On harm reduction
On harm reduction
There's a key insight missing from autonomous cars, and I can finally talk to you straight about it
I spent the last seven years of my life trying to convince the autonomous vehicle industry of two things: first, that driving is fundamentally a social process; and second, that if they ever wanted their vehicles to work as advertised they needed to buy the solution offered by Perceptive Automata, a company I co-founded to commercialize research from my doctoral thesis. I failed on both counts: the industry does not yet understand the issues it faces and the harm that could cause, and Perceptive Automata no longer exists as a company.
Nevertheless, driving is fundamentally a social process, whether autonomous vehicle companies understand that or not. They underestimate the centrality of that insight to the problem they’re trying to solve at their own peril, not to mention the peril of the drivers, pedestrians, and cyclists who interact with their products. I hope that I can convince you, my reader, how important it is to understand driving as a social process. I hope, too, to convince you that autonomous vehicles that are not designed around that basic fact will never work correctly or safely.
While we're at it, I have a bunch of other things to say. Things that I couldn't say when I was selling something. If I was trying to sell software to, say, Tesla, I could hardly go around saying that Tesla should be criminally liable for unleashing wildly premature and nonfunctional safety-critical driving software on the world. If I was trying to sell software to, say, Ford, I couldn't say that the design of their best-selling vehicle increasingly makes drivers of those vehicles act like murderous sociopaths. It's freeing not to be selling anything. I can say what I think.
I'm writing this, in part, because I know a lot about driving (autonomous and otherwise), about the behavior of pedestrians, about how our brains work, about the mechanisms and limits of machine learning, and about how all those things interact. Contemplating and explaining the subtle interactions between all of the above was one of my favorite parts of my job at Perceptive Automata. I don't want to stop doing it, and I think sharing what I know, while continuing to learn more about this field, can be useful to others.
I'm also doing this because it's important for people who really understand the autonomous vehicle world to speak up. The industry is shrouded in secrecy. This is for many reasons, some good and some bad. But the real story of how well these vehicles work, and how and why they fail, hasn't been told. What will happen when these vehicles are deployed widely? How will the world change? Will these vehicles be drivers of social good or social harm? Industry insiders tend not to give their real answers publicly.
The public conversation around autonomy (both autonomous vehicles and the broader landscape of autonomous robots out in the world) is dominated on one side by boosters of the technology—or of automobile transportation more generally—and on the other side by an ad hoc and outnumbered coalition of urbanists, bicycle activists, and iconoclastic Cassandras. This coalition is wildly skeptical of autonomous vehicles, not least because they seem like the precise opposite of a solution to the real mobility problems facing us: a failed transportation system that relies on too many cars driving too many miles.
The skeptical coalition is not wrong.
The coalition isn’t wrong about the problem: we need to figure out ways to vastly reduce our dependence on cars, especially in cities. We need many, many more people to walk, ride ebikes, take trains. We need to move away from the tyranny of the automobile. It's an existential question for our nation and our planet.
The coalition also isn't wrong to be skeptical of autonomous cars: the social and environmental implications of this technology could be hugely harmful. The deployment of autonomous cars could replicate all of the policy failures that accompanied the automotive boom and amplify them tenfold. It's a scary, overwhelming prospect, and I think it will take a lot of active effort to avoid it.
That's another thing I couldn't really say when I was selling something.
But because the skeptical coalition doesn't understand autonomous cars from the inside, their rhetoric and activism tends to be a blunt instrument. If banning autonomous cars—or autonomous mobile robotics around people, really, since that's what we’re actually talking about when we talk about “autonomous cars”—were on the table as a serious policy idea, I don't think I'd oppose it. That idea isn't on the table, and it won't be, in any world I can imagine. And autonomous mobile robots really can make for a better world, sometimes, if deployed carefully and thoughtfully. The anti-Roomba lobby has been notably quiet, for instance. Autonomous technologies in some form are going to increasingly be part of our lives. So how do we help that reality not become a disaster?
I used to think of what I was doing at Perceptive Automata as harm reduction for autonomous cars. Harm reduction, in the substance addiction world, refers to an approach to managing the harms of drugs that accepts that you can't stop people from using harmful drugs entirely. You offer treatment if it’s wanted, and options for living life a different way, you make harmful or lethal uses of drugs less appealing, and you do everything you can to keep people who do use those drugs from hurting themselves or others. Harm reduction is a metaphor with surprising applicability to cars, especially cars in cities.
SoI guess that's my mission statement for the collection of writing I'll be putting here. I am sure I'll wander off topic and talk about other things: brains and machine learning and behaviorism and startups and who knows what. I had a lot of lives before starting an autonomous vehicle company – lousy bike messenger, mildly notorious troublemaking hacktivist, old-school political blogger, face perception researcher, house DJ – and my interests are correspondingly broad. But central to the work I've left unfinished is the question of how to think about autonomous technologies correctly, and hence to understand how they really ought to fit into the world. That’s what I hope to explore here.