay,
So Robert, I can recall discussions going back for years
about whether we're going to make it out of the
Uncanny Valley in the realm of robotics or animation, and
I think from here on I want to focus primarily
(26:08):
on animation, just to just to keep us focused and
I think there are actually two separate questions here, assuming
that the Uncanny Valley is to some extent to coherent idea.
We've already explained all the ways in which it's obviously
way more complicated than the naive popular culture culture understanding
of it. But the two big questions. Number one, can
(26:30):
we make realistic looking humanoid characters that aren't creepy? I
think the answer here is yes. I think it's not
a two dimensional graph. I think you can make things
that aren't quite photo realistic but look realistic that aren't creepy.
Current generation video games have been doing this, and as
I mentioned in the last episode, I think that there
are tricks to doing this. It's apparently in achieving like
(26:52):
the right combination of realistic traits and unrealistic traits that
maybe you would just land on by doing trial and
error in design over time. You would never mistake these
characters for photographs of real humans. But they're also not cartoony.
They've got this feeling of really real ishness, if you
know what I mean. But they've they've they've attained sort
(27:15):
of a generally acceptable plateau of realistic affect. But they're
not skewing into these different danger zones adjacent photo realism,
where the shortcomings become creepy and off putting and we
don't like it. Then there would be another question, and
that's just can we make animated characters that are robustly
(27:35):
indistinguishable from human like, can we get all the way
up the other side of the mountain, up to the
peak of reality? And last week, if you'd ask me,
my personal answer would have been no, not yet. But
I think that's actually not as clear cut as we
would first guess, because you think that perhaps many of
the humans we see on TV are actually digital creations.
(27:58):
Oh I know, for a act that that who's that guy,
that Jimmy Fallon guy? Oh yeah, Jimmy Fallon might be
a computer generated that is not a person. He has
been generated by a computer that's in Palo Alto, California.
It's a super computer. I mean, it's a really good computer.
But yeah, well, I often feel the same way about
(28:21):
Michael Fastbender, And granted it's complicated by the fact that
he has he has a thing for playing androids recently,
but at times you're just like, no, he's just a
little too handsome. There's something in human about this this
man's handsomeness in his charm. So I want to talk
about one thing that is that has given me pause
on this subject. And it's going back to what we
(28:41):
talked about at the beginning of the last episode Rogue one. Yes,
so back to the c G I Grand mof Tarken
when I saw When I first saw Rogue one, I
liked a lot of things about the movie, but I
did not really like the the c g I. Tarken.
The the almost Peter Cushing was very, very good, and
I really mean that, I'm shockingly good, but still not
(29:03):
quite real to me, still kind of distracting because of
how slightly off it was. I would not have mistaken
it for a real person. But the other day I
was talking in the office to Holly fry Who it's
It's Holly, a Star Wars fan. She's so, she's one
of the hosts of Stuff You Missed in History Class,
one of our one of our podcasts in the podcast
(29:25):
family here. Yeah, she has hands down the most Star
Wars knowledgeable person in the office in an office full
of nerds. Yeah, I should point out that Holly has
slash had like a golden ticket to go see Rogue
one anytime she wanted to at the theater. What are
you serious? And serious? This is? Yeah, yeah, legit, where'd
that come from? I don't know, I don't know. I'm
(29:45):
not I'm not at that level of a fandom where
I'm even offered such things. I know. So if if
you want to have a funny experience, go up to
Holly and just like, ask some really obscure random question
about Star Wars where you think it could not be
possible that there's an actual answer to this, like oh,
that stormtrooper on the left, where did he? What planet
(30:06):
was he born on? Holly will have an answer. She'll
be like, oh, that was actually addressed in dialogue in
the Greek dub of this episode of Clone Wars. So
Holly has amazing Star Wars knowledge. She's super fun to
talk to about the Star Wars universe. But anyway, Holly
pointed out that while a lot of people like me
were saying that the c G I. Tarkan was a
few pixels short of escaping the valley, then again, there
(30:29):
were plenty of people, including some older people that she knew,
who couldn't tell that it wasn't a real person they
literally they couldn't tell. Well, I'm worried about my viewing
upcoming viewing of the film because I have I've been
preconditioned to have a certain response to the to the
Tarkan bot here. Oh, I'm sorry that we've had this
(30:49):
discussion before you were able to see the movie for yourself.
