Through the Wormhole s04e07 Episode Script
Are Robots The Future of Human Evolution?
Freeman: We are in the midst of a revolution so insidious, we can't even see it.
Robots live and work beside us.
And now we're designing them to think for themselves.
We're giving them the power to learn to move on their own.
Will these new life-forms evolve to be smarter and more capable than us? Or will we choose to merge with the machines? Are robots the future of human evolution? Space, time, life itself.
The secrets of the cosmos lie through the wormhole.
We humans like to think of ourselves as the pinnacle of evolution.
We are the smartest, most adaptable form of life on earth.
We have reshaped the world to suit our needs.
But just as homo sapiens replaced homo erectus, it's inevitable something will replace us.
What if we're building our own successors? Just as we learn to move, think, and feel for ourselves, we're now giving robots those same powers.
Where will this lead? Is this what humanity will become? When I was a teenager, I built a bicycle from spare parts.
My bicycle was so well-balanced, i could jog alongside it without holding onto it.
Nobody was that impressed, but it made me think.
Would we one day have machines that truly could move on their own? Would they even need us anymore? Daniel wolpert of the university of Cambridge believes that if robots are to be the future of human evolution, they're going to have to learn to move as well as we do Because movement is the supreme achievement of our powerful intellect.
The most fundamental question I think we can ever ask is, why and when have animals ever evolved a brain? Now, when I ask my students this question, they'll tell me, we have ones to think or to perceive the world, and that's completely wrong.
We have a brain for one reason and one reason only, and that's to produce adaptable and complex movement, because movement is the only way we have of affecting the world around us.
Freeman: All of our brains' intellectual capacity grew from one primal motivation -- to learn how to move better.
It was our ability to walk on two legs, to speak and emote with complex facial movements, and to manipulate our dexterous limbs that put humans on top of the food chain.
There can be no value to perception or emotions or thinking without the ability to act.
All those other features, like memory, cognition, love, fear play into movement, which is the final output of the brain.
Freeman: No machine could handle the huge variety of complex movements we perform every day.
Just imagine a robot trying to play one of england's most famous pastimes.
So, although that shot looks simple and it felt effortless to me, the complexity of what's going on in my brain is really quite remarkable.
I have to follow the ball as the bowler bowls it and predict where it's going to bounce and how it's gonna rise from the ground.
I then have to make a decision as to what type of shot I'm going to make.
And finally, I have to contract my 600 muscles in a particular sequence to execute the shot.
Now, each of those components has real mathematical complexity, which is currently beyond the ability of any robotic device.
Freeman: One of the greatest challenges in getting robots to move like we do is teaching them to deal with uncertainty -- something our brains do intrinsically.
A ball will never come at you the same way twice.
You must instantly adjust your swing each time.
The question is How does the human brain deal with all this uncertainty? Daniel thinks it uses a theory of probability estimation called bayesian inference to figure it out.
Wolpert: So, a critical thing the batsman now has to do is decide where this ball is going to bounce, so as they can prepare the correct shot, and for that, they need bayesian inference.
What bayesian inference is all about is deciding, optimally, the bounce location of the ball from two different sources of information.
Freeman: One source of information is obvious.
You look at the ball.
Wolpert: So, you can use vision of the trajectory of the ball as it comes in to try and estimate where it's going to bounce.
But vision is not perfect, in that we have variability in our visual processors, so at least where distribution's shown in red here as the probable bounce locations.
But bayes' rule says there's another source of information.
It's the prior knowledge about possible bounce locations.
If you're a good batter, then you can effectively look at the bowler and maybe know by his particular bowling style or small cues -- and that's shown by the blue shading -- which is a different area.
So, bayesian inference is a way of combining this red distribution with the blue distribution, and you do that by multiplying the numbers together in each to generate this yellow distribution, which is termed the belief.
And using that information, the batsman can now prepare his shot.
Okay, I should probably get out of the way.
Freeman: The batsman's brain, like all of ours, is doing this math automatically.
We are a species that's honed for movement prediction.
It's what has made us the planet's best hunters and toolmakers.
We already have robots that are faster and more accurate than we are.
But we have to program their every move.
For robots to walk down the evolutionary road we've already traveled, they're going to have to learn to move on their own.
What happens then? Will they evolve complex brains like ours? Robot builders Josh bongard of the university of Vermont and hod lipson of Cornell university are trying to answer that question.
Increasingly, we see that interaction with the world, with the physical world, is important for intelligence.
You can't just build a brain in a jar.
Freeman: Hod and Josh's goal is to build a machine that's smart enough to learn how to move around all by itself.
They've created a menagerie of strange robotic forms along the way.
But their work starts with the computer program designed to evolve robot bodies.
It simulates various body plans and then tries various strategies to get them to move.
Okay, so, let's walk our way through -- no pun intended -- an actual evolutionary simulation.
So, in this case, we've told the computer that we want a robot that has two legs, but we want the computer to figure out how to get the robot to orchestrate the movement of the robot's legs.
And here, we see something a little bit surprising, that evolution hasn't discovered the solution that we use.
Sometimes when we run this evolutionary process, it produces something familiar like walking, and in other cases, it produces something that's not familiar, something we wouldn't have come up with on our own.
Freeman: It's survival of the fittest or perhaps the least awkward.
Just as mother nature selects generations based on their ability to survive, so does the simulation.
The computer deletes the robots that aren't doing a very good job, and the computer then takes the robots that are doing a slightly better job and makes modified copies of them and repeats this process over and over again.
And after a while, the computer starts to discover robots that, in this case, are able to walk from the left side of the screen to the right side of the screen.
Freeman: This is evolution on steroids.
What took mother nature millions of years takes the computer just a few hours.
Overnight, the computer tests thousands of generations, and eventually it produces a robot that meets the goal.
When the simulation makes something that looks particularly interesting, hod and Josh take that body plan and build it.
Now they can test whether the strategies for moving learned in simulation work as well in the real world.
So, this robot is called the quadratot, and it's basically a robot that learns how to walk using evolutionary robotic techniques.
And so, what we can see here is a particular example of how this robot learns.
This is one of the earliest gaits that it did, and we can see that it's not moving very far or very fast.
It's kind of like a child during its very early behaviors of crawling.
It's trying out different things.
Some things work better.
Some things work less well.
