Flipping back over to UpWing for a little while, throwing down some notes while it’s still fresh in my mind from recent conversations.

A lot of these notes are based on your casual tech blog’s impression of the future and so they may be a bit more trite than DownWing thought seems to be. That’s the feeling I get talking the same accelerationist talk that I hear everywhere anyway, when there’s a dark exciting world of dour DownWing cynicism to be had out there. Maybe it’s the novelty that provides the exhilaration. I don’t know.


But today, notes on the conventional understanding of autonomous cars. Much of my own text is edited from my own posts on Facebook in conversation.

EDIT: I was mulling through, “what about talking about contemporary technology is necessarily UpWing”? Nothing, really, I’m just being sloppy. I meant to go further into a bizarre, expensive future that futurists argue these cars enable, but I only wrote about the state and immediate expectations of the industry. That’s what happens when you don’t respect your own deadlines (and are too stubborn to change your deadlines, evidently). More on that tomorrow.

The Audacity of Gradualism

In class, I remember being told a story by Red Whitaker about the DARPA Grand Challenge: The challenge was to construct autonomous vehicles that could complete a 150 mile course through roads and pathways in the Nevada desert. That first year, in 2004 more than 100 teams signed up for the challenge. Twenty-one teams would complete enough of the prelimary test to qualify.

In the first DARPA Grand Challenge (2004), not a single competitor got even close to finishing the course. The farthest distance was my Alma mater, and we had only traveled 7 of the 150 miles before getting stuck.

The next year, of the twenty-three finalists who started, 22 of them all beat the previous year’s best. Five cars completed the course (Stanford first, and Carnegie Mellon second and third- Gray Insurance company and Oshkosh Truck company after that).

By 2007, DARPA released the Urban Challenge, a 60 mile (under-6 hour) track through traffic and obstacles. Eleven teams were allowed to qualify (and Carnegie Mellon would win that too, followed by five other completionists)

The point Red was making had to do with the rapid speed by which the field was developing: from the embarrassing failures of the 2004 Grand Challenge it would’ve been hard to take seriously the complexity of the challenges that they’d be able to cover in just three years. That being said, there was no big revolution. It was a long grind of development, the ability to learn from many people’s failures, and admittedly some very clearly defined goals that allowed for such a drastic-seeming jump in effectiveness (quick, effective feedback loops). And there are other elements at play- falling costs of hardware, for example.

In all, it’s a good story. Great fodder for the accelerationist.


Useful Vocabulary

Some useful vocabulary taken from a Human Rights Watch report on Autonomous Weapons. Great for my inevitable drones post:

Human-in-the-Loop Weapons: Robots that can select targets and deliver force only with a human command;
Human-on-the-Loop Weapons: Robots that can select targets and deliver force under the oversight of a human operator who can override the robots’ actions; and
Human-outof-the-Loop Weapons: Robots that are capable of selecting targets and delivering force without any human input or interaction.

“On-the-Loop” is the space I’m most worried by when it comes to continuous, immediate tasks like driving. When machines automate most of a task, human attention will become less reliable. I expect 98% of all driving to be automated easily in the near future, but the final stretch- edge cases, odd judgment calls, etc- will probably be a long trudge to reduce, and we may end up requiring policy or outside systems to try to mitigate that dark 2% (making up a number, it could be smaller) of driving time where the system is not reliable.

And the legal issues involved are obviously still really hairy.


The Evolution of the Driverless Car

Driverless cars are coming faster than most observers would have thought.  One big reason, according to Bryant Walker Smith in a recent article in Slate, is that people predicting the driverless car future assumed that they would have to be part of centrally-run systems, with corresponding changes to physical infrastructure, such as special roads embedded with magnets.  Or for that matter, we can add, centralized computers to take control of all the vehicles in the system.  The changeover has to be centralized and take place for a given area all at once; it doesn’t scale incrementally.  That was the thought, anyway, and Smith (who is a fellow at Stanford’s Center for the Internet and Society) says that as a consequence, ever “since the 1930s, self-driving cars have been just 20 years away.”

The Volokh Conspiracy

And so now it’s no longer even really a blue-sky idea that the Autonomous Car we’re all been waiting for is being constructed out of the systems (both technical and otherwise) that already exist. Functions of driving are slowly becoming autonomous, which allows for gradual technical development and also allows for drivers to become more comfortable with future reduced autonomy, in a frog-in-boiling-water type scenario that one should expect over a violent, sudden revolution.  As the New York Times reports on this paradigm: “On the road to autonomous, a pause at extrasensory.” This framing makes sense anyway, since in this case it seems that car buyers are not sure this is the future they want. I’d love to watch changes in belief there over the next decade.

I have to imagine that the main reason for “sluggishness” here is risk-reduction. And I don’t think we’ve been all that sluggish at all.

The actual play-by-play isn’t as interesting to me, but to get a sense of the breadth of this trend (and in no particular order):


Life with Autonomous Vehicles

The most apparent benefits to autonomous cars:

How exactly the autonomous car might impact society is a more open, (more-likely-to-be-incorrect) question.

Are we going to step away from vehicle ownership altogether? In the short