It'll be interesting being preconditioned. Will I go into it,
you know, expecting an abomination and like, but the level
of detail will overwhelm me and it won't matter. Or
Am I going to go into there and nothing is
gonna fool me because I'm gonna be looking for the cracks?
You know, I can, I can put myself in a
(31:10):
mindset where I think it's possible. I might not have
known that it was a c g I effect if
I wasn't familiar with what to look for, Like if
I didn't watch a lot of movies that had c
g I effects in them, and if I didn't complain
about c G I a lot. I'm sorry, guilty is charged,
I'm guilty of that. Uh. If I wasn't aware that
(31:32):
Peter Cushing was dead, um, if it wasn't if I
wasn't sort of prepared to see a lot of high
tech c g I by virtue of the fact that
I'm sitting in a theater for a Lucasfilm movie, all
those things. If you took away all the context and
my pre knowledge, I'm very might possibly have fallen for it.
I think I might have just it might have just
gone past me. If I was absorbed in the story,
(31:54):
I might have thought, Yeah, it's kind of strange looking dude,
but it's just a dude. And I think it gets
better or worse, depending on your perspective. So I have
in October news piece from BBC Asia here about a
character called Saia, a computer animated character created by the
(32:14):
Japanese husband and wife graphic design team Terry Yuki and
Yuka Ishikawa. And I mentioned this one in particular because
before we did this episode, I went and I looked
up what are considered a lot of the most realistic
c g I character creations, the most impressive animations. This
one came up, and I think this was the most
impressive to me. It's probably the most photo real computer
(32:37):
generated human I've come across so far. So Siah is
supposed to be a seventeen year old Japanese student and
The creators have been working on her design for a
couple of years now, and as versions of Saia have
been posted on the Internet, people have widely reacted with
comments like, I can't believe that's not a real person.
And I kind of have to agree. I'm looking at
(32:59):
these these pictures of her. There are a couple of
different generations of her design up and the most recent
one just looks like a photograph of a person. Yeah,
I I can't tell that that is not a person.
I have no recourse to critical faculties in my mind
that would say no, here's where you can tell that
(33:20):
that's not a real person. Now. At the same time,
I do have to come back to a comment I
made earlier that that this is also a it's a
it's a it's a pretty face, it's a very standard
face like this is this is leading lady material, Whereas
I think it gets more problematic when you look at
character actor type figures such Peter Cushing, because they have
(33:41):
such a distinctive face. Yes, and Peter Cushing. So this
is one thing that helps, I think, is that it's
a young character who has very smooth features. Uh, Peter
Cushing has a lot of cracks and crags, a lot
of wrinkles, and I think that that may actually be
simply having more texture on your face could make it
much more difficult to make a photo real copy of you.
(34:03):
That that's entirely possible. But to get back to Saia,
So in October of last year, the Artist's debuted the
first animated clip of Siah, which they created using motion
capture technology, and they debuted at a Japanese consumer electronics show.
I watched this footage and I think Morey's distinction about
(34:23):
having different standards for motion and still images does apply
because while with the still image, I can't tell that's
not a real person, with the short animated clip, I can.
I can tell it's not a real person, but it's
still very very impressive, not as absolutely photo real as
the still images. But I don't know. I wonder to
(34:46):
what extent this gap is just um that motion animation
is a bigger technical project, it takes more investment and
money and all that, um, And to what extent the
gap is within the viewers mind essentially, to what ext
and it's caused by the fact that the climb out
of the uncanny valley is steeper if you're moving. Now.
(35:06):
I know we're not talking about robots here, but this,
of course, this brings up the thought that as where
as we attempt to conquer this in the realm of
humanoid robotics, you're gonna inevitably have situations where, oh, it
looks just like a person if it's walking down the street,
but if it climbs stairs. There you go. It's not
necessarily going to add to oh nine the lights in
(35:27):
the air, but maybe there would be something telling like, oh,
it looks like a human most of the time. But
they're gonna be certain movements, certain environmental reactions that are
just not going to hold up to scrutiny, right, So,
I don't know. Looking at these things, looking at Tarkan
in Rogue one, looking at Siah, I think it's clear
that we're getting closer and closer to really bridging the
(35:49):
gap on indistinguishable photo real humanity in computer animation. Whether
you'd call that an uncanny valley or not, obviously this
is a related but maybe different an issue. What we're
definitely drawing near is that peak of reality where there
are synthetically generated images of humans that you can't tell
from the real thing, and since things in the Uncanny
(36:13):
Valley are creepy, I think we usually just assume that
overcoming it is a good thing, right, like designs getting better? Uh.
