Do We Need Humanoid Robots?

Josh Clark joins the podcast to talk about humanoid robots. Why are we so concerned about building them and are they a good idea?

Learn more about your ad-choices at https://www.iheartpodcastnetwork.com

See omnystudio.com/listener for privacy information.

Episode Transcript

Speaker 1 (00:04):
Get in text with technology with text Stuff from stuff
dot com either everyone, and welcome to Tech Stuff. I'm
Jonathan Strickland and I'm joined by a special guest host today,
someone who is on the most popular podcast in the
How Stuff Works suite of shows, Josh Clark Glow, co

(00:27):
host of Stuff You Should Know. Josh, thank you so
much for joining us. Thank you for having me here.
And Josh has been very kind. He's had me on
Stuff you Should Know. Way back in the day when
you guys were doing one on the Necronomicon. Yeah, we
were starting to get knee deep into it and realizing
that we were going to get a lot, a lot
of listener mail, so we came and got you. You

(00:49):
know what was funny was that no warning, you came
to me asking about pronunciation I think two times, and
then on the third time you're like, just come on
inside the studio and you came and said down and
talked about the Necronomicon like a promicron. I know this
by the way, Yeah, Nickronomicon, Necronomicon. Okay, Yeah, and um

(01:10):
I am still to this day impressed. And that was
five years ago. Yeah, yeah, and then they haven't had
me on since But anyway, we had Josh on today. No,
you guys are super busy even out. Yeah, I know,
I was the one dragging my feet. Yeah, And I
asked Josh what he would like to cover because with
the fact that I've got all these guests coming in
to sit down with me. Um. You know, some people

(01:32):
like to come up with their own suggestions. Some people
preferred if I pick a topic and then they research it.
I asked Josh what he would like to talk about,
and you were really interested in the idea of humanoid robots. Well,
you have this awesome spreadsheet of um of listeners suggestions
and it might as well have been a neon when
it was going down the sheet. I'm like humanoid robots,

(01:53):
of course, and this is a great topic. In order
to really get into it, I was going to define
a few terms, even though a lot of these are
ones that I think most of us just kind of
understand just from the fact that this is in our
culture now. It's not just not just a reality as
far as technology goes, but it plays a large part
in fiction. In fact, that's where the term robot comes

(02:15):
from is from fiction. Uh, it was from Carrel Chopeck,
a Czechoslovakian playwright. I did, in fact listen to it
a couple of times. Correl Chope Yeah, because it's his
last name, is spelled c A p e k. And
it includes uh symbols that are not in the English alphabet,

(02:36):
like squiggly lines and little UFOs and things he wrote.
Are you are also known as Rossum's Universal Robots, and
the word robot comes from the check word robot to,
which means forced labor. Yeah, so a robot is a
an entity, a synthetic construct that is forced to do work.

(02:58):
Then we have humanoid, which just means resembling a human being.
That's a term that is relatively young. It started showing
up around the turn of the twentieth century, and uh
it started I think the first few times it was
ever mentioned was around nineteen twelve, and it was mostly
used then to describe fossils, saying these are humanoid fossils,

(03:21):
like yeah. And then we have android, which is we're
probably not gonna be using that word very often, but
android is a robot that's in the form of a human.
So all androids or robots, but not all robots are androids.
And you know, I ran into I looked up android
as well. Yeah, and apparently that's from like the early
eighteenth century. It's a little odd that it actually predates robots. Yeah,

(03:45):
but uh, when we look at myths and legends, there's
so many stories that involve a human like entity that's
not actually a person that you can see where it
gets translated in there. This gets a little confusing too,
because Star Wars they called all their abouts droids, but
there they aren't androids are two? T two is not
an android because he's not um or it's not human shaped.

(04:09):
You could even argue that C three Po is not
a true android because some people say to be an
android you have to appear, at least on casual glance
to be human. He's way too shiny, way too shiny.
Data from Star Trek next Generation might be uh android,
but he's an android who's had, you know, a long
time in the man cave. He hasn't seen much sun, right, Yeah,

(04:31):
because he's definitely got a weird complexion. Thing goes the
kid in a I would be an android David yes,
which turns out to be a popular name for droids
because there was well, there's David in AI and there
was also David and Prometheus. Oh you know, I never
saw Prometheus. I don't know if your listeners are going
to agree with me or not, because I could see

(04:52):
um getting shouted down. But I thought Prometheus was a
great movie. Even even upon second viewing, I thought it
was good. You know, I know that the from the
artistic level, a lot of people really loved it. And
then there were some people who said, how can you
get lost if you have a three dimensional map with
you at all time? But you know, plot versus artists,

(05:12):
I don't know. Then you've got replicants from Blade Runner.
These are more like cyborgs because they have some sort
of organic material attached to them. They're not completely you know,
synthetic material. So Terminator is another example. They have a
fleshy over skin on top of their metallic bodies. But

(05:32):
was that real skin that he had there was synthetic skin,
because that would make a difference. That's a good question.
And I uh, you know, I know that they referred
to them as cyborgs at least a few times in
the movies, which would suggest that it's actual skin. Maybe
it's lab grown human skin. So you know, there's some
fuzzy lines around these definitions. I would think that cyborgs

(05:56):
would be the hardest of all of the the humanoid
robots to make, because the flesh were just rot Like
you have your normal looking humanoid robot, your cyborg, and
it's ear would just fall off. Yeah, and it's not
as easy as you might think to wire together the
wet wear that's in our heads with hardware that runs

(06:18):
on circuits. We will often think of computers working in
a way that's similar to our brains, but in fact
the two work in very different ways. Well, it seems
like running into this, though Strickland um. The the more
we got into humanoid robotics, the more we started to
understand just how complex we are. Yeah, that's one of