And it's taking the experiences and learning from them and gradually improving its gait.
Freeman: There are many robots that can move well while executing a specific predesigned task.
But hod and Josh's robots must start to learn by themselves from scratch in an unknown environment.
It can sense its own progress, and like a baby learning to crawl, it becomes more aware of its body with every step and every tumble.
Hod and Josh believe this self-awareness gradually builds into a basic form of consciousness.
Often, we've raised the question, is something conscious, or is it not? But it's really not a black-and-white thing.
It's more about to what degree an entity's able to conceive of itself, to simulate itself, to think about itself, to be self-aware.
As robots learn to move in more complex ways, it's possible they will develop levels of consciousness equal to ours and maybe beyond.
But according to one scientist, for a robot to become truly conscious it must develop feelings.
What is consciousness? The answer depends on who you talk to.
A doctor's definition would be different from a priest's.
But we all agree that our high-level consciousness is what separates us from other organisms and, of course, from robots.
What would it take for robots to become conscious? Can they get there on logic alone? Or must they also learn to feel? Professor pentti haikonen from the university of Illinois believes machines will only become conscious when they can experience emotions.
It's a belief he has held since he was very young when he first contemplated what it meant to be conscious.
When I was 4 or 5 years old, I was standing in our kitchen, and suddenly I was struck by the mystery of existence how and why I was me and why I was not my sister or my brother.
How did I get inside myself? Freeman: As he got older, pentti realized that what made him feel conscious of being inside his head were his emotional reactions to the people and objects in the world around him.
The neural processes that are behind our consciousness take place inside our brain, but we don't see things that way.
For instance, when you cut your finger, the pain is in the finger or so it appears, but, actually, the pain is in here.
To feel is to be conscious.
Freeman: Our brain's raw experience of the world around us is just a series of electrical impulses generated by our senses.
However, we translate these impulses into mental images by making emotional associations with them.
A sound is pleasing.
A view is peaceful.
Consciousness, according to pentti, is just a rich mosaic of emotionally laden mental images.
He believes that to have a truly conscious machine, you must give it the power to associate sensory data with emotions.
And in this robot, he has begun that process.
This is the first robot that utilizes associative neural networks.
It is the same kind of learning that we humans use.
When we see and hear something, we make a connection between those things, and later on when we see or hear the other thing, the other thing comes to our mind.
Freeman: The xcr-1 experiences the world directly through its senses like we do.
On board are the basics of touch, sight, and sound.
Pentti has begun the process of giving it emotional associations to specific sensory data, like the color green.
Pentti places a green object in front of the robot, which it recognizes.
Green.
Then he gives green a bad association -- a smack on the backside.
The associative learning is similar to little children.
Hurt.
And you say that this is not good or this is good, or you may also smack the little child.
Hurt.
I don't recommend that.
Green bad.
Freeman: The robot's mental image of the green object is now associated with the emotion bad.
And from now on, it will avoid the green bottle.
But it's not all pain for the xcr-1.
Just like we teach the robot to associate pain with the green object, we can teach the robot to associate, also, pleasure with objects, in this case with the blue object, like this.
Blue.
Freeman: To give blue a good association, pentti gently strokes the top of the robot.
Blue good.
This simple experiment demonstrates that this robot has mental images of objects and mental content.
Freeman: It's still early in its development, but the xcr-1 has learned the basics of emotional reaction from fear Green bad.
Green bad.
to desire.
Blue good.
Now's my time for love lonely moments seem to As a more advanced version of the xcr-1 fills its memory with mental images Dentist bad.
it will start to be able to react to new situations on its own and eventually experience the world much like any emotionally-driven being.
It is my great dream to build robot that is one day able to ask, "How did I get inside myself?" Freeman: Once robots reach this point, what's to stop them from moving on and becoming conscious of things we're not? This man thinks robots will become the future of humanity because they'll have something we lack.
Their brains will have the capacity for genius long after the last human ever says "Eureka.
" for Archimedes, Eureka happened in the bathtub.
Einstein was riding a streetcar when relativity dawned on him.
These brilliant minds had a flash of inspiration and drove all of humanity forward.
But the scientific questions of today, probing shoals of subatomic particles and our vast genetic code, have become so complex that they take teams of thousands of researchers to solve.
Is the age of the single scientific genius over? Not if machines have their way.
Data scientist Michael schmidt sees the world filled with intricate beauty -- the flowering of a rose, the veins branching on a leaf, the flight of a Bumblebee.
But below the surface of nature's wonders, Michael also sees a treasure trove of uncharted mathematical complexity.
Schmidt: Well, I love coming out here.
Nature is beautiful.
There are equations hidden in every plant and every bee and the ecosystems involved in this garden.
And part of science is figuring out what causes those things to happen.
Freeman: Science is our effort to make sense of nature, and this quest has given us some very famous discoveries.
In Newton's time, he was able to figure out a very important rule in physics, which is the law of gravity.
It predicts how this apple falls and the forces that act upon this apple.
Today in science, we're interested in similar problems but not just about how the apple falls but the massive complexity that follows from this very simple dynamic to the world around us.
For example, when I drop this apple, the apple stirs up dust.
This dust could hit a flower, and a bee may be less likely to pollinate that flower.
And the entire ecosystem in this garden could change dramatically from that single event.
Freeman: Scientists understand the basic forces of nature, but making precise predictions about what will happen in the real world with its staggering complexity is overwhelming to the human mind.
So, one of the reasons why it's extremely difficult for humans to understand and figure out the equations and the laws of nature is literally the number of variables that are at play.
There could be thousands of variables that influence a system that we're only just beginning to tease apart.
In fact, there are so many of these equations, we'll never be able to finish analyzing them if we do it by hand.
Freeman: In 2006, Michael began developing intelligent computer software that could observe complex natural systems and derive meaning from what seems like chaos.
So, what I have here is a double pendulum.
If you look at it, it consists of two arms.
One arm swings along the top axis, and the second arm is attached to the bottom of the first arm, and it's two pendulums that are hooked together, one pendulum at the end of the other.
Now, the pendulum is a great example of complexity because it exhibits some of the most complex behavior that we're aware of, which is called chaos.
So, when you collect data from this sort of device, it looks almost completely random, and there doesn't appear to be any sort of pattern.
But because this is a physical deterministic system, a pattern does exist.