In't it kind of cool that we can generate these
photo real images without without actually having to photograph someone.
But I'm not so sure that's a good thing. I
think we should maybe think about the implications of this,
like what would happen in the world where human simulations,
(36:35):
especially computer animation, can reliably climb up that second peak.
So we are going to take a quick break and
when we come back, we will go beyond the Uncanny Valley.
All right, we're back. You know, I'm glad you brought
up this idea of you know, is it a bad thing?
Is it a good thing? It does remind me of
(36:56):
a fabulous book that came out a few years back.
I think I just referenced it in our Sex Spots
episode titled The wind Up Girl by Paolo baglapi Um.
It's a near future science fiction tale, just really a
wonderful novel, very fun, but that the wind Up Girl
in question is a essentially a sex Spot character. It
(37:16):
goes you know, that ends up rebelling and you have
sort of a typical narrative with her, but they call
her a wind up girl because she's she's she's very
convincing as a humanoid, except that her skin pours are
too small and she has an intentionally herkey jerky movement
to uh to as she walks around. That they did
(37:38):
so that she could not be mistaken as a person.
So that so that apparently, like all the I think
they were called the new people, it's in some cases
so that the new people could not be mistaken for
the old people. Whoa It reminds me of how they
had to add artificial sounds to electric cars for safety purposes.
Because the cars are too quiet, they can really sneak
(38:00):
up on you from behind, so they had to make
them rumble a little bit. Yeah, I think that's an
app comparison. So here's a question for you, Robert, what
is the gold standard of evidence that somebody did something?
Imagine you're on a jury. I'm the defendant. I've been
accused of offering a cash bribe to a police officer
(38:23):
if she'll let me borrow her gun for five minutes.
She says, I did it. I plead not guilty. What
what's the best evidence to convince you that I really
did that. Well, it's not human memory, because if we've
touched on many times before, human memory is falliable and
and it's a legitimate problem when it comes to to
eyewitness testimony. But when the eyewitness is a video camera,
(38:47):
a digital camera that has a long photographic evidence as
well to certain degrees, like this has been held up
as the gold standard. I mean, assuming the footage is
clear enough that the individual's face is visible, all of that. Uh.
Like even in our science fiction, right, we have so
many examples of like on Star Trek, again, there would
be scenes where Picard would command it we zoom in
(39:09):
and enhance and it was never questioned that there were
any problems with the enhancing of the image. It was
just something that was done. It's like, oh, yeah, the
image is enhanced, and now we see exactly who the
killer is. Huh. So we should look at this this
booming new research field. I shouldn't say booming, I just
mean there are some papers on it, okay, called facial reenactment.
(39:30):
So this employs some of the same techniques that you
would see used in uh in studios. If if people
are doing motion capture for c g I characters In
movies and video games. You have an actor performer who
puts on special gear and a special environment surrounded by
lights and cameras, and the performer acts out motions. These
motions are captured from multiple angles and different lighting conditions,
(39:53):
and then they're translated by a computer into the motions
of a c g I character. You could make a
c g I me that was doing all the same
things I did with my body. But what if instead
of a c g I character you used captured motion
to manipulate existing video or images of a real person,
(40:14):
not a c g I character. This technology is already
in development today, and one example is the research being
done under the heading as I said, of facial reenactment.
There are a couple of papers along with accompanying video
demonstrations by a group of researchers based out of Stanford,
out of the Max Planck Institute for Informatics and the
University of Erlangen Nuremberg, and in their own words quote,
(40:38):
we present a method for the real time transfer of
facial expressions from an actor in a source video to
an actor in a target video, thus enabling the ad
hoc control of the facial expressions of the target actor.
And so if you haven't seen video of this, you
should look it up trying facial reenactment video. If you
(40:59):
have efficient sample video of your target, you can use
a regular camera to project new facial expressions, including mouth
movements which form the shapes of words, onto your your
target in your video. So I could take video of
Robert talking if I had all this technology, if I
could take video of Robert talking, and then I could
(41:21):
film myself saying Halloween five is the best entry in
the Halloween franchise, and then map that onto Robert's face
to make his lips say those words, to make his
face move along with my face as it's being recorded.
And in the demonstrations of this it looks nearly photo real.