(06:38):
the things that I think is a benefit of study
of humanoid robotics. The idea of pursuing the goal of
creating a humanoid robot is not just that we learn
more about all the different areas and robotics, and they're
a lot, and we'll talk about some of them, but
we also learn more about ourselves. We're trying to figure out, okay,
if we're going to make something that is able to

(07:00):
performed tasks the way human does. Then we really got
to take a close look at humans. That's that's the
first place to start. So what makes a humanoid robot?
And generally speaking, we're talking about a robot that has
basic features, usually at minimum a torso, arms and legs,
and is walking up right. Uh. It may have a

(07:22):
head or it might not. Early humanoid robots didn't, or
at least their sensory uh, instruments were all located within
the top part of the torso there wasn't like a
separate head. Did you see a picture of Minerva at
the Smithsonian. No? I did not. It's a robot to
her guide, But um, she came up in the humanoid

(07:44):
robot research and I think she's stretching it a little bit. Yeah,
she looks a bit like a washing machine with a
couple of UM cameras on top. So just those alone,
I guess makes her eligible for the humanoid robot realm.
But to stretching that, that seems like that's a bit
of a you know, if it has an appendage, that

(08:04):
doesn't necessarily make a humanoid. I mean you could look
at the Mars Curiosity rover, which has several appendages. But
I don't think anyone would ever describe it as humanoid, right, So, uh,
ideally a humanoid robot would be able to interact with
humans within a human environment. Because here's the thing about we,
we people, we have defined our environments to a large extent,

(08:26):
especially in developed nations, where the stuff that's around us
we have shaped so that it works within our capabilities,
right with our with our human environment. Thus far, if
you look at technology though, on the whole, we've pretty
much been forced to adapt to it. So for example,
like a keyboard, we don't normally, naturally you know, um

(08:49):
express ideas through our fingers on a little a little board, right,
we don't normally do that, So we had to adapt
to the technology and learn to type and get good
at it. With humanoid robots, it's basically going the exact opposite.
It's saying we already have an environment, We already are um,
you know, good at all this other stuff. If we're
gonna make humanoid robots, one of the great benefits is

(09:10):
they can adapt to us. Right, Yeah, we don't have
to uh create a unitask or robot that's really good
at one thing. Um that may or may not be
something that humans can do easily. We can make a
robot that's good lots of things. Uh. I also think
usually when I think of humanoid robots, when I think

(09:30):
of robots in general, I normally think that they are
at least semi autonomous. That's that's one of the things
I usually think of. It doesn't necessarily have to be.
You could have a tel operated robot, but I almost
think of that closer to the realm of like a
remote controlled car or a puppet even um So I

(09:51):
often one of the definitions I use is that it's
an autonomous or semi autonomous machine in human form, mechanical
and electronic, and it can thus do the sort of
things humans do, but do it in a totally synthetic way.
And you have to be careful when you say autonomous

(10:11):
or semi autonomous, because the state of the art right
now appears to be that robots display autonomy because they
just kind of wander off in places they're not supposed to.
But it's not because they want to. It's because their
their program just ran a foul of program. Right. There's
no determination there on the part of the robot. It's
not exploring its environment on its own accord. It's someone

(10:35):
made a mistake in the code somewhere and where the
robot was supposed to take a left hand turn at
this one you know, predetermined spot, and instead continued forward
or something exactly. Um, So I wanted to talk a
little bit about the history of humanoid robots and if
you wanna look way way way back, I mean, we're
talking about the first people to really kind of attempt

(10:58):
to build a humanoid machine that could mimic are the
movements at least of a person. You gotta go all
the way back to Greece between the years ten and
seventy Common era. That's when Hero of Alexandria started to
create various machines. He's also the person who made the
first working steam engine style tool, which is pretty impressive. Yeah.

(11:23):
He he had come up with a lot of very
clever designs. Whether they were built or not depends upon uh,
certain accounts and and if they're true. But the stuff
he designed is completely build a bowl. So he didn't
come up with any ideas where it was so you know, outlandish,
that was impossible. He wasn't just like drawing in the
margins of his diary or something. No, Yeah, he came

(11:44):
up with specific plans that people today have recreated, they
built their own versions. Yeah. So he created a lot
of designs for automata, although these are things that were
controlled by pulleys and ropes and cogged wheels, and uh
needed some form of outside influence to make them work,
so they're not all fully self contained like the show

(12:07):
biz pizza rockefire explosion band. Yeah exactly. Yeah, yeah, some
one of those audio animatronic figures that that looks very robotic,
but you realize it's really just one tiny piece of
a giant system. Uh. In fourteen, we get up to
Leonardo da Vinci. He designed an automaton in the form
of a mechanical night which uh again supposedly he built.

(12:31):
There's no actual record of an existing one from history,
but they have created ones based on the design since then,
and it could do things like move its arms and
raise its visor. Who's doing this? Who's doing this? Crazy
engineers who are also uh, very excited about history and
very wealthy too, I would imagine. Yeah, so in this case,

(12:53):
you're talking about Mark Rossheim, who recreated this particular machine
and it it's a night. It's a night in German
medieval armor. And it can sit down, it can stand up,
it can move its arms, it can raise a visor,
it can work its jaw. Um, I imagine. So I

(13:15):
tried to find video of Rossheim's version working, and I
couldn't find it. He did, however, make another of da
Vinci's inventions, which was a self driving cart that used Yeah,
you you wound a spring and it had cam stops
that would allow it to steer a predetermined path. You
would actually program the cart by putting cam stops in

