“If you want, we can debate whether a human being is perfectly rational or not,” says Adam Bahlke. Adam is a Chief Technology Officer and co-founder at Motor AI, Germany’s brightest hope for self-driving cars. He’s in the car with his co-founder Roy Uhlmann and they tell me – swearing their choice of meeting room isn’t a car company marketing gimmick – that they’re taking the call while they’re driving to their next meeting.
The limits of human rationality might seem like a question that’s more appropriate for a philosophy seminar than an interview about self-driving cars. However, it’s this very question which Motor AI’s new system pivots on. It’s this same system that means that they’re closer to getting the required certification to put their self-driving cars on the road than any of their EU competitors.
When AI companies want to create decision making systems, Adam explains, they usually copy the part of the human brain that carries out pattern recognition: “So, the classic example is, you show a child three pictures. Two of them are cats and one of them isn’t and you say to the child, which picture is different?”
He explains this was because initially, a lot of the companies were trying to harness this aspect of the brain for image recognition for e-commerce.
“Those AIs were never actually designed to carry out decision making. As a human being, you don’t make decisions by considering every possible option — if you’re trying to turn right at an intersection, you don’t go through every right turn that you’ve ever made in your life to then figure out what to do in this particular decision.” Instead, he argues that a human brain generally tries to connect different pieces of information in a logical way so they can make a decision, even if they haven’t encountered that particular situation before.
This, he argues, is why Motor AI’s competitors are stumbling at the first hurdle, and the most essential one: certification. Don’t fall asleep! Certification is essential and sexy and great because certification is the difference between a self-driving car being an idea on someone’s whiteboard and you climbing into a self-driving car and spending your journey catching up on Succession and emails instead of inching through traffic at rush hour yourself.
But for self-driving software companies in Europe to get certified, one of the various independent entities responsible for inspecting your vehicle (in Germany, the most obvious are TÜV or DEKRA) and the national government of your country and the European Union have to be sure that your system is safe enough to be used on a public road.
With a system based on pattern recognition and deep learning, AI has driven billions of miles, but there’s no logic behind it. It hasn’t reached the Holy Grail of autonomous driving: a deterministic system, where you always have a reliable outcome. It’s thanks to this, Roy argues, that none of the companies who use this system of AI will be able to get certification. Only Motor AI will — and possibly Bosch, he says. There’s rumors they’re working on a similar system of self-driving AI as Motor AI are.
Adam also argues that this certification issue is the reason why the US companies are lagging behind. The US’s biggest players in the self-driving car field release their own reports saying that their self-driving software is safe. “But why should we trust them? I mean, look at Tesla,” Adam says, referring to the incident this year in which a Tesla in autopilot mode crashed into a truck in Taiwan. He implies the American approach isn’t terribly careful. “They see the whole thing like a software, like Windows. ‘There’s a problem? Then we’ll just do an update.’”
Since the US doesn’t have an equivalent organisation to Germany’s TÜV — an independent third body organization which objectively evaluates the AI’s safety — these companies will not be able to sell their software in Europe. This is a big deal. Roy notes that while China leads the world as a market for self-driving cars, in second place is Europe. “Logistics-wise, Europe is just a much bigger market than the US for transport.”
Is there any way that they could imagine America catching up? I ask. It’s hard to imagine one of the global leaders in tech innovation – in making shedloads of money – missing out on something as big as this.
“I’d never say never. But at the moment, what you’re seeing in the American market is a lot of consolidation: Argo are “VW and Ford’s lovechild”), Amazon bought Zoox, Google already had Waymo. I don’t think any of the established companies are necessarily going to change their ways very quickly because they’re focused on the US market and they need to show their investors results there.”
But by the time the US gets certified, they may well have a lot of catching up to do. When I ask the pair where they think self-driving cars are headed, Adam talks about the possibility of it not just changing transport but transforming the retail sector — instead of you visiting a pharmacy, maybe the pharmacy visits you, it tours your city, it brings your prescriptions to you.
“If you think about what is possible, there’s a lot that is — there are so many positive things could be brought into the world that are currently too difficult that self-driving technology could solve.”