What am I?

Hi there.

I'm chris.

Atkinson.

Chris is a professor at the robotics institute at Carnegie Mellon University.

Today.

I'll be breaking down clips from movie and tv about artificial intelligence and robotics out of control robots.

I robot detective.

What are you doing?

You said they've all been programmed with the three laws.

Isaac Asimov was one of the earliest science fiction writers who focused on robots and he came up with this scheme of the three laws.

Yeah, I know the three laws.

Your perfect circle of protection robots don't hurt humans but doesn't the second law state that a robot has to obey any order given by a human being, except if it causes a violation of the first law, right?

But the third law states that a robot can defend itself unless it causes a violation of the second law.

And the first law, we have 1000 robots that will not try to protect themselves if it violates a direct order from a human.

And I'm betting one who will got you get the hell out of here.

What am I?

You saw a robot that somehow it's three laws were disabled and that left it with an existential crisis, what am I?

Which is sort of similar to what am I supposed to do next?

If I don't have any guiding purpose and if it was a case that robots ran on what we call expert systems or sets of laws that might actually be a reasonable way to program robots.

You know what they say, laws are made to be broken.

But nowadays we're actually programming robots in a very different way by giving them lots of training examples and having them essentially learn parameters in formulas so that they do the right thing.

So we have this mismatch between logical rules and numbers in a formula.

So in robot movie after robot movie, they're obsessed with the robots actively turning against the humans and started killing humans.

And I get that the goddamn robots john far more likely is the robot will screw up, it will make a mistake and that mistake will have bad consequences.

It's a robot.

It doesn't need a motive.

It just has to be broken.

Deactivating an android Blade Runner.

Should I really shoot the crazy robot that's coming to attack me?

Yeah, shooting a robot.

It is a potentially a pretty good way to stop a robot.

Now there might be parts of its chassis that you shoot a bullet through.

It doesn't do anything, it doesn't cut any wires.

Doesn't open up a hydraulic fluid hose.

So you might have to shoot it a bunch of times.

There are other ways to disable a robot that might not damage it in the same way.

For example, there's something called an electromagnetic pulse which will fry all its circuits but leave the mechanics intact.

Blade Runner is a fantastic movie, has this incredible vision of the future, which by the way, we've already caught up with learning from imitation terminator.

Hey buddy got a dead cat in there or what you were going to program robots to have conversations and dialogue and in many, many situations you can anticipate what happens next.

But sooner or later the robot's going to face the situation this new, it doesn't know what to say.

Dead cat in there.

You know, the robot's going to have to wing it in.

This clip was the way to go a screen came up and english words were there and it sort of moved a cursor down and picked one that's only there for the audience of the movie.

The robots aren't going to do that.

You know, it's all electronics, little transistors going blip, blip, blip and it's gonna make the decision decided our fate in a microsecond programming the new Westworld.

Did you see it?

No, give it a second.

She'll do it again.

Her finger that's not standard.

Occasionally in Westworld you see a black box looks like a fancy keyboard in which they're sort of programming various sub behaviors.

Let's call them or primitives.

There's been this controversy of, do you build up behaviors from twitch here and a twitch there or do you have a few fundamental behaviors and then you combine them.

You must have I slipped it in there without telling anyone.

It turns out you can learn a lot faster if you combine these fundamental behaviors rather than adding up a lot of twitches, he calls them reveries.

The old gestures were just generic movements.

These are tied to specific memories and this is a spoiler things in Westworld go bad because they couldn't completely wipe the memory of a robot.

The memories are purged at the end of every narrative.

They're still in there waiting to be overwritten.

He found a way to access them.

It turns out in complex machines so you've got lots of different kinds of memory in lots of different kinds of places.

For example the CPU is gonna have a little bit of memory in it.

It will also have something called cache memory.

We have what's called a memory hierarchy.

There's fast memory and then there's slower memory like a subconscious a hooker with hidden depths.

If you totally fried the machine, all you'd end up with is a broken robot.

So you can't totally wipe the machine.

It's the tiny things that make them seem real, robotics.

Lab making mr.

Right okay.

You look at that scene you say why is chemistry happening in robotics lab?

The future of robotics is where we make materials using chemistry to make soft materials.

So that's actually it looks very similar to what goes on.

And some of my colleagues labs at Carnegie Mellon.

I thought showing you these tapes might help to make you more familiar with the drawing.

We video record all our experiments and in fact often you know when we press an on button to say do something that simultaneously turns the cameras on programming the android takes him just so far.

The rest must be learned.

Watching video feedback is very helpful to debug behavior.

We had some difficulty with those gross motor functions before we modified his cerebral muscular coordination.

Okay, so you just saw some mumbo jumbo.

He's mixing things we'd say about a human who had a disease.

We modified his cerebral muscular coordination and a robot that doesn't know how to walk gross motor functions.

Oh, my robots actually fell down much more than their robot did robot malfunction.

Austin powers.

Mm hmm.

Can you design an input to a robot that will cause it to malfunction or crash?

Only Austin Powers can do that.

Yeah.

Baby multi agent robotics.

Minority Report One area of research and robotics is what we call multi agent robotics.

Which is to get a bunch of robots to work together.