(13:37):
particular locations along the cam and that would tell it
when to turn left or when to turn right. So
it couldn't it couldn't navigate through uh an obstacle course
unless you had already previously seen the obstacle course and
you could figure out when it needed to turn ahead
of time. So you're essentially programming the device. Um. There
are lots of examples in the renaissance of automata and

(14:00):
semi automata, things that are really more like puppets. You've
heard about the mechanical turk, the chess playing robot uh
So it looked like it was a robot that could
play chess and was really really good at playing chess
and it turned out eventually to be a hoax. It
was actually it was actually a puppet, and there was
an actual chess master hidden in a cabinet beneath the

(14:23):
mechanical turk who sat kind of Indian style with a
chessboard in front of him and could move the pieces
to where it needed to be. But it was all
being guided by an actual chess master who was hidden
under When that that was late Renaissance early Enlightenment, pretty impressive, Yeah,
it was. It was neat that people were thinking about
these sort of things. Uh. By ninety six we get

(14:45):
the first humanoid robot to appear on film, uh Metropolis,
the character of Maria and at nine at the World's Fair,
Westinghouse Electric Corporation showed off a robot called Electro. Now
have you ever seen this? So he kind of looks
like the ten Man from the Wizard of Oz film.
I have seen him. Yes, he smokes. I have seen

(15:07):
one of the things he can do. He had a
little bellows in his head that allowed him to puff smoke.
He could also kind of speak. He had a seventy
eight revolution per minute UH record player essentially inside of him.
I suppose if the if the needles skipped, it would
at least sound like it. Um. He would repeat himself over.

(15:28):
He called the audience Twitts. That was Grandpa. Yeah, he's apparently.
The main reason he was retired from being shown off
at exhibitions was because it was very dated kind of lingo.
But he was used at the ninety nine World's Fair,
which was It's funny because I'll be talking about that

(15:50):
again in another episode very shortly. The World's Fair would
have been an amazing thing to visit. Um, I'm more
of I can understand that. But if you're if you're
going to get to the point where you're looking at
full scale anthropomorphic robots, you gotta get up to about nine.
That's when the Waybot one from the Waseda University came out.

(16:12):
That was the first full scale anthropomorphic robot developed in
the world which had limb control, a vision system, so
I had an optical system that could recognize its environment
and objects and measure distance, and it also had a
conversation system. Uh. It was actually a collection of a
bunch of very complex machinery. Like its hands had been

(16:34):
previously developed independently of the robots, so had its legs.
So it's like all these people coming up with these
various pieces saying, all right, let's connect all this together
and see what happens. So that was a huge, huge
leap forward. Yeah. You know, if you'll notice, we went
basically from um, HOAXI chess playing turks to you know,

(16:55):
a robot that could converse and interact with its environment,
and then um, it seems like we we kind of
went off course for a little bit and now we're
coming full circle back to that where, like you said,
a lot of different disciplines are contributing these different pieces
to what will eventually be all of the best practices
from each little sub discipline put together in you know,

(17:17):
the true humanoid robot. Well yeah, I mean, if we're
talking about a humanoid robot that's capable of interacting with people,
uh as if they were you know, their own person,
even though maybe an odd person, not like the kind
of person you would typically run into. There's it's a
multidisciplinary approach. I mean, artificial intelligence by itself is multidisciplinary

(17:39):
because you have sensing, you have so all the perception,
there's all these different just that's multi disciplinary. Then you've
got the processing, the cognition, things like planning, navigation. There
are so many things that come together to make a
humanoid robot a a possibility. And that's just the the

(17:59):
men toll side, right. Then you have all the physical side,
the how do you make it walk? How do you
make it keep its balance? So lots of stuff to
consider there. Waybot two came out in nineteen eighty that
was a specialist robot. It could play a keyboard trying
to keyboard. He could read sheet music and play music. Uh.
It was because it was a specialist. It was not

(18:21):
able to do the general functions that it's predecessor could do.
And that's why of the issues in robotics today is
that it's very challenging to build a general purpose robot.
It's much easier to take a specific task that you
need to have done and to sign a robot to
do that, because I mean, we already have those rombas

(18:41):
and rovers. I mean, you know, there's there's also all
the robots and manufacturing, all the welding robots things like that. Uh.
Nine Pacific Northwest National Laboratory built a robot called Manny
that was the first full scale android body and it
had forty degrees of freedom, but no AI or autonomy.

(19:02):
It was completely teleoperated. And um, it took me a
couple of times to figure out what degrees of freedom meant. Yeah, um,
I thought it meant like the it could move its
arm forty two degrees basically. But a degree of freedom
is say, like it can move its wrist. That's a
degree of freedom. It can turn its head left and right.
That's a degree of freedom right right. And if you

(19:23):
look at the human hand, the human hand has about
thirty degrees of freedom, meaning that you look at the
way each finger and your thumb can move. You look
at the way you can clench a fist, you can
twist your hand with your wrist. Um, those are all
different degrees of freedom. And uh. In fact, one of
the cool things about robots is that as we get

(19:44):
better and better at designing them, we can create robots
that have far more degrees of freedom than the human
body does. So I like the idea of a humanoid
robot in the future that has sty degree motion with
its wrist and then just having it changed light bulbs
is spin and not have to do the little twisty
turning motion. How many robots does it take to change

(20:04):
the light bulbs. Just just the one, just that one,
just one billion dollar room. Yeah. Yeah, I'm not saying
it's a fishing system. I'm just saying I'm a supremely
lazy human being with tall ceilings. Uh In nine, Honda
introduced the P two Robot, which was a self contained
robotic humanoid. It could walk and climb stairs. The P

(20:25):
three followed in nine and in two thousand two, Honda
introduced My good buddy as Emo. As the first article
I wrote for How Stuff Works as works an episode
on it. I have not done a full episode on
as Amo. I even was offered the opportunity to meet
as Amo when I first wrote the article, but it