Freeman: Finding a pattern amidst the chaos of the double pendulum has stumped scientists for decades.
But then Michael had a flash of inspiration.
Why not grow new ideas the same way nature created us, using evolution? He called his program Eureka.
Eureka starts with a primordial soup of random equations and checks how closely they fit the behavior of the double pendulum.
If they don't fit, the computer kills them.
If they do, the computer moves them into the next generation, where they mutate and try to get an even closer fit.
Eventually, a winning equation emerges, one that Archimedes would be proud of.
Eureka! Schmidt: And I'm running our algorithm now.
On the left pane are the lists of the equations that Eureka has thought up for this double pendulum.
Walking up, we can see we increase the complexity, and we're also increasing the agreement with the data.
And eventually, as you go up, you start to get an extremely close agreement with the data, and eventually you snap on to a truth where you get a large improvement in the accuracy.
And we can actually look in here and see exactly what pops out.
For example here, you might notice we have a 9.
8, and if you remember from physics courses, that is the coefficient of gravity on earth.
What's very important is the difference between the two angles of the double pendulum.
This pops out.
Essentially, we've used this software and the data we've collected to model chaos, and we've teased out the solution directly from the data.
Freeman: Eureka has not only discovered a single equation to explain how a double pendulum moves.
It has found meaning in what looks like chaos -- something no human or machine has done before.
Schmidt: So, we could collect an entirely new data set, run this process again, and even though the data is completely different -- we could have different observations -- we can still identify the underlying truth, the underlying pattern, which is this equation.
Freeman: To Michael, the future of scientific exploration isn't inside our heads.
It's inside machines.
Whether they're looking at patterns of data from genetics, particle physics, or meteorology, programs like Eureka can evolve inspiration on demand, finding basic truths about nature that no human ever could.
We're gonna reach a point where we decide what we want to discover and we let the machines figure this out for us.
Eureka can find these relationships without human bias and without human limitations.
We created robots to serve us.
As the machines learn their own ways to move, feel, and think, they will eventually grow out of that role.
What if they start working together? Could they build their own society, one made by the robots for the robots? There's no species on earth more successful than us.
We owe that success to the powerful computer inside our heads.
But it takes more than one brain to conquer a planet.
Homo sapiens thrive because we have learned to make those computers work together as a society.
What will happen when robots put their heads together? Roboticist by day and gourmet chef by night, Professor Dennis Hong of Virginia tech is a specialist in building cooperative robots.
But he also sees cooperation outside the lab.
So, we don't really think about it, but everything in our daily lives involves cooperation.
For example, cooking oftentimes is thought of as a solo act, but if you think about it, a lot of people are involved and a lot of careful coordination is required to make it happen.
Oh, thank you, charli.
Take this tomato as an example.
This tomato most likely started its life as a seed, where a group of breeders need to choose the right sequence of genes for a plump, juicy, tasty tomato.
The seeds needed to be planted, grown, harvested, then the tomatoes needed to get to the market.
Freeman: Food production is a complex web of coordination.
But as good as it is, human cooperation has its limits.
Hong: Oops.
Freeman: Every day, like most of us, Dennis has to contend with the prime example of human cooperation gone wrong -- traffic.
The problem is, us being human, we all need to, want to get to our destination as quick as possible, thus we have traffic jams.
Freeman: If it wasn't for traffic lights, which are, in reality, very simple robots, it will be almost impossible to get anywhere.
These traffic lights, they talk to each other.
They communicate with other traffic lights at other intersections.
And they have cameras, so they actually see the traffic patterns and make decisions for us, for humans.
Oh, there you go.
Thank you, traffic light.
Freeman: Traffic is a nuisance.
But other failures of human cooperation are much more serious And often deadly.
[ Machine-gun fire .]
Dennis believes a society of robots can be much better collaborators than we are.
So, in collaboration with Daniel Lee at the university of Pennsylvania, he designed a group of robots to compete in the robocup, an international robotic soccer championship.
Robocup is an autonomous robot soccer competition, which means that you have a team of robots, you press "Start," And then nobody touches anything.
And the robots need to look around, see where the ball is, need to coordinate and actually play a game of soccer.
Freeman: Dennis' soccer robots, called Darwin-op, are fully autonomous.
They use complex sensors and software to navigate the playing field.
And they have a serious competitive edge over their human counterparts.
Teammates can read each other's minds.
So, if you look at human soccer players, obviously they're great at what they do.
They communicate sometimes by shouting, sometimes by a subtle gesture, but, again, it's not really accurate, and they cannot share all the information together at the same time in real time, but robots can do that.
Freeman: Each robot knows the exact location and destination of the other robots at all times.
They can adjust their strategy and even their roles as necessary.
Hong: Depending on where the ball is, where the opponents are, they didactically switch their roles.
So the goalie becomes a striker, a striker becomes a goalie or defense.
Freeman: They may not be as agile as pelé or bend it like Beckham, but they are able to dribble past their opponents, pass the ball, score a goal And even celebrate.
Dennis believes robocup is just the beginning of robot societies.
Dennis imagines a connected community of thinking machines that would be far more sophisticated than human communities.
He calls it cloud robotics.
Hong: Cloud robotics is a shared network of intelligence.
It's similar to what we call common sense in humans.
So, just like those smaller robots that play soccer for robocup, they share a common data, team data, to achieve the goal, in this case, winning the soccer game.
For cloud robotics, robots from the furthest corners in the world, they can all connect to the cloud and share information and intelligence to do their job.
Freeman: Humans spend a lifetime mastering knowledge, but future robots could learn it all in microseconds.
They could create their own hyper-connected network using the same spirit of cooperation that built human society without the selfishness and greed that hold us back.
Robots operate by a very well-defined set of rules.
The human impetus to break them is just not there.
Freeman: Robots already know how to talk to one another.
But now a scientist in Berlin has taken robotic communication a step further.
His machines are speaking a language he doesn't understand.
Motakay tokima.
Did you understand what I just said? Of course you didn't because I wasn't speaking any known human language.
But it wasn't nonsense.
It was a robot language.
We humans took tens of thousands of years to develop our complex means of communication.
Now robots are following our lead, and they're doing it at light speed.
Someday soon, robots may decide to exclude us from their conversation.
Robot: Tokima.
Lucabo.
Miyoto.
Motakay.
