What Boston Dynamics robots mean (and what they

Boston Dynamics posted another video. Atlas, their humanoid robot, is doing parkour now. Jumping over logs. Doing backflips. Moving with a fluidity that’s unsettling in a way that’s hard to pin down.

And right on cue, the internet splits into two camps.

Camp one: “This is so cool!”

Camp two: “This is terrifying, we’re all going to die.”

I think both camps are missing the point.

What Atlas actually is (and isn’t)

Let me be clear about something. Atlas is a research platform. It’s not a product. You can’t buy one. It’s not going to show up at your door. It’s a very expensive, very fragile, very limited machine that can do a small number of things in a controlled environment after extensive programming and many, many failed attempts.

The videos on YouTube are highlight reels. They don’t show the hours of calibration, the failed runs, the team of engineers standing just off camera. Atlas didn’t learn to do backflips the way a gymnast does, through practice and body awareness. It was programmed with specific trajectories and force profiles for that specific movement on that specific surface.

So the “cool!” camp is overreading what Atlas can do right now.

But the “terrifying!” camp is also wrong, just in the opposite direction. They’re overreading the present and underreading the trajectory.

The trajectory is the thing

Here’s what I think matters.

In 2013, Boston Dynamics released a video of Atlas walking. Just walking. On flat ground. And it looked drunk. Stiff, lurching, clearly on the edge of falling at any moment. If you showed that video to a physical therapist, they’d diagnose a serious neurological condition.

Five years later, the same platform is doing backflips and parkour.

Five years.

What does the five years after that look like? What about fifteen? I don’t know. Nobody does. But the rate of improvement is the signal, not the current capability.

Think about it this way. The Wright Brothers flew 120 feet in 1903. Twelve seconds of flight. A strong kid on a bike could have kept up. Twenty years later, we had commercial airlines. Forty years later, we broke the sound barrier. Sixty years later, we were on the Moon.

The first flight wasn’t impressive on its own terms. It was impressive because of what it implied.

The uncanny valley, sideways

There’s a concept called the uncanny valley: the idea that a robot that looks almost-but-not-quite human triggers a deep revulsion in us. We’re fine with robots that look like machines. We’d be fine with robots that look perfectly human. It’s the in-between that makes our skin crawl.

I think there’s a movement equivalent. An uncanny valley of motion.

Watch a video of Atlas and pay attention to the moment your brain shifts from “that’s a machine” to “that’s… something.” It happens when the robot recovers from a stumble. When it adjusts its balance mid-jump. When it does something that feels reactive rather than programmed.

A perfectly smooth, mechanical motion reads as machine. A perfectly fluid, natural motion would read as alive. Atlas is in between. It moves with a fluidity that doesn’t match its metal body. Your eyes see a machine. Your brain reads an animal. And something in the gap between those two signals feels wrong.

I wonder if we’ll get used to it. I wonder if in twenty years, a robot moving like an animal will be as unremarkable as a plane flying through the sky. Probably. We get used to everything.

But right now, in 2018, watching Atlas do a backflip makes the ancient parts of my brain very uncomfortable, and I find that reaction more interesting than the robot itself.

The real question nobody’s asking

Everyone argues about whether robots will take our jobs or kill us all or become our friends. Those are fine questions, I guess. But I think they skip over a more basic question that nobody seems to be asking.

What is the physical world going to look like when machines can move through it as fluidly as animals?

Right now, robots are mostly confined to factories. Bolted to the floor. Moving in precise, repeating motions. The world comes to them (on conveyor belts, in organized trays, in predictable sequences).

Boston Dynamics is working on the opposite problem. Robots that go out into the world. The unstructured, chaotic, cluttered world where the floor isn’t flat, the lighting isn’t perfect, and someone left a garden hose in the middle of the path.

If they solve this (and I think they will, eventually), the implications aren’t about jobs or warfare or any single application. The implications are about the fundamental relationship between machines and physical space.

Machines in our homes. Machines in our forests. Machines on construction sites and disaster zones and Mars. Not wheeled platforms limited to smooth surfaces, but machines that can go anywhere a human (or a dog, since Spot exists) can go.

The part from Asimov nobody quotes

Everyone quotes Asimov’s Three Laws of Robotics. Don’t harm humans, obey orders, protect yourself. They’re famous.

But there’s a moment in one of Asimov’s later robot novels where a character realizes that the most dangerous thing about a robot isn’t its strength or intelligence. It’s that it doesn’t get tired. It doesn’t get bored. It doesn’t lose motivation. It just keeps going, forever, at the same level of performance, until it physically breaks down.

I think about that when I watch Atlas do parkour.

The robot isn’t impressive because it can do a backflip. A human gymnast can do a better backflip. The robot is impressive because it can do a backflip at 3am on a Tuesday in the rain without complaining, without a coach, without a desire to do anything else.

And the next version will do it better. And the one after that. And the one after that. And none of them will ever get tired.

That’s not cool or terrifying. It just is. And I think we’d better start thinking about what it means.

The MIT CSAIL robotics lab is working on similar problems from the AI side. Boston Dynamics attacks it from the mechanical engineering side. Somewhere in the middle, these two approaches are going to merge. And when they do, the videos won’t be highlight reels of single tricks.

They’ll be Tuesday afternoon.