It's time to rebuild the world for robots

We smoothed the world for cars, but assume robots will have eyes and ears

We redesigned the world for automobiles and now it's time to redesign it for robots.

To understand why, consider the sad story of the clothes-folding robot.

It turns out that after many years and much research you can get a robot to fold your laundry. But the robot basically sucks at it for pretty much the same reasons that printers, which are robots too, have a great deal of trouble handling individual sheets of paper.

That’s unsurprising. For five hundred years paper has been designed with the human hand in mind, and has a ‘tooth’ that’s evolved to engage the whorls of our fingerprints. So while it's easy for us to pick up a single sheet of paper, robots need all sorts of springs and wheels to move one a few inches. And forget a robot lifting a piece of paper off a desk any time soon.

For robots, clothing isn’t much easier to handle than paper, because nothing about clothing is regular. Even towels pose huge problems in edge detection, object identification and modeling. It all means that a robot can spend minutes ‘staring’ at a towel, another few minutes trying to grasp it, then - once it’s been successfully grasped, a few moments folding it.

Does that mean humans are smart and robots are stupid? No. It means we humans have cheated the exam. Clothing is pre-eminently an artifact of human culture. We ‘get’ clothing because it’s an extension of our skin. Robots, in that context, act like aliens dropped in from another universe where there’s no concept of clothing.

So when we hear claims that autonomous vehicles are decades further away than they appear - because they can’t deal with a graffitied stop sign or a hopping marsupial their programmers never countered - we have to sit back and ask ourselves why for the love of all that is digital did we ever get the idea that cars had to think like us, see like us, react like us and work with the same range of inputs as us?

Just because driving is a messy and complex task for us meat sacks doesn’t mean it should be anywhere near as complex for robot. Robot needs the affordances suitable to a robot, like sensors in the roadway. They don't need a stop sign, but do need an accurate and continuously-updated map of where the stop signs are.

And although autonomous vehicles will need to rely on their lidar prevent collisions, we shouldn’t expect that we can simply drop an autonomous vehicle into a road system that’s been designed around human capacities and expect them to perform like we do. There’s no way to shortcut the nearly four billion years of evolution that gave us our unique configuration of senses and cognition. That’s a test an autonomous vehicle can only fail.

We therefore need to redesign our material world to give robots all they need to find their way through this world. We need to put those sensor networks everywhere, map the world and all of its peculiarities, digitise all of the relevant data, then make it accessible to robots as they move autonomous through that world.

If this sounds like a big ask - or worse, like putting the cart before the horse - consider this: a hundred years ago we completely rewrote the fabric of the material world to accommodate the high-tech invention of the time - the automobile. Out went cobblestone and dirt roads, in came the bitumen. Up went stop lights and street signs. We completely redesigned culture to accommodate the automobile and the freedoms it provides.

Why wouldn’t we do that again, this time for autonomous vehicles? Why wouldn’t we want to make it easy for these new robots to do their best possible job? Why would we want to see them struggle to be human? Are we that worried about losing the race that we want robots to run it blindfolded, with both arms tied behind their back?

That we haven’t even begun to build a robot-friendly world tells me that the race to autonomy reveals more about ourselves than our creations. I wonder if we should feel a bit embarrassed about that. ?


Biting the hand that feeds IT ? 1998–2017