Tokima, kymamu.
Tokima.
Simeta.
Tokima.
Motakay.
Steels: Without language, our species would never be where it is today.
It's the most magnificent thing that has ever been created by humanity.
If you look at ourselves, then it's pretty clear that without language, we would not be able to do the kinds of things that we're doing.
Freeman: Luc steels, a Professor of artificial intelligence, sees language as the key to developing true robot intelligence.
Steels: What I'm trying to understand is, how can we synthesize this process so that we can start up a kind of evolution in a robot or in a population of robots that will also lead to the growth of a rich communication system like we have.
Freeman: Machines already communicate with each other, but these are based on predetermined, human-coded languages.
Luc wants to know how future robot societies might communicate given the chance to make a language on their own.
Luc gives his robots the basic ingredients of language, like potential sounds to use, and possible ways to join them together.
But what the robots say is up to them.
We put in learning mechanisms, we put in invention mechanisms, mechanisms so that they can coordinate their language.
They can kind of negotiate how they're gonna speak, but we don't put in our language or our concepts.
Freeman: It's not enough for the robots to know how to speak.
They need to have something to speak about.
Luc's next step is to teach the robots how to recognize their own bodies.
Steels: In order to learn language, you actually have to learn about your own body and the movements of your own body.
So, what you see here is an internal model that the robot is building of itself.
This robot is trying to learn here, is the relationship between all these different sensory channels and its own motor commands.
Freeman: As a robot watches itself move in the mirror, it forms a 3-d model of its limbs and joints.
It stores this information in sense memory and is now ready to talk to another robot about movement.
Tokima.
So, now, this robot is talking.
He's asking an action.
This robot is doing -- you know, stretching the arm.
No, this is not what was requested, and he's showing again what the right action is.
Freeman: She's unsuccessful in her first attempt, but eventually the robot learns that "Tokima" Means "Raise two arms.
" after repeating this process with different words, they try again.
Homakey.
Another request.
He's doing the action.
Yes, this is the right kind of action.
So, in other words, this robot has learned the word from the other one and vice versa.
They now have a way to talk about actions.
Freeman: The robots' language is already so well-developed, they can teach it to Luc.
Let's see what, you know, what he asks me to do.
Motakay.
Okay, motakay.
No, this is not right, so he's showing it to me.
Okay, I'm learning this gesture now.
Motakay.
Motakay.
Motakay is this.
Okay, I got it right.
So, now I'm going to use the gesture with him.
Motakay.
Okay, yes, you're doing the right thing.
Thank you.
As the robots repeat this process, they generate words and actions that have real meanings for one another.
And so the robots' vocabulary grows.
Every new word they create is one more that we can't understand.
Is it only a matter of time before they lock us out of the conversation completely? Steels: And I think it's actually totally possible, but society will kind of have to find the balance between what it is that we want robots for and how much autonomy are we willing to give them.
Freeman: If we're giving robots autonomy to move, to feel, to make their own language, could that be enough for them to surpass us? After all, what's robot for "Exterminate"? But one Japanese scientist doesn't see the future as robots versus humans.
In fact, he is purposefully engineering their intersection.
We know that homo sapiens cannot be the end of evolution.
But will our descendents be biological or mechanical? Some believe that intelligent machines will eventually become the dominant creatures on earth.
But the next evolutionary step may not be robot replacing human.
It could be a life-form that fuses man and machine.
This is yoshiyuki sankai.
Inspired by authors like Isaac Asimov, he has always dreamed of fusing human and robotic life-forms into something he calls the hybrid assistive limb system Or h.
A.
L.
Sankai: That one is one of my dreams.
We could develop such kind of devices, like a robot suit h.
A.
L.
System, for supporting the humans and human's physical movements.
Freeman: And now, after 20 years of research, he has succeeded.
H.
A.
L.
Assists the human body by reading the brain's intentions and providing assistive power to support the wearer's movement.
If she wish to or try to move as her brain generates intentions, and the robot detects these intention signals and wants to assist her movements.
Freeman: When the brain signals a muscle to move, it transmits a pulse through the spinal cord and into the area of movement.
This bioelectric signal is detectable on the surface of the skin.
Yoshiyuki designed the h.
A.
L.
Suit to pick up these impulses and then activate the appropriate motors in order to assist the body in its movement.
The human brain is directly controlling the robotic suit.
It's not just a technological breakthrough.
Yoshiyuki already has h.
A.
L.
Suits at work in rehabilitation clinics in Japan.
So, if some of the body has some problems, like a paralyzed So this would help patients, who are such kind, and a handicapped person can use it.
Freeman: People who haven't walked in years are now on the move again thanks to these brain-powered robot legs.
Yoshiyuki has also developed a model for the torso and arm that can provide up to 200 kilograms of extra lifting power, turning regular humans into strongmen.
But it's not all about strength.
He believes the merging of robotic machinery and human biology will allow us to preserve great achievements in movement.
Athletes like Tiger Woods or Roger federer bring unique skill and artistry to their sports.
However, when they die, so does their movement.
But since the h.
A.
L.
Suit can detect and memorize the movements of its wearer, that knowledge doesn't have to disappear.
If some of these athletes like Tiger Woods, if they wear it and they swing it, every data -- motion data and physiological data -- also gather in the computers.
Freeman: We once built great libraries to preserve knowledge expressed through writing for future generations.
Yoshiyuki wants to create a great library of movement.
By merging our bodies with robotic exoskeletons, we will not only be stronger.
We will all move as well as the most talented athletes and artists.
The last century of popular culture has focused on apocalyptic scenarios of robotic mutiny.
But the h.
A.
L.
Suit opens up a different future.
We tend to think about robotics as an alternative life-form that may someday replace or compete with humans.
But I think the reality of the matter is that, increasingly, we'll see humans and robots cooperate and actually become one kind of species both physically and mentally.
Schmidt: Absolutely, I think robots are the future.
I think we need to rely on them.
Otherwise, we will stagnate and make no more progress.
Eventually, life on the earth will come to an end.
What is our legacy? We will leave nothing unless we leave consciousness.
We need conscious robots everywhere.
That will be our legacy.
That will be the legacy of mankind.
Robots are rapidly becoming smarter, more agile, and are developing human traits like consciousness, emotions, and inspiration.
