Smart Infrastructure for Smart Cars: A Conversation with Dr. Stilgoe on the Hidden Dependencies Shaping Automated Vehicles

Automated vehicles navigate the road using built in sensors, and can be augmented with digital connectivity allowing the vehicle to communicate with nearby vehicles, with surrounding road infrastructure like traffic signs, and with the cloud-based networks providing real time driving updates, warning of changing weather or traffic conditions. 

These technologies have the potential to create safer driving conditions and decrease crashes, allow new populations to access transportation through car sharing, bring mobility to disabled populations and the elderly, and support environmental gains including through increased fuel efficiency. Policies on the digital infrastructure enabling their connectivity, like 5G networks, as well as policies on the physical infrastructure, like road design, lane markings, and traffic signs, shape how the vehicles are developed, where they will be deployed, and who will be able to access their benefits.

To cut through the hype around automated vehicles and discuss the technology’s digital and physical challenges, we spoke with Dr. Stilgoe, a Professor at University College London in Science and Technology Studies and lead of Driverless Futures. Dr. Stilgoe has engaged public dialogue on connected and automated vehicles with the British government, and his new book, Who’s Driving Innovation?, examines collaborative governance for automated vehicles.

You’ve previously written that “autonomous vehicles” are not really that “autonomous.” Can you explain what you mean by that?

Historians like David Mindell have written about how all so-called autonomous systems are conditional – they depend on particular things in the real world in order for them to function. The real world cannot be ignored and is sort of swept to the sidelines because of this insistence that a technology is possible that can come along and solve problems by simply replacing an aspect of human labour. That narrative of autonomy is quite a powerful one and it is particularly within self-driving cars where the problem frame is so often the one of human error. Humans are terrible drivers, they get drunk, distracted, crash and kill each other all the time. Therefore, if we can replace humans with autonomous systems, we can solve that problem….

That’s the technical critique of autonomy. There’s also a political critique of autonomy and machines which comes from philosophers like Langdon Winner, for example, who said technology is itself not autonomous: it has a direction, it has purposes, it is built with values inside it, therefore to talk about technology having an autonomous driving force is to depoliticize it and make it impossible for consideration through democratic debate. If we want to have a democratic debate about new technologies, we have to reject that idea of technological autonomy as well.

Speaking of the entanglements with the real world, what does the infrastructure for automated vehicles need to look like?

At the moment, we have some of the market leaders in self-driving cars saying that we can do what we need to do with existing infrastructure. They say the infrastructure doesn’t need to change at all. I don’t buy that. I don’t buy that as a short-term consideration and I certainly don’t buy that as a long-term consideration.

You do have some developers who are admitting that it would be convenient if bits of the physical infrastructure were upgraded to make life easier for self-driving cars. For example, the addition of a smart traffic light, that just transmits its status to a vehicle....There are some developers that are saying having pedestrians and self-driving cars mixing on the open road is too hard, so it would be convenient if pedestrians just behaved a bit more predictably. Even though pedestrians aren’t part of the infrastructure, they’re part of the world surrounding the self-driving car - the physical, material world.

Then you have aspects of digital connectivity. Self-driving car developers often talk about it in terms of Vehicle to Vehicle communication, V2V, and then Vehicle to other things, V2X, or vehicle to infrastructure communication. [Developers] say it would be so much easier if these things could talk to each other. 

Do you see approaches varying by country?

If you compare the approaches taken by governments and developers in the US to those taken by people in Singapore, or China there are very different emphases placed on infrastructure. Policy makers in Singapore are much more ready to say we’re going to design a new place, a new neighborhood, and we’re going to design it around AVs -- we’re going to design an infrastructure that suits this technology, even though we don’t know what this technology is going to look like. In China, new cities are being designed with AVs as part of the brief. In the US, if you talk to developers, a lot of them have given up on infrastructure. When it comes to digital infrastructure, I think that’s where you’ll see some really interesting battles. Around 5G and connectivity, I think you will see China and the US take very different approaches. I think European countries will mix and match a bit of those. You see very different types of technologies, so to even talk about an automated vehicle is to talk about [different things]. You have the Waymo idea, where you take a normal car, take out the driver, and put in a computer. You have, in some places like the Heathrow airport in London, these low speed shuttles trundling along in very tightly defined design domains. And you have, in China, state-owned enterprises working with cities on developing infrastructure at the same time as developing the artificial intelligence that goes in the car.