(20:46):
would have meant having to travel to Disneyland to do it,
and at the time How Stuff Works was not prepared
to do such a I went to Disneyland by myself,
uh well with my wife, and we went and saw
the Asimo production. And at the end of it, I
talked to one of the disney cast members and I said, yeah,
I wrote the article about how Asthma works for How

(21:07):
Stuff Works, and she said, hang on a minute. And
I got to meet Asthma and it was a man
in a suit. Right. It was actually it was actually
a collection of cats that the duct taped together and
then plastic. No, it was a working robot that's pretty new.
Blown away, I was very much blown away. It was
cool seeing it up close. I mean, it looks like
a little tiny astronaut, right, because he's got like the

(21:28):
face plate. Especially. I love the people call it he
people give And I do this all the time too
with robots. I'll assign a gender even though technically many
of them are specifically genderless. As amo is is supposed
to be genderless, but I often refer to asthmosa he
as well. Well, you know why. It's the shoulders, I
would guess. Yeah, they go straight across and and and
far out. That's very masculine no matter what. Yeah, and

(21:51):
you need that, I would guess you need shoulders in
a humanoid robot with flexible arms. Well, and also, I'm
sure every single element Asimo is built with the balance
in mind, because Asimo is the first robot that can run. Yes,
I've seen him run. It's gawky. Yeah, it kind of
looks like someone who really needs to get to the
bathroom as a little bit of a hoppy kind of run.

(22:14):
But the the definition of run here is that there
are moments where both feet are off the ground. So
walking you always have one foot in contact with the ground,
and running both feet at some point are out of contact.
And that's a huge deal for robotics, right. I mean,
you have a machine that completely separates itself from contact

(22:34):
on the ground. It has no propulsion to keep it upright,
you know, it doesn't have like propellers or jets or
anything like that, so you have to design it so
it can it can propel itself off the ground and
then catch itself when it comes back down without falling over.
And that is a non trivial challenge, No, it's an
enormous challenge that robot is just UM really kind of

(22:55):
started to tackle lately. Um. One of the ways that
they've overcome it is with rounded feet, which are very
helpful in keeping balanced and allowing it to run. UM.
But there's drawbacks to it as well, Like the robot
can't start itself, it also can't stop, so it can't
stop moving, which is not something you want. Like, there's

(23:15):
still some challenges there ahead of the robotists who are
learning to teach a robot to walk, and even the
ones that have taught robots to walk, um, they typically
can just walk over flat surfaces with no obstacles. When
they encounter stairs, there in trouble. But then you have
robots that know how to go upstairs, but they can't
walk on a flat surface. Eventually, all this information, all

(23:38):
this knowledge will be brought together and you'll have a
robot that can walk, no problem. Right. In fact, this
kind of transitions nicely into those challenges that face designers
of humanoid robots and and locomotion is the probably one
of the top ones, at least from the physical engineering side.
For example, you know, as Amo can can go up

(23:58):
and downstairs, but that is a little deceptive because as
Amo has to be programmed to go up or down
the staircase and know exactly how many stairs are involved.
It's not so much it's not a case of Asimo
detecting a staircase and then uh and then navigating through
oh up or down. It it's the fact that all

(24:19):
right now we're initiating your stair climbing programs. Yeah, exactly.
It's kind of like smoke and mirrors robotics. Basically, it's
at but that's you know, those are the little no
no pun intended, Those are the little steps you have
to take in order to get to the destination. What
do you mean, no pun intended. I don't buy that
at all. I started saying it without thinking about it,

(24:41):
and then I mean, but then I did follow through
with it, so I guess there was some intention there
at the end. But uh yeah. They're also not very
good at going across any kind of uneven terrain, right, So,
humanoid robots in particular find it very difficult to maintain
balance over anything that's not either a flat surface or,

(25:02):
in the case of robots, that can go up or
downstairs and stairs. So if you're talking about like a
sidewalk that is not completely even, that would be enough
to give a robot trouble because it's going to try
and put its foot down to where it would believe
the ground to be, and if the ground is not
exactly level, then that's yeah, because they can't really catch

(25:23):
their balance very well. There are robots that can, but
they are four legged. Yeah, you've seen you've seen the
Big Dog video. No, I saw the Army. It's very similar.
Big Dog is essentially a robo mule type development. It
is a four legged robot that is able to maintain
its balance even when pushed. And the famous video shows

(25:47):
the robot dog, the big dog kind of jogging, and
then a guy just casually lifts his leg up and
kicks the robot dog like he puts essentially puts a
bomb of his foot against the side of it and
pushes really hard, and you see the big dog stumble.
It actually stumbles and then catches itself and then rights

(26:09):
itself and continues on. And almost everyone has an emotional
reaction to this, like, how dare that evil man kick
that poor, defenseless robot. The robot can't feel anything, but
that robot is um gasoline powered, so oh yeah, that's right,
that's one of the big keeping it inside. Yeah, you

(26:30):
wouldn't want to have one of these indoors. No, you
don't bring indoor no. And uh and the the pistons
that allow it to do this are quite loud. You know.
It's not not a subtle system at all. So a
lot of work has to go into creating better systems
for robots to maintain their balance in order for the
locomotion problem to really be solved. And again, I mean,

(26:53):
like anybody who's seen short circuit knows that you can
build a robot like Johnny five with arms in the
head and a tore so and then like traction um
shreds uh, and it can go anywhere over terrain, it
can go up steps. Probably. The thing is is, again
you have to remember when it comes to humanoid robots,
you're trying to make the robot that can adapt to

