At t Siemens’ 2017 Innovation Day, Kai Wurm and his team of researchers unveiled a world first: a pair of robotic arms that could assemble something by itself. The team built cameras into robot’s wrists to detect objects which, on two large screens either side, displayed what the the machine was thinking. On one screen, the audience could watch exactly what the robot was seeing: simple basic geometric shapes against a blank background. On the other screen they could witness, in real-time, the formula the robot used to work out the problem.
The prototype is a huge step in creating a machine that can make decisions by itself. “There are many other researchers who are trying to solve this problem. But there is nothing comparable to what we have developed on the market — yet,” says Kai, who finished his PhD thesis at the Department of Computer Science at the Albert-Ludwigs University of Freiburg, and now manages a team of engineers and researchers at Siemens’ Corporate Technology headquarters in Munich, Germany.
“It took around two years to get from conception to this point,” he says. “With the current technology, you can program a robot to do anything you want it to; it’s just so tedious, only a very few people want to actually do it. By designing the basic mechanism of the brain, we can just tell a machine what to achieve, sit back, relax and let the machine figure out what it needs to do.”
It takes more than a single part to solve a puzzle
When we think about creating intelligent machines, most of us picture a lone eccentric scientist — but the reality is far from that. Machines are made up of many different components: from drives and motors to sensors and controllers, and a lot of software. It all relies on several teams, working across various departments, utilizing different skill sets.
Based in Munich, Kai’s team of researchers at Siemens Corporate Technology constantly collaborate with colleagues in the USA and China, each one piecing together a different part of the puzzle. This teamwork is essential because when it comes to building smart machines it’s not only the hardware that’s a challenge, it’s making everything work in harmony.
A game of strategy
Another challenge, says Kai, is teaching a robot to make decisions by itself. Most of Kai’s work is about comparing how humans and robots think; how to plan something and make the right choices. It’s one of the reasons he first started playing Go, an ancient Chinese game of strategy and the only game in which a computer can’t outsmart humans. Consisting of nothing more than black and white pebbles on a wooden board, Go is famous for its complexity. Considered more difficult than chess, it’s often described as the hardest game in the world because of the level of foresight and planning needed to win.
But in 2015, much to the dismay of Go players across the world, Google’s AlphaGo became the first machine to outsmart a real player. Two years later in 2017, Google’s Deepmind wanted to see if a computer could learn how to play the game from scratch, with no prior training. AlphaGo Zero quickly became the strongest player in Go’s history. “In just three days, it surpassed the standard professional player,” says Kai.
The reason why AlphaGo Zero quickly learned what took humans thousands of years is because computers are far better at simulating things then we are. Problem solving requires a linear thought process; you have to know how to support an action before it happens in order for the next step to work. In general, people can only imagine the first few steps of an activity, but computers can envisage an entire process — from start to finish — in immense detail. And unlike humans, machines don’t need to be taught individually in order to learn new things. “If we had a digital model of the world as we know it, we could share it among all machines,” says Kai.
Robots are vital to saving the economy
People are understandably wary of making machines that are better than humans, but we need innovation for society to grow. “This debate has been going on for centuries, probably ever since the first steam engine,” says Kai. “Everyone appreciates a raise in salary. And that money has to come from somewhere, usually from increases in productivity. The alternative is to make products more expensive, because manufacturers have to pay their workers.”
Factories are always competing with one another for clients and increasingly, customers want their suppliers to be flexible. As demand for different products continues to ebb and flow, manufactures need to increase their turnaround time. Flexible manufacturing requires one product to be altered in many different ways, but at the moment someone has to program all these different options. “Ideally, we’ll create smart machines that could manufacture electronics, rubber ducks — whatever you put in front of it,” Kai says. “But we’re still in the very early stages.”
Humans and robots, working together in harmony
Kai is passionate about creating automation that helps rather than hinders humans. “It’s all very well creating these robots, but if they don’t work with us they’re useless,” he says. “Because we don’t have the option of getting rid of hundreds of years of industrial infrastructure and starting all over again, it’s vital that robots are able to work with what we’ve already got. We have to work with existing factories, and slowly automate from there. It’s why robots tend to be based on humans; for example their arms have the same dimensions as us so they have the same reach.”
Designing robots to look like us comes with the risk that we accidentally end up treating them like humans. “On the flip side, one of the biggest dangers we face is assuming robots have the same capabilities as us,” says Kai. “Take robotic arms. We assume they move the same as ours but they can’t. They move up to 360 degrees, at incredible speeds, and are much stronger than we are. We need to take care of such assumptions whenever machines and people are working side-by-side.”
We’ve still got a long way to go before autonomous robots become common place, but in the meantime Kai is content with watching it pan out. “People are fascinated by automation,” he says. “I mean, I could watch it all day; machines doing things by themselves, figuring things out. It’s just amazing.”
At Siemens, Kai is a project manager in the Corporate Technology research facility based in Munich, Germany. Find out more about working at Siemens.
Photography: Here
Illustration: Peter Henderson
Icons: Atif Arshad; Arthur Shlain and Clockwise from the Noun Project