Will they leave us behind on the evolutionary highway, or will humans join the machines in a new age? Evolution is unpredictable And is bound to surprise us.
Robots live and work beside us.
And now we're designing them to think for themselves.
We're giving them the power to learn to move on their own.
Will these new life-forms evolve to be smarter and more capable than us? Or will we choose to merge with the machines? Are robots the future of human evolution? Space, time, life itself.
The secrets of the cosmos lie through the wormhole.
We humans like to think of ourselves as the pinnacle of evolution.
We are the smartest, most adaptable form of life on earth.
We have reshaped the world to suit our needs.
But just as homo sapiens replaced homo erectus, it's inevitable something will replace us.
What if we're building our own successors? Just as we learn to move, think, and feel for ourselves, we're now giving robots those same powers.
Where will this lead? Is this what humanity will become? When I was a teenager, I built a bicycle from spare parts.
My bicycle was so well-balanced, i could jog alongside it without holding onto it.
Nobody was that impressed, but it made me think.
Would we one day have machines that truly could move on their own? Would they even need us anymore? Daniel wolpert of the university of Cambridge believes that if robots are to be the future of human evolution, they're going to have to learn to move as well as we do Because movement is the supreme achievement of our powerful intellect.
The most fundamental question I think we can ever ask is, why and when have animals ever evolved a brain? Now, when I ask my students this question, they'll tell me, we have ones to think or to perceive the world, and that's completely wrong.
We have a brain for one reason and one reason only, and that's to produce adaptable and complex movement, because movement is the only way we have of affecting the world around us.
Freeman: All of our brains' intellectual capacity grew from one primal motivation -- to learn how to move better.
It was our ability to walk on two legs, to speak and emote with complex facial movements, and to manipulate our dexterous limbs that put humans on top of the food chain.
There can be no value to perception or emotions or thinking without the ability to act.
All those other features, like memory, cognition, love, fear play into movement, which is the final output of the brain.
Freeman: No machine could handle the huge variety of complex movements we perform every day.
Just imagine a robot trying to play one of england's most famous pastimes.
So, although that shot looks simple and it felt effortless to me, the complexity of what's going on in my brain is really quite remarkable.
I have to follow the ball as the bowler bowls it and predict where it's going to bounce and how it's gonna rise from the ground.
I then have to make a decision as to what type of shot I'm going to make.
And finally, I have to contract my 600 muscles in a particular sequence to execute the shot.
Now, each of those components has real mathematical complexity, which is currently beyond the ability of any robotic device.
Freeman: One of the greatest challenges in getting robots to move like we do is teaching them to deal with uncertainty -- something our brains do intrinsically.
A ball will never come at you the same way twice.
You must instantly adjust your swing each time.
The question is How does the human brain deal with all this uncertainty? Daniel thinks it uses a theory of probability estimation called bayesian inference to figure it out.
Wolpert: So, a critical thing the batsman now has to do is decide where this ball is going to bounce, so as they can prepare the correct shot, and for that, they need bayesian inference.
What bayesian inference is all about is deciding, optimally, the bounce location of the ball from two different sources of information.
Freeman: One source of information is obvious.
You look at the ball.
Wolpert: So, you can use vision of the trajectory of the ball as it comes in to try and estimate where it's going to bounce.
But vision is not perfect, in that we have variability in our visual processors, so at least where distribution's shown in red here as the probable bounce locations.
But bayes' rule says there's another source of information.
It's the prior knowledge about possible bounce locations.
If you're a good batter, then you can effectively look at the bowler and maybe know by his particular bowling style or small cues -- and that's shown by the blue shading -- which is a different area.
So, bayesian inference is a way of combining this red distribution with the blue distribution, and you do that by multiplying the numbers together in each to generate this yellow distribution, which is termed the belief.
And using that information, the batsman can now prepare his shot.
Okay, I should probably get out of the way.
Freeman: The batsman's brain, like all of ours, is doing this math automatically.
We are a species that's honed for movement prediction.
It's what has made us the planet's best hunters and toolmakers.
We already have robots that are faster and more accurate than we are.
But we have to program their every move.
For robots to walk down the evolutionary road we've already traveled, they're going to have to learn to move on their own.
What happens then? Will they evolve complex brains like ours? Robot builders Josh bongard of the university of Vermont and hod lipson of Cornell university are trying to answer that question.
Increasingly, we see that interaction with the world, with the physical world, is important for intelligence.
You can't just build a brain in a jar.
Freeman: Hod and Josh's goal is to build a machine that's smart enough to learn how to move around all by itself.
They've created a menagerie of strange robotic forms along the way.
But their work starts with the computer program designed to evolve robot bodies.
It simulates various body plans and then tries various strategies to get them to move.
Okay, so, let's walk our way through -- no pun intended -- an actual evolutionary simulation.
So, in this case, we've told the computer that we want a robot that has two legs, but we want the computer to figure out how to get the robot to orchestrate the movement of the robot's legs.
And here, we see something a little bit surprising, that evolution hasn't discovered the solution that we use.
Sometimes when we run this evolutionary process, it produces something familiar like walking, and in other cases, it produces something that's not familiar, something we wouldn't have come up with on our own.
Freeman: It's survival of the fittest or perhaps the least awkward.
Just as mother nature selects generations based on their ability to survive, so does the simulation.
The computer deletes the robots that aren't doing a very good job, and the computer then takes the robots that are doing a slightly better job and makes modified copies of them and repeats this process over and over again.
And after a while, the computer starts to discover robots that, in this case, are able to walk from the left side of the screen to the right side of the screen.
Freeman: This is evolution on steroids.
What took mother nature millions of years takes the computer just a few hours.
Overnight, the computer tests thousands of generations, and eventually it produces a robot that meets the goal.
When the simulation makes something that looks particularly interesting, hod and Josh take that body plan and build it.
Now they can test whether the strategies for moving learned in simulation work as well in the real world.
So, this robot is called the quadratot, and it's basically a robot that learns how to walk using evolutionary robotic techniques.
And so, what we can see here is a particular example of how this robot learns.
This is one of the earliest gaits that it did, and we can see that it's not moving very far or very fast.
It's kind of like a child during its very early behaviors of crawling.
It's trying out different things.
Some things work better.
Some things work less well.
And it's taking the experiences and learning from them and gradually improving its gait.
Freeman: There are many robots that can move well while executing a specific predesigned task.