(27:15):
the human world. So if you had somebody like Johnny
five as your house butler or something, you you couldn't
You couldn't have an island in your kitchen, and who
doesn't love an island in their kitchen. Johnny five couldn't
maneuver around it because he's too wide. Yeah, you wouldn't
have You wouldn't be able to have any any space
that would be narrower than the robot's body. Exactly. That's
not what you want with humanoid robots. It wouldn't work

(27:39):
well in my house. I've got a I've got a
a flat style house where there's three floors. Yeah, it's
like flat like European flat. Uh, not flat as in
there's only one level. They're actually three of them, four
if you count the rooftop Denck, So that counts. It
makes it. You know, any robot that would not be
able to navig it stairs easily would definitely have an issue,

(28:02):
which is the main reason why I don't have a roomba,
because I don't want to hear the sound of a
room bag going falling down a flight of stairs. Um.
But at any rate, Uh, that's a great point. Moving
on from locomotion, there's also dexterous manipulation. Yeah, I think
we should. I think that point bears repeating. What we
just talked about was locomotion. Yeah, and this is you

(28:23):
and I a couple of non robot experts talking about
the problem with locomotion. That's just one of myriad challenges
facing humanoid robotics designers. Yeah, yeah, exactly. It's It's one
that's easy to point to because it's something that we all,
you know, end up at least observing or participating in

(28:45):
all the time. Can we take it for granted? But
then when you think, okay, well, how do I make
a machine that does that? You start to realize this
is you know, even if I have a leg that
has lots of different degrees of freedom and points of articulation.
I still have to design the upper part of the
robots so that it does not unbalanced the lower part,
and if it does unbalance, that it's able to catch itself.

(29:06):
You know, some people just describe walking as falling and
catching yourself over and over again. Yeah, yeah, they're walking
right now. No, I I described walking as something that
other people do. I like to keep my walking to
a minimum. I thought you walked alot, Actually I do.
I just joke about being lazy. I think moving forward,

(29:28):
falling down and catching your balance every time it's lurching. Yeah,
that's well. As an Adam's family fan, I'm okay with that. Yeah,
but uh, Dexter's manipulation would be the ability to pick
up and manipulate objects. Now, we're really good at that,
we humans, You know. We can we can feel an
object and decide at that point how to handle it,

(29:50):
even if we've never encountered that kind of object before.
So if I encounter something I've never seen before, and
i've I've ascertained it's safe for me to touch it,
I can touch it. I can feel like, I can
get a feel for how heavy it is, I can
get a feel for how delicate it might be, and
then I can adjust on the fly so that I
can handle it appropriately. Well, I'm not gonna hurt myself.
I'm not gonna hurt the object. Robots are not so

(30:13):
good at that, Yeah, exactly, even if they don't mean
to right, Yeah, if the robots, If the robots grip
is too strong, it can break the object. If it's
too weak, the object slips from its grip and it falls. Uh.
And it may not be able to distinguish between different types.
So getting those tactile sensors where a robot can tell

(30:34):
how tightly it's gripping something and how much pressure a
particular object can take before you've reached the failure point
is a big deal. Now, this is also a big
deal for just making robots safe for humans to be around.
It is a big deal. You know that the first
fatality by robot occurred um at the business end of
a robotic arm in a flat rock, Michigan in nine nine.

(30:57):
A man named Robert Williams, who's working on a Ford
line YEA, his robot arm was moving a little slow
for his taste in getting supplies down, so he climbed
up to where the supplies were the robot arms suddenly
sped up and hit him in the head and killed
him instantly. Wow. Yeah, I've I've had a chance to
see some of these industrial robots. Uh. And I would

(31:18):
say up close, but you can't because because of instances
like that, industrial robots usually have lots of of safety
barriers around them because it's not safe to be near
those robots when they're in operation. They they can't react exactly.
So you leave it up to the humans to stay
away from the robots because the road we haven't gotten

(31:39):
to the point where the robots no, did not crush
you or hit you in the head. Right, Yeah, I
got the robots fall. I'm looking forward to that day
when they figure out not to crush me. Yes, it's
been pretty lousy days so far. Uh. Yeah. When I
when I toured the Georgia Tech Robotics Lab, they talked
specifically about this. This is a real challenge having robots
recognize and react in a way that's going to be

(32:02):
safe around humans. And uh, but dexterous manipulation is only
that's only a part of dextrous manipulation. Obviously, the rest
of it is again that object recognition and handling so
that you're not destroying whatever it is you're trying to
pick up. Um. Another big challenge in designing robots in general,
not just humanoid robots, is just the the perception, the

(32:22):
sensory perception of the robot. Yeah, so you know, whether
it's optical systems like actual cameras in the place of eyes,
or infrared so that you can see even in low
light situations, radar, light ar. I mean, there's tons of
different ways of sensing. Yep, there's uh, there's you know

(32:43):
sensing obviously, it's not just site. Then you have to
have the sound. That's a really tricky one actually, because
for us humans, we we can kind of zero in
on what's important, right, So if we're if you and
I were at a party, which you know, someone made
a mistake and invited me, we could have a conversation
and be able to carry that conversation on even within

(33:05):
the context of a big, bustling party because we can
focus on what the other person is saying. It's called
latent inhibition. Yeah, so when you don't have that, that schizophrenia, Yeah,
you you can't separate the the signal from the noise
and everything either. Becomes noise or everything becomes signal. Um.
So for a robot that might, for example, require verbal commands,

(33:26):
that's really tricky. What if you have the television on
and someone's saying something on TV and uh, you are
trying to get your robot to do something, and it's
not quite sure what to do because it's hearing these
different commands and isn't sure who to obey you the
person pitching the bacon bowl right right exactly. Let's say
that you are, you know, trying to get some help