They do it with with public figures, making them move
(41:44):
their faces around, move their lips to say things. In
some cases, I think the average observer already would not
be able to tell the difference in this video and
in fact that a similar thing appears to be happening
with voice. This thing you might have read about last
year of this thing Adobe Voco, where they came out
with this announcement that Adobe is working on software where
(42:05):
you can take a twenty minute sample of your audio
to learn from. If you've got a recording of somebody
talking for twenty minutes, I can take it, make a
recording of you saying things you never said in your
own voice. I wanna. I seem to remember that before
his death, Roger Ebert was involved with the project with
some of this technology. The idea of being that, of
(42:25):
course he had lost his his availability to talk due
to illness, but there was so much Roger Ebert audio
out there from all of his years as a as
a as a film critic and a TV personality that
they had this they had, they had everything they needed
to enable him to say anything new. Yeah, and that
(42:45):
that explores the totally non nefarious aspect of this. I mean,
I I don't think people who are pursuing these lines
of research are just trying to create a world where
we can fake video evidence of things. But wouldn't that
be a wonderful thing for somebody who lost their capacity
for speech they had recordings of their voice to be
able to create a text to speech voice box that
(43:05):
could speak with their own voice. That's amazing, that's kind
of beautiful. But there are these other ways of looking
at this, and the authors also that they point out that,
you know, they they in their own defense, they're like, look,
we're not trying to create a world where people can
fake video. We also try to show how you can
detect altered video. So that's another thing they're trying to
(43:28):
explore and make public because you know, they're not the
only people pursuing this research. Obviously, people all over the
place are doing stuff like this, and stuff like this
has been in the use, has been in use in
the movie industry for years. Yeah, I mean it makes
me think that what you would need to go for
is the equivalent of a water mark. I don't know
exactly what that watermark would be and what form it
(43:50):
would take, but it does make me think, well, we're
going to reach a point where any kind of footage
has to have the watermark of authenticity, otherwise doubt will
be asked upon it. Yeah, I'm concerned about the idea
of living in a world where you can make very
convincing looking fake video evidence of things, and not just
(44:13):
because of the specific example of somebody can make a
video of me or somebody I like, you know, saying
or doing something that they didn't do. It's not just
the specifics, it's the general degrading of our trust in
the ability to look at things and know that they're true. Yeah.
I mean, we look at the current news cycle and
there's been a lot of discussion about the reliability of information,
(44:36):
of so called fake news, the idea that you read,
do you reach this point where nobody knows what to
trust anymore? Set of just trusting nothing? And you already
know this because you don't trust any weird looking picture
of somebody you see, right, because you know what can
be done with photoshop. We're already there with still images.
Photoshop is is basically in many contexts become a verb
(45:01):
for for the the distortion of truth. But what what
if we had photoshop to undermine moving video evidence to
the same extent that photoshop has undermined still images. I mean,
this is what a lot of these viral fake news
stories are based on, is a photoshopped image. If you
go to Snopes or something and you look at what
a lot a lot of what they're debunking is, it's
(45:22):
just an image that claims to be real of somebody
doing something, and they have to track down where it
came from Yeah, it's difficult to imagine what that what
would slash will be like when we reach that point
to where there's where where video even digital footage is
no longer the gold standard it was. I mean, I
think it's actually a very important project to maintain a
(45:47):
version of the Uncanny Valley, to to help people find
a way to all to separate real video evidence from
fake video evidence of things, to understand that there are
things you can look for that separate real moving imagery
from falsified or synthetic moving imagery. Now, one way you
can approach this is to do what the authors of
(46:08):
this research I was talking about do is they say, look,
here are things that are are signs that video has
been manipulated. And that's one thing. And maybe there will
be a lot of experts on this in the future,
Like it could be a whole field of people who
are just there to have expertise in authenticating purportedly real
video of you doing things or not. Uh. Then on
(46:32):
the other hand, we could hope that there is in
fact an adaptive response in our discernment in general. And
this is where I want to go to the concept
of an of the uncanny Wall. So paper in the
International Journal of Arts and Technology, authored by Tinwell Grimshawn
Williams offers this interesting counter hypothesis to the Uncanny Valley.
(46:56):
And I want to emphasize again we've sort of shifted
back and forth betwe in the Uncanny Valley itself in
terms of what causes negative affinity, and then over slightly
just to the issue of nearing photo realism or not.