But hod and Josh's robots must start to learn by themselves from scratch in an unknown environment.
It can sense its own progress, and like a baby learning to crawl, it becomes more aware of its body with every step and every tumble.
Hod and Josh believe this self-awareness gradually builds into a basic form of consciousness.
Often, we've raised the question, is something conscious, or is it not? But it's really not a black-and-white thing.
It's more about to what degree an entity's able to conceive of itself, to simulate itself, to think about itself, to be self-aware.
As robots learn to move in more complex ways, it's possible they will develop levels of consciousness equal to ours and maybe beyond.
But according to one scientist, for a robot to become truly conscious it must develop feelings.
What is consciousness? The answer depends on who you talk to.
A doctor's definition would be different from a priest's.
But we all agree that our high-level consciousness is what separates us from other organisms and, of course, from robots.
What would it take for robots to become conscious? Can they get there on logic alone? Or must they also learn to feel? Professor pentti haikonen from the university of Illinois believes machines will only become conscious when they can experience emotions.
It's a belief he has held since he was very young when he first contemplated what it meant to be conscious.
When I was 4 or 5 years old, I was standing in our kitchen, and suddenly I was struck by the mystery of existence how and why I was me and why I was not my sister or my brother.
How did I get inside myself? Freeman: As he got older, pentti realized that what made him feel conscious of being inside his head were his emotional reactions to the people and objects in the world around him.
The neural processes that are behind our consciousness take place inside our brain, but we don't see things that way.
For instance, when you cut your finger, the pain is in the finger or so it appears, but, actually, the pain is in here.
To feel is to be conscious.
Freeman: Our brain's raw experience of the world around us is just a series of electrical impulses generated by our senses.
However, we translate these impulses into mental images by making emotional associations with them.
A sound is pleasing.
A view is peaceful.
Consciousness, according to pentti, is just a rich mosaic of emotionally laden mental images.
He believes that to have a truly conscious machine, you must give it the power to associate sensory data with emotions.
And in this robot, he has begun that process.
This is the first robot that utilizes associative neural networks.
It is the same kind of learning that we humans use.
When we see and hear something, we make a connection between those things, and later on when we see or hear the other thing, the other thing comes to our mind.
Freeman: The xcr-1 experiences the world directly through its senses like we do.
On board are the basics of touch, sight, and sound.
Pentti has begun the process of giving it emotional associations to specific sensory data, like the color green.
Pentti places a green object in front of the robot, which it recognizes.
Green.
Then he gives green a bad association -- a smack on the backside.
The associative learning is similar to little children.
Hurt.
And you say that this is not good or this is good, or you may also smack the little child.
Hurt.
I don't recommend that.
Green bad.
Freeman: The robot's mental image of the green object is now associated with the emotion bad.
And from now on, it will avoid the green bottle.
But it's not all pain for the xcr-1.
Just like we teach the robot to associate pain with the green object, we can teach the robot to associate, also, pleasure with objects, in this case with the blue object, like this.
Blue.
Freeman: To give blue a good association, pentti gently strokes the top of the robot.
Blue good.
This simple experiment demonstrates that this robot has mental images of objects and mental content.
Freeman: It's still early in its development, but the xcr-1 has learned the basics of emotional reaction from fear Green bad.
Green bad.
to desire.
Blue good.
Now's my time for love lonely moments seem to As a more advanced version of the xcr-1 fills its memory with mental images Dentist bad.
it will start to be able to react to new situations on its own and eventually experience the world much like any emotionally-driven being.
It is my great dream to build robot that is one day able to ask, "How did I get inside myself?" Freeman: Once robots reach this point, what's to stop them from moving on and becoming conscious of things we're not? This man thinks robots will become the future of humanity because they'll have something we lack.
Their brains will have the capacity for genius long after the last human ever says "Eureka.
" for Archimedes, Eureka happened in the bathtub.
Einstein was riding a streetcar when relativity dawned on him.
These brilliant minds had a flash of inspiration and drove all of humanity forward.
But the scientific questions of today, probing shoals of subatomic particles and our vast genetic code, have become so complex that they take teams of thousands of researchers to solve.
Is the age of the single scientific genius over? Not if machines have their way.
Data scientist Michael schmidt sees the world filled with intricate beauty -- the flowering of a rose, the veins branching on a leaf, the flight of a Bumblebee.
But below the surface of nature's wonders, Michael also sees a treasure trove of uncharted mathematical complexity.
Schmidt: Well, I love coming out here.
Nature is beautiful.
There are equations hidden in every plant and every bee and the ecosystems involved in this garden.
And part of science is figuring out what causes those things to happen.
Freeman: Science is our effort to make sense of nature, and this quest has given us some very famous discoveries.
In Newton's time, he was able to figure out a very important rule in physics, which is the law of gravity.
It predicts how this apple falls and the forces that act upon this apple.
Today in science, we're interested in similar problems but not just about how the apple falls but the massive complexity that follows from this very simple dynamic to the world around us.
For example, when I drop this apple, the apple stirs up dust.
This dust could hit a flower, and a bee may be less likely to pollinate that flower.
And the entire ecosystem in this garden could change dramatically from that single event.
Freeman: Scientists understand the basic forces of nature, but making precise predictions about what will happen in the real world with its staggering complexity is overwhelming to the human mind.
So, one of the reasons why it's extremely difficult for humans to understand and figure out the equations and the laws of nature is literally the number of variables that are at play.
There could be thousands of variables that influence a system that we're only just beginning to tease apart.
In fact, there are so many of these equations, we'll never be able to finish analyzing them if we do it by hand.
Freeman: In 2006, Michael began developing intelligent computer software that could observe complex natural systems and derive meaning from what seems like chaos.
So, what I have here is a double pendulum.
If you look at it, it consists of two arms.
One arm swings along the top axis, and the second arm is attached to the bottom of the first arm, and it's two pendulums that are hooked together, one pendulum at the end of the other.
Now, the pendulum is a great example of complexity because it exhibits some of the most complex behavior that we're aware of, which is called chaos.
So, when you collect data from this sort of device, it looks almost completely random, and there doesn't appear to be any sort of pattern.
But because this is a physical deterministic system, a pattern does exist.
Freeman: Finding a pattern amidst the chaos of the double pendulum has stumped scientists for decades.