(33:47):
in the kitchen, but it just keeps hearing uh C
s I Miami and say kill Billy. And then next
thing you know, you're like, you're just desperately trying to
kill the robot. Please don't kill Billy. Um. Yeah, well
that's a silly example. It's a real problem. Uh. And
then there's the tactile the responses, the tactile sensors, like
the making sure you don't crush something that's delicate that

(34:09):
falls into perception. Smell can also fall into perception. You
might want to have humanoid robots that work in areas
where the humanoid robot can alert humans to the presence
of things that might be toxic. You know, this isn't
necessarily just the robot Butler we're talking about. This could
be robots that work in areas that might be dangerous

(34:31):
for humans, and that would be an important element too.
It's not a robot, but NASA already has a sensor
that senses things like ammonia or smoke. It can actually
sense smoke artificially, smell smoke before the fire has actually
started ignited. Interesting because you know it's such a dangerous proposition, right, yes,

(34:51):
clearly for for anything NASA related. But they can also
sense ammonia because you know a lot of the refrigeration
systems run on ammonia and you can't have a pneumonia
leak on the space basis, right right, And I mean
the same thing is true for I mean I've heard
of of robots that are used in in mining operations,

(35:12):
which you know, if you come upon a pocket of
natural gas that can be a real danger that sort
of stuff. Then you've got the the back end of
the sensory perception. That's where you have the actual interpretation
of the data. Where that's the big one. That's huge,
because not only do you need to have a robot
that can have that has binocular vision, so it has

(35:32):
a depth of field. Right. Um, it also has to
know what it's doing, what what the information means, and
how to apply it to adapted changes, right right. So,
so if I were to show you, Josh a series
of pictures of various types of dogs, you would very
quickly pick up on the the things that mean mean

(35:54):
dog like. You would understand the concept of dog pretty quickly.
Robots and various other computers machines they have a lot
harder problem with this. If whatever they're looking at doesn't
exactly match the parameters of the example, it's very difficult
for a machine to extrapolate and say, oh, this other
thing I'm looking at relates to this thing I know,

(36:17):
even though the two examples don't aren't identical. So the
same thing could be true for any object. Let's let's
just use a coffee mug. And let's say that you
use a a plain white coffee mug of average size
as h the example for the robot, and then the
robot encounters a larger blue coffee mug and the handles

(36:39):
turned the other way. The robot might be completely befuddled
by this. So this is a real problem in artificial
intelligence is object identification, so that a robot knows what
it's looking at and also understands the context that that
object fills within the environment. So it's not just that, oh,
that's a mug, it's oh, that's a mug. A mug

(37:02):
is a container. I can put things into that mug.
Here are the things that can go in the mug.
You are the things that absolutely should not not go
in the mug, like Billy. Those are the kind of
things that yeah, Billy, well, you know we're gonna need
another Billy. Um, we'll make a robot, which in which
case you can just turn those suckers out right mass
production of Billy. But yeah, artificial intelligence an enormous problem.

(37:23):
And that, of course is not just with robotics. That's
that's a field unto itself, and robotics is just one
branch that relies upon artificial intelligence. And it's from what
I came across, it looked like just out of the gate,
I guess that was said a university they tried to
build a robot that was just like high functioning, yeah,

(37:43):
and they realized, like, we have no idea what we're doing. Yeah,
that that Waybot one was able to converse at the
level or was able to have a a a cognitive
function equivalent to a one and a half year old person,
which I have to when you're talking right out of
the gate. Very impressive, Very impressive because we're not that

(38:05):
much further along now. But what they found from making
way about one was, Okay, this is way more difficult
than we thought. You can't just program every kind of
coffee cup in the world, and even if you could,
then you also have to program every kind of table
and every kind of light. And we need to come

(38:26):
at this in a different way. And so they realized,
number one, humans are extraordinarily more complex than we thought before.
And then number two, humans make a pretty good model
for a humanoid robot in the realm of things like
perception and UM information systems and UH learning. So they

(38:48):
went to these different these different disciplines like neurobiology, neurology,
UM psychology, and they said, what can we learn from
you guys about how humans do this that we can
apply to robots. And since they started taking those steps,
it seems like, uh, humanoid robotics has gotten it's it's

(39:09):
footing a little more. Yeah, and we're seeing so many
developments in other areas of a I that are really promising.
I always bring up IBMS Watson because it's natural language
recognition was phenomenal, the ability for it to parse clues
in Jeopardy and come up with the appropriate answer, knowing
that Jeopardy those clues are not always straightforward. Uh. And

(39:31):
it again illustrates the complexity that we humans navigate without
much trouble because this is the world we've created. But
then we realize if we make a machine that's mostly
when you get down to it based on yes or no,
a one or a zero, true or false, and you're
trying to build complex behaviors off of something that is
incredibly simple. When you boil it down to its basic element,

(39:55):
that's where you're like, oh, this is this is gonna
require a lot of work. I mean, IBMS Watson was
an enormous machine with with thousands of microprocessors just so
it could be able to play Jeopardy. That's a very
specific function too. So creating a robot that is able
to navigate and interact with a human environment and be

(40:16):
able to interact with humans in a way that makes
sense is a big challenge. Also, just the way that
a robot would socialize with humans is a huge challenge.
How do how do you make a robot that is
able to respond to commands and cues in an appropriate way? Uh?
An appropriate way is the key there, because there are
humans are pretty complex and we can be very subtle

(40:38):
in many ways. Yeah, we speak unplainly, we use sarcasm,
we we uh yeah, we use a lot of gestures
rather than just words. Yep. There's a lot that goes
into human communication that, if you are a human is
pretty much natural, especially I mean if you're a human
within that particular culture and you're familiar with that culture,

(40:59):
because anyone who has traveled extensively knows there are cultures
where things that would be commonplace at home are very
different in the place where you happen to be, right then,
and it may be that something that is completely innocent
at home is an uh, offensive gesture in the place
where you are. Now, we'll imagine a robot that is

(41:20):
not programmed to handle these kind of subtle uh communication
methods and exact Yeah, he doesn't know, he's just he's
just doing as he was programmed. But even even beyond that,
something that you brought up in our research when we
were planning this was the Uncanny Valley. It's a big one. Yeah.