So keep in mind the difference in those subjects. But
they proposed this idea of the uncanny wall. To put
(47:17):
it succinctly, they proposed that quote, increasing technological sophistication in
the creation of realism for humanlike virtual characters is matched
by increasing technological discernment on the part of the viewer.
In other words, as humanoid characters become more real, our
(47:38):
standards for what looks realistic go up. And I do
think they're just anecdotally personally that I think there's some
support for this, and I kind of hope this is
true so we can avoid this world where all video
evidence is in question. Because I immediately think, right now,
I've got I've got antennae for photoshopped images, unlike had
(48:00):
ten years ago. I think stuff that would look obviously
photoshop to me today would have fooled me ten years ago.
I think I've simply adapted. And another thing is it
makes me flashback to the early days of c g
I in movies and like the thirty two bit video
game era, or think about like PlayStation one games and
(48:21):
back then. I remember looking at games for the original
PlayStation and thinking, wow, that looks so real. And you
try and play him now and it's painful. Yeah, you
can't blocky polygons. People's faces have all these sharp corners.
It's it's hilarious. Uh, there was some kind of geometrical
nightmare world everything was taking place in where there's just
(48:45):
lots of sharp angles. But at the time it looked
so real to me. And another fun trick is go
back and read movie reviews for movies with bad c
g I from the nineties. Professional movie reviewers at the time,
we're often praising the X. One example is like the
Mortal Kombat movie, the original Mortal Kombat movie. Remember this.
(49:06):
You can find reviews at the time where people are like, well,
the stories then an immature but dazzling special effects. Now
even mentioning those special effects conjures a kind of delirious hilarity.
You just start laughing when you think about the c
G I in Mortal Kombat. But that being said, the
GOA puppet was above reproach, really good. It did have
(49:27):
kind of nasty beady eyes, kind of crypt keeper ask. Yeah,
oh yeah, it was like a very It was like
a bloated, buffed up crypt kicker. But yeah, it's so
ugly it provokes uncontrollable laughter. But at the time people
were like, dazzling, looks amazing. Uh. So it makes me
think that I hope that there is something to this,
(49:51):
this idea that these authors have that as things continue
to chase photo realism, as synthetic imagery of humans gets
closer and closer to the real thing, we just get
more and more attuned to the minute problems with them
with them and never really get fully fooled. Well. I
have two thoughts here, One on the whole water mark thing.
(50:14):
Maybe it would be some sort of a Bitcoin type
of authentication system that would be in place. Uh. The
other is maybe you have to go beyond the real
maybe safe for a head of state to appear in
a video and it'd be authentic. They have to appear
as an as a computer generated avatar so advanced than
it is that it is beyond the ability of any
(50:37):
like non state production or company to create like something
I'm I'm something that at this point in what would
be their past, I cannot even conceive of, like a
I don't know, like a five dimensional unfolding c g
I god being, because how he's going to fake that?
You can fake a person, but good luck faking the
(50:57):
fifth dimension hole of avatar or a fish new Okay,
like like the algorithms for faking a person with a
lot of photo and and audio visual cues to sample from.
If you've got a lot of footage out there, you
could simulate that person, But you couldn't simulate this brand
new creation that is it requires you know, supercomputers to generate.
(51:20):
And yeah, and it was probably built from the bottom
up with completely alien physiology and movements. Just a thought.
That's kind of a crazy idea, but I like it. Well,
that's what I'm here for with the crazy ideas. I
don't know, have you got anything else, Robert, any anything
else you can think of to save us from the
future of uh synthetic human imagery? Oh? You know, I
could sit around here all day and talk about c
(51:41):
g I, monsters and UH and Uncanny Valley and films
and video games and whatnot. But you don't have to
what to save that for another time. Maybe save some
of it for trailer talk, which will If you're listening
to this on a Thursday, hopefully you can tune in
tomorrow around eleven am on our Facebook page. I like
our Facebook page. While you're at it, follow us there,
but tune into a little discussion of trailers that are
(52:04):
associated with the Uncanny Valley. Oh yeah, And in the meantime,
heading over to stuff to Blow your Mind dot com.
That's the mothership. That's we will find all of our
blog posts, our podcast, our videos, and links out to
our various social media accounts such that Facebook page or
the Twitter page, the Instagram account, you name it. And
if you want to get in touch with us directly,
as always, you can email us at blow the Mind
(52:27):
at how stuff works dot com for more on this
and thousands of other topics. Is that how stuff Works
dot com has atance by the bar