It seems there is a combination of both new and old infrastructure that must interact in order to allow for AVs. What are some of the risks and benefits of laying a new transportation system on top of older network infrastructures?

I would be more considered about the generic aspects of any technological infrastructure, which are things like the risk of technological lock in that then makes reversibility really hard. People like me, when we think about technological lock in, we might look at examples like the qwerty keyboard: the classic case in technological lock in, a sub optimal technology we’re currently hooked on and it’s impossible to develop any alternative. We can think about car-based infrastructures in the same way. If you live in Phoenix, Arizona, you are in a car dependent society and it is really hard to imagine an alternative to car dependence there. The danger would be that as soon as we start imagining infrastructure around one set of technologies, it locks out other technologies and it reduces our ability to adapt. For example, we can’t have mixed economies of public and private transport. We may well be designing streets to suit AVs, but in doing so we may be making it harder for cyclists, pedestrians and other people. 

We see some international cross-border trials popping up in specific places, coming with their own unique set of challenges for coordination. What role do standards play in facilitating this international cooperation?

I think everybody – all the developers - are starting to recognize that standards are going to be absolutely vital. In the world of cars, there are so many standards. It is highly standardized, which means in almost all cases, you can sell a car in any country in the world. They are developed to global standards and the global standards have made that form of trade and global development work.

I think one really interesting question, is whether self-driving car systems are going to be hyper-localized in the way that some public transport systems are, or whether they are going to be developed according to one internationally agreed upon standard, and then rolled out in different places. It may well be that the standards setting means that all the companies come together and agree upon a set of standards….. If that does happen, what you’re likely to see is the standardization of the world in the same way as roads in most places look pretty like other roads. There are some points of disagreement, but it is possible to drive in one place even if you learned to drive in a completely different car from another place. It would be really interesting to see if that happens when it comes to self-driving cars, or if it goes in the other direction and places develop their own bespoke systems and the technology looks very different in different places.

How might this technology and its dependencies interact with existing societal inequalities? Essentially, who benefits and who might not?

The guilty secret about technological innovation is that it tends to widen inequalities rather than close them. And that’s not at all the story we like to tell ourselves about innovation. The question is, when it comes to a new technology, how do we make sure that the benefits don’t just go to the same people who have always benefited from technological change? How do we make sure that the technology benefits the people who it is claimed will benefit from the technology?

In the case of self-driving cars, there are safety benefits. So, we look at who is at most at risk from current modes of transport and how might we alleviate those risks. We might think about disabled groups, people who for some reason can’t drive, and how might we target those needs in innovation. We might think about people who don’t currently have access to many modes of transport, people in ‘transport deserts,’ and we might seek to target those needs.

But the story of autonomy doesn’t really do that: the story in which what you’re simply doing is taking out the driver and putting in a computer, you’re not considering particular groups of beneficiaries. You’re just hoping for a trickle-down model of innovation, where we hope the richest people will be early adopters and eventually the benefits of the technology will reach other people.

A more deliberate approach to technology policy would be to say, no let’s actually target those needs. If we started with the requirements of those groups, how might we develop technology differently? This might look like, for example, targeting particular places, designing particular vehicles, thinking about the ownership of the technology. Is it owned by private individuals or is it a public transport system? And that might change the design of the system.

It is easy to imagine a future where gaps in digital networks prevent populations from accessing automated vehicles, where the automotive industry must build specialized parts to meet regionalized technology requirements, and where automated driving across international borders is hindered by lack of technological interoperability. By asking questions now on the direction of autonomous vehicles, such challenges can be identified and can provide opportunities for strategic collaboration. Engaging in public conversation, participating in international standard settings bodies, and leveraging international agreements to coordinate infrastructure like 5G and facilitate data sharing, can help realize and distribute the benefits of autonomous vehicles.