(41:42):
I've never read that paper before, and I'm glad I did.
It's really interesting, right, Yeah, So the Uncanny Valley. For
those who are not familiar with the term, uh, it's
it describes when we start to approach artificial humans that
look almost but not quite like real humans. And and
by look, I don't necessarily just mean the physical appearance.
I also mean their behaviors, their movements. So if you

(42:06):
are if you were to look out and see a
figure that from the from a distance looked like it
was a human figure, and you start walking towards it,
just thinking this is another person, and then they start
moving in a very herky jerky motion, very mechanical motion,
then you're likely going to have a negative emotional response.
Um often, revulsion is one of the words used very frequentness. Yeah,

(42:30):
I mean I remember the C G I movies that
would that were almost to the point of photo realism,
where they look like people for the most part, but
there's not something is just not quite right with the eye.
There's a good example. And polar express polar expresses they

(42:51):
do very well because of the Uncanny Valley. That's what
they blame it on. Yeah, and the same thing applies
to robots. So in fact, I saw a robot that
was really disturbing to me. It was really as an
um art exhibit, an installation, and um it was a
robot of Confucius in a cell filled with monkeys, and

(43:12):
uh they're live monkeys, real monkeys, and the robot would
just thrash around wildly. It was It was the stuff
of nightmares. I'll show it to you after the show. Yeah.
So these are all big challenges and some of them
are going to be harder for us than others. And
maybe that the engineering challenges of locomotion are solved well

(43:34):
before we ever get a real grip on all the
artificial intelligence problems. Or it could be the other way around.
Uh we but it is multidisciplinary. It's a big, big issue.
So there are some people who argue for humanoid robots,
there are people who argue against humanoid robots. Um. I think.

(43:55):
I think research and development with humanoid robots is important
because by having the goal of creating a humanoid robot,
you drive the research and development process. You have a
specific goal in mind, and in order to achieve that goal,
you know what sort of problems you have to solve.
And even if we never enter a future where humanoid
robots are a common thing, even if they are mostly

(44:20):
used as something in an exhibition or uh for pr
or whatever, even if that's the only use for them,
we're going to benefit from the research and development of
making that possible in ways we can't anticipate. Well. Yeah,
and then the more we get into humanoid robotics, the
more we understand humans, which is pretty much the only

(44:43):
argument I've seen that stands up in favor of doing
humanoid robots. Yeah, because it's it's expensive, and it's hard.
I mean, it's it's really a difficult problem, and it's
and and to build a humanoid robot something that is
capable of being a general purpose robot, it's you know,
it's hard to anticipate all the things you're going to

(45:03):
need to be able to do. If you're talking general
purpose and adaptable, that's really tough. I mean, we didn't
even talk about the adaptability problem very much. We talked
a little bit about a robot capable of learning from
other people, which I find fascinating. Um you know, that's
one way. Just watching humans and then mimicking humans, that's
one way of robot learning. There's also, um the way

(45:27):
where it's controlled through virtual reality by human and it
just kind of logs the motions the humans making it do.
Like there's a NASA has a robot not that learns
like that. So I think I think there's a lot
of benefit to investigating artificial intelligence. Like you want your
nest at home to learn so you don't have to

(45:49):
keep adjusting the thermostat that counts. That's the that's machine learning.
To me. The big argument against having humanoid robots, and
the quag meyer that seems to begin, is sociability. That
seems to be the whole reason anybody wants a humanoid robot,
because you can make you know, you have a ruma vacuums,

(46:10):
you can make a driverless car um As Olivia Solan
wrote and wired a couple of years back. Um, you know,
why why do you have to make a robot butler
to park the car? Just make a car that parks itself.
And it seems like that's where we're going right now. Um.
So when you add this extra layer of humanoid, you

(46:30):
add all of this additional problems and troubles and redundancies,
like like, for example, if you're gonna make a humanoid
robot that throws a ball, this this humanoid robot to
appear real needs to have a little bit of follow through.
But as far as the robot, the machine is concerned,
it can throw the ball and just stop right where

(46:53):
the releases. It doesn't need to go anymore. But it's
gonna look weird and robotic if you want to get
past that uncanny valley, which is an other problem, um,
the thing has to have followed through. That's totally unnecessary.
You can make a robot that can throw a ball
and the goal is to throw the ball. You don't
have to add the follow through, but you do when

(47:13):
you're making it a sociable humanoid robots. So it seems
like that's the path that will lead everyone is straight
that I don't get. Yeah. I I like the idea
of designing robots for specific tasks because you can really
focus on getting the task done. So there I see
this as two separate branches. I see the branch of

(47:33):
developing the humanoid robot as pushing forward a lot of
different areas of thought. That could be applied in multiple disciplines,
so that will benefit from that. I see the development
of robots as unitaskers as being important to actually handle
the jobs that are the three D s. That's dirty, difficult,
and dangerous, all right, So those are the jobs that

(47:58):
maybe they revolve a lot of repetition, which can cause
injury over time, or it can lead to mistakes because
you've done the same task so many times that you
start to kind of zone out. Robots will never zone out. Um.
If it's a dirty job where it's something that's undesirable
by people, robots don't care. They'll do that. Or if
it's dangerous, if it's bomb disposal, or if it's something