But then Michael had a flash of inspiration.
Why not grow new ideas the same way nature created us, using evolution? He called his program Eureka.
Eureka starts with a primordial soup of random equations and checks how closely they fit the behavior of the double pendulum.
If they don't fit, the computer kills them.
If they do, the computer moves them into the next generation, where they mutate and try to get an even closer fit.
Eventually, a winning equation emerges, one that Archimedes would be proud of.
Eureka! Schmidt: And I'm running our algorithm now.
On the left pane are the lists of the equations that Eureka has thought up for this double pendulum.
Walking up, we can see we increase the complexity, and we're also increasing the agreement with the data.
And eventually, as you go up, you start to get an extremely close agreement with the data, and eventually you snap on to a truth where you get a large improvement in the accuracy.
And we can actually look in here and see exactly what pops out.
For example here, you might notice we have a 9.
8, and if you remember from physics courses, that is the coefficient of gravity on earth.
What's very important is the difference between the two angles of the double pendulum.
This pops out.
Essentially, we've used this software and the data we've collected to model chaos, and we've teased out the solution directly from the data.
Freeman: Eureka has not only discovered a single equation to explain how a double pendulum moves.
It has found meaning in what looks like chaos -- something no human or machine has done before.
Schmidt: So, we could collect an entirely new data set, run this process again, and even though the data is completely different -- we could have different observations -- we can still identify the underlying truth, the underlying pattern, which is this equation.
Freeman: To Michael, the future of scientific exploration isn't inside our heads.
It's inside machines.
Whether they're looking at patterns of data from genetics, particle physics, or meteorology, programs like Eureka can evolve inspiration on demand, finding basic truths about nature that no human ever could.
We're gonna reach a point where we decide what we want to discover and we let the machines figure this out for us.
Eureka can find these relationships without human bias and without human limitations.
We created robots to serve us.
As the machines learn their own ways to move, feel, and think, they will eventually grow out of that role.
What if they start working together? Could they build their own society, one made by the robots for the robots? There's no species on earth more successful than us.
We owe that success to the powerful computer inside our heads.
But it takes more than one brain to conquer a planet.
Homo sapiens thrive because we have learned to make those computers work together as a society.
What will happen when robots put their heads together? Roboticist by day and gourmet chef by night, Professor Dennis Hong of Virginia tech is a specialist in building cooperative robots.
But he also sees cooperation outside the lab.
So, we don't really think about it, but everything in our daily lives involves cooperation.
For example, cooking oftentimes is thought of as a solo act, but if you think about it, a lot of people are involved and a lot of careful coordination is required to make it happen.
Oh, thank you, charli.
Take this tomato as an example.
This tomato most likely started its life as a seed, where a group of breeders need to choose the right sequence of genes for a plump, juicy, tasty tomato.
The seeds needed to be planted, grown, harvested, then the tomatoes needed to get to the market.
Freeman: Food production is a complex web of coordination.
But as good as it is, human cooperation has its limits.
Hong: Oops.
Freeman: Every day, like most of us, Dennis has to contend with the prime example of human cooperation gone wrong -- traffic.
The problem is, us being human, we all need to, want to get to our destination as quick as possible, thus we have traffic jams.
Freeman: If it wasn't for traffic lights, which are, in reality, very simple robots, it will be almost impossible to get anywhere.
These traffic lights, they talk to each other.
They communicate with other traffic lights at other intersections.
And they have cameras, so they actually see the traffic patterns and make decisions for us, for humans.
Oh, there you go.
Thank you, traffic light.
Freeman: Traffic is a nuisance.
But other failures of human cooperation are much more serious And often deadly.
[ Machine-gun fire .]
Dennis believes a society of robots can be much better collaborators than we are.
So, in collaboration with Daniel Lee at the university of Pennsylvania, he designed a group of robots to compete in the robocup, an international robotic soccer championship.
Robocup is an autonomous robot soccer competition, which means that you have a team of robots, you press "Start," And then nobody touches anything.
And the robots need to look around, see where the ball is, need to coordinate and actually play a game of soccer.
Freeman: Dennis' soccer robots, called Darwin-op, are fully autonomous.
They use complex sensors and software to navigate the playing field.
And they have a serious competitive edge over their human counterparts.
Teammates can read each other's minds.
So, if you look at human soccer players, obviously they're great at what they do.
They communicate sometimes by shouting, sometimes by a subtle gesture, but, again, it's not really accurate, and they cannot share all the information together at the same time in real time, but robots can do that.
Freeman: Each robot knows the exact location and destination of the other robots at all times.
They can adjust their strategy and even their roles as necessary.
Hong: Depending on where the ball is, where the opponents are, they didactically switch their roles.
So the goalie becomes a striker, a striker becomes a goalie or defense.
Freeman: They may not be as agile as pelé or bend it like Beckham, but they are able to dribble past their opponents, pass the ball, score a goal And even celebrate.
Dennis believes robocup is just the beginning of robot societies.
Dennis imagines a connected community of thinking machines that would be far more sophisticated than human communities.
He calls it cloud robotics.
Hong: Cloud robotics is a shared network of intelligence.
It's similar to what we call common sense in humans.
So, just like those smaller robots that play soccer for robocup, they share a common data, team data, to achieve the goal, in this case, winning the soccer game.
For cloud robotics, robots from the furthest corners in the world, they can all connect to the cloud and share information and intelligence to do their job.
Freeman: Humans spend a lifetime mastering knowledge, but future robots could learn it all in microseconds.
They could create their own hyper-connected network using the same spirit of cooperation that built human society without the selfishness and greed that hold us back.
Robots operate by a very well-defined set of rules.
The human impetus to break them is just not there.
Freeman: Robots already know how to talk to one another.
But now a scientist in Berlin has taken robotic communication a step further.
His machines are speaking a language he doesn't understand.
Motakay tokima.
Did you understand what I just said? Of course you didn't because I wasn't speaking any known human language.
But it wasn't nonsense.
It was a robot language.
We humans took tens of thousands of years to develop our complex means of communication.
Now robots are following our lead, and they're doing it at light speed.
Someday soon, robots may decide to exclude us from their conversation.
Robot: Tokima.
Lucabo.
Miyoto.
Motakay.
Tokima, kymamu.
Tokima.
Simeta.
Tokima.
Motakay.