Mm hmm.

We model that on humans working together like a sports team or when we're searching for somebody lost in the woods.

Small robots in this clip, it would seem very smart and complicated.

The dream in robotics is can we make a lot of stupid cheap robots that by working together?

Get the job done.

What we saw in this clip is they were using some kind of imaging radar to figure out where people are likely to be tom cruise hidden in the bathtub that water shields him from radar as you saw.

But it also tries to hide his thermal signature.

So he's hotter than everything else around him And any kind of infrared imaging would have found him.

That's why he dumped ice in the water as well.

But you know, if the robots have any kind of camera that could have just looked down and there's tom Cruise.

So I think we're gonna have to do a little better than that if we're going to hide from the robots, thermal vision Westworld, the West Road clip is using what we call thermal imaging.

Far infrared imaging where you essentially can see an image of the temperature of things that are out there.

When the human got next to something that was much hotter.

The movie tried to suggest that that would hide the human.

That's actually probably not the case because the human isn't any less hot and as long as the image doesn't do what we call bloom and the whole thing saturate that robot should have been able to see the human just as well.

Mhm.

Thanks.

Human limitations, battlestar galactica.

The five of us designed you to be as human as possible.

I don't want to be human.

I want to see gamma rays.

I want to hear x rays and I want to I want to smell dark matter.

It's highly unlikely robots are going to complain about their bodies.

What they might complain about is our crappy computational hardware.

I can't even express these things properly.

A lot of it gets back to how do we build computers and right now for largely historical reasons.

We separate out the thinking part.

We'll call that the processor and the memory part.

And in order to really think about things, we've got to move everything in the memory part into the processor so it can process it.

And it turns out that's really slow.

And if you want to save stuff, you got to move it back and that's really slow.

If I'm so broken, then whose fault is that?

It's my maker's fault.

Google has something called a tpu, which they're optimizing to run something called neural networks and they're building bucket loads of these things and they're very, very different from standard computers.

But I know I want to reach out with something other than these prehensile paws.

I do believe that the way we build computers now is going to change completely in many ways.

I'm a machine and I could know much more if the robots want to help us do a better job at that.

More power to them robotics.

Lab rising sun, why do we have guys in gold suits?

I don't know why you'd wear gold suits.

You certainly in a clean room where some kind of suit.

So your dandruff doesn't ruin the chips.

They look ridiculous.

Jim.

How are you, Captain Connor, the people making the movies that we got to get us some cool looking robots.

But I've been told by reliable sources, but only you have the next generation of technology to do this kind of work.

So they actually reached out to what was then the leg lab, which was at that time at M.

I.

T.

You're looking at early robots that eventually became boston dynamics.

I'm getting out to a lot more dodger games lately.

The robot on the boom I believe was a una rue, which it was a kangaroo like robot.

So the three D.

By pad is quite famous.

It was, you know, one of the first robots that didn't have a sort of protective system to keep it from falling down.

Well, your reliable sources are wrong and it did a lot of amazing things.

Walk, run it even did flips.

In contrast the robot you saw earlier inside that that building was on a boom and it runs in a circle.

What do they think of next designed bicentennial man?

Have you given any thought whatsoever as to what age you'd like to be Officially?

I am 62 years old.

Mhm.

Let's take off 25 years.

What do you say?

15, 20, perfect.

They use soft materials to make the face of the robot.

Soft materials typically do what we call creep over long periods of time.

The animatronic figures such as the Disney presidents and whatnot all begin to sag just keeping you on your toes.

I don't have any toes.

A big problem with robot skin is after a while it gets worn, It gets cuts it sags.

It's actually a big problem because if you're loaded up with sensors and wires, It's $1 million dollar piece of skin and fixing it is a big problem.

But otherwise that's pretty much how you make a robot face.

You have some mechanism and you cover it with some soft, gooey stuff.

Origami transformers.

Mhm.

Mhm.

There are a lot of roboticists out there who want to build origami robots, robots that can reconfigure themselves by folding.

That turns out to be a great way to build robots.

Another area of robotics is more like traditional origami where you have a flat sheet and you have a folding pattern and that makes a three dimensional robot.

Why is it a good idea to start things from the flat sheet?

You can use any printing technology you want to lay down things like surfaces or materials.

But most of these folding robots on a smaller scale for structural reasons.

So they're on the meter or centimeter scale rather than the 100 or tens of meter scale.

Big guys, big guys with big guns, robotic insects, Black mirror colony collapse disorder.

We still don't know what's behind it, bees themselves were virtually extinct.

So what are ladies do is effectively stand in for them?

This clip is very close to the truth in that.

People are trying to build a robot insects right now, they can build things that are almost as small as a B.

And that can fly, we simply set the behavior and leave them to it.

They need rudimentary pattern recognition in order to locate compatible Thora navigate.

They even construct these hives themselves.

They reproduce.

How could you each hive is a replication point.

I was entertained by the hives which had these sort of squarish honeycombs.

It's like a three D printer basically.

They also had a little bit of confusion about reproduction.

They created duplicates of themselves, create more hives and spread out exponentially.

They seem to suggest that there's this big three D printer that's churning them out.