(48:21):
like the Mars curiosity rover. These are dangerous jobs that
you wouldn't necessarily want to put a human into if
you had the alternative. So all of these things, those
that's where robots really makes sense to me UM to
to go to the places that are difficult for us
to go to. Maybe that's you know, deep sea exploration,
space exploration, that kind of thing, or to do jobs

(48:45):
that might be dangerous, having a first respond to robot
to survey a scene, to make sure that a structure
is remaining uh solid while maybe there was a fire
and it has to make sure that the it's it's
not going to collapse in on rescue mission that kind
of stuff. Um, but do you need those things to
be to come out and be able to tell a

(49:06):
joke or something? And most of them don't need to
be any sort of humanoid form factor either, which greatly
simplifies the actual development of the robot and thus cuts
down on the cost, so you can you can achieve
the task you're trying to achieve for less money than
if you are trying to build this this general purpose machine.

(49:26):
So then we come to this ultimate question, why what's
the purpose of humanoid robot? Well, I think the I
think the purposes too fold. One is again to have
that specific goal in mind that allows you to define
where your in point is. I believe that when you

(49:49):
have that defined in goal, it makes it easier for
you to build on the things you need to achieve it,
as opposed to having an open goal where it's just
I want to approve improve a I that's so open
that it's hard to get direction from it. But if
you think I need to have an artificial intelligence that
will allow a robot to Uh, here's a great example.

(50:09):
Let's say that the challenge is to have a robot
leave a room, go down a flight of stairs, leave
a building, get into a vehicle, drive the vehicle to
a different location, get out of the vehicle, go into
another building, break through a wall, and put out a fire.
That's a real, actual robotics challenge challenge. Yeah. Dr Henrik

(50:32):
Kristensen told me about this. Is I mean, it really
is a challenge. It's not just a real challenge. It's
a real challenge. It's like a DARPA challenge. So it's
a uh he was telling me about this, and you
start to think about all the things that have to
fall in line for you to be able to achieve
the skull. That is a valuable thing. But I think
the other thing is the social aspect. I think that
there are people who would benefit from a robot that

(50:56):
is able to give some form of social comfort. Let's
say for the elderly who need to have some form
of interaction. Um, you know, that could actually be a
really valuable tool. And in fact, there's a lot of
work that's going into robotics to help people like the
elderly who may have real emotional and psychological problems, Um,

(51:20):
due to loneliness. Do you think that robots are the
answer to that? I think that robots can help. I
don't know. I would never go so far as to
say answer. But couldn't you also make the argument that
if you created robots that displaced human jobs and also
simultaneously said, hey, this, this nursing home sector is about

(51:46):
to explode because we've got a bunch of baby boomers
and ways the society of now decided that our elderly
need human interaction more than we've more than we've carried
it out before. So let's create this whole other industry.
Or let's expand this industry of elderly caretakers and fill

(52:06):
those jobs with people who have been displaced by roots.
Wouldn't that be better? That might be? Or you could again,
looking at the way a lot of roboticists framed this,
they say, all right, well, it is a reality that
robots are taking over actual jobs, but the hope is

(52:27):
that it also ends up creating new jobs that are
better paying jobs, less dangerous jobs friendly, more old books friendly.
But yeah, like um, the idea being that that it
frees up people and encourages the pursuit of jobs and
engineering in computer science. Now we live in the real

(52:50):
world and we understand that it's a lot more complex
than telling someone who's been working on a manufacturing line, Hey,
I'm sorry your job's gone because there's a robot here.
But guess what, we have an opening and engineering. So
if you just go and pursue a four year degree
and then some post graduate work, you'll be right back
to work. That's that's obviously not uh something that's going

(53:10):
to be easy, especially in the short term. But the
long term hope is that more and more of these
jobs that are are dangerous for people, are less desirable
for people, will be taken up by robots, and then
the will there will be the creation of better jobs
that are higher up on the food chain. And I
think that makes sense to me. It's just the once

(53:32):
you enter the sociability, yeah, because without sociability, there's no
reason to create a humanoid robot. Everything else can look
like a robot. Yeah. Um, so it's it's when you
enter sociability that you lose me. Not only does it
can it look like a robot, but we can still
socialize with it, even if it doesn't look like a
human people there. You know, there's the story that Rumba

(53:54):
owners named the Yeah yeah, so you we we end
up having these kind of emotional attachments and investments in
things that don't look not only do they not like human,
they don't look like any other animal that we would
interact with on a like owner or and pet or whatever.
I mean, they they're they're just a robot. So I

(54:16):
think at the end of the day, Josh, I think
we're on the same page. We think humanoid robots are
an interesting idea, but not necessarily the end goal. There's
there's not a whole lot of of incentive to go
after it for its own purposes. We can see the
benefits of going after it in the sense of the
developments that are made in that pursuit help us in
other ways. But I don't need a human robot, Butler, Yeah,

(54:42):
I just know I don't. Well, I'm curious to hear
what my listeners have to say about this. You can
get in touch with me my email addresses, text stuff
at hell stuff works dot com, or drop me a
line on Twitter or Facebook or tumbler. Tech Stuff HSW
is the handle for all three of those. Josh, thank

(55:02):
you so much for being on the show. Thank you
for having me. I'm let do this begin now. If
you have not heard stuff, you should know, you definitely
need to go and check that out because it is
a phenomenal podcast. Thank you. Josh is one of the
two hosts, the other being Chuck Bryant, who I hope
to have on tech stuff in the near future, so
keep your ears up for that. And Josh, I hope

(55:23):
I can grab you back in here for another episode
in each time. Man awesome. Well, thank you again, and folks,
we'll talk to you again really soon for more on
this and thousands of other topics. Because it has to
works dot Com