Steels: Without language, our species would never be where it is today.
It's the most magnificent thing that has ever been created by humanity.
If you look at ourselves, then it's pretty clear that without language, we would not be able to do the kinds of things that we're doing.
Freeman: Luc steels, a Professor of artificial intelligence, sees language as the key to developing true robot intelligence.
Steels: What I'm trying to understand is, how can we synthesize this process so that we can start up a kind of evolution in a robot or in a population of robots that will also lead to the growth of a rich communication system like we have.
Freeman: Machines already communicate with each other, but these are based on predetermined, human-coded languages.
Luc wants to know how future robot societies might communicate given the chance to make a language on their own.
Luc gives his robots the basic ingredients of language, like potential sounds to use, and possible ways to join them together.
But what the robots say is up to them.
We put in learning mechanisms, we put in invention mechanisms, mechanisms so that they can coordinate their language.
They can kind of negotiate how they're gonna speak, but we don't put in our language or our concepts.
Freeman: It's not enough for the robots to know how to speak.
They need to have something to speak about.
Luc's next step is to teach the robots how to recognize their own bodies.
Steels: In order to learn language, you actually have to learn about your own body and the movements of your own body.
So, what you see here is an internal model that the robot is building of itself.
This robot is trying to learn here, is the relationship between all these different sensory channels and its own motor commands.
Freeman: As a robot watches itself move in the mirror, it forms a 3-d model of its limbs and joints.
It stores this information in sense memory and is now ready to talk to another robot about movement.
Tokima.
So, now, this robot is talking.
He's asking an action.
This robot is doing -- you know, stretching the arm.
No, this is not what was requested, and he's showing again what the right action is.
Freeman: She's unsuccessful in her first attempt, but eventually the robot learns that "Tokima" Means "Raise two arms.
" after repeating this process with different words, they try again.
Homakey.
Another request.
He's doing the action.
Yes, this is the right kind of action.
So, in other words, this robot has learned the word from the other one and vice versa.
They now have a way to talk about actions.
Freeman: The robots' language is already so well-developed, they can teach it to Luc.
Let's see what, you know, what he asks me to do.
Motakay.
Okay, motakay.
No, this is not right, so he's showing it to me.
Okay, I'm learning this gesture now.
Motakay.
Motakay.
Motakay is this.
Okay, I got it right.
So, now I'm going to use the gesture with him.
Motakay.
Okay, yes, you're doing the right thing.
Thank you.
As the robots repeat this process, they generate words and actions that have real meanings for one another.
And so the robots' vocabulary grows.
Every new word they create is one more that we can't understand.
Is it only a matter of time before they lock us out of the conversation completely? Steels: And I think it's actually totally possible, but society will kind of have to find the balance between what it is that we want robots for and how much autonomy are we willing to give them.
Freeman: If we're giving robots autonomy to move, to feel, to make their own language, could that be enough for them to surpass us? After all, what's robot for "Exterminate"? But one Japanese scientist doesn't see the future as robots versus humans.
In fact, he is purposefully engineering their intersection.
We know that homo sapiens cannot be the end of evolution.
But will our descendents be biological or mechanical? Some believe that intelligent machines will eventually become the dominant creatures on earth.
But the next evolutionary step may not be robot replacing human.
It could be a life-form that fuses man and machine.
This is yoshiyuki sankai.
Inspired by authors like Isaac Asimov, he has always dreamed of fusing human and robotic life-forms into something he calls the hybrid assistive limb system Or h.
A.
L.
Sankai: That one is one of my dreams.
We could develop such kind of devices, like a robot suit h.
A.
L.
System, for supporting the humans and human's physical movements.
Freeman: And now, after 20 years of research, he has succeeded.
H.
A.
L.
Assists the human body by reading the brain's intentions and providing assistive power to support the wearer's movement.
If she wish to or try to move as her brain generates intentions, and the robot detects these intention signals and wants to assist her movements.
Freeman: When the brain signals a muscle to move, it transmits a pulse through the spinal cord and into the area of movement.
This bioelectric signal is detectable on the surface of the skin.
Yoshiyuki designed the h.
A.
L.
Suit to pick up these impulses and then activate the appropriate motors in order to assist the body in its movement.
The human brain is directly controlling the robotic suit.
It's not just a technological breakthrough.
Yoshiyuki already has h.
A.
L.
Suits at work in rehabilitation clinics in Japan.
So, if some of the body has some problems, like a paralyzed So this would help patients, who are such kind, and a handicapped person can use it.
Freeman: People who haven't walked in years are now on the move again thanks to these brain-powered robot legs.
Yoshiyuki has also developed a model for the torso and arm that can provide up to 200 kilograms of extra lifting power, turning regular humans into strongmen.
But it's not all about strength.
He believes the merging of robotic machinery and human biology will allow us to preserve great achievements in movement.
Athletes like Tiger Woods or Roger federer bring unique skill and artistry to their sports.
However, when they die, so does their movement.
But since the h.
A.
L.
Suit can detect and memorize the movements of its wearer, that knowledge doesn't have to disappear.
If some of these athletes like Tiger Woods, if they wear it and they swing it, every data -- motion data and physiological data -- also gather in the computers.
Freeman: We once built great libraries to preserve knowledge expressed through writing for future generations.
Yoshiyuki wants to create a great library of movement.
By merging our bodies with robotic exoskeletons, we will not only be stronger.
We will all move as well as the most talented athletes and artists.
The last century of popular culture has focused on apocalyptic scenarios of robotic mutiny.
But the h.
A.
L.
Suit opens up a different future.
We tend to think about robotics as an alternative life-form that may someday replace or compete with humans.
But I think the reality of the matter is that, increasingly, we'll see humans and robots cooperate and actually become one kind of species both physically and mentally.
Schmidt: Absolutely, I think robots are the future.
I think we need to rely on them.
Otherwise, we will stagnate and make no more progress.
Eventually, life on the earth will come to an end.
What is our legacy? We will leave nothing unless we leave consciousness.
We need conscious robots everywhere.
That will be our legacy.
That will be the legacy of mankind.
Robots are rapidly becoming smarter, more agile, and are developing human traits like consciousness, emotions, and inspiration.
Will they leave us behind on the evolutionary highway, or will humans join the machines in a new age? Evolution is unpredictable And is bound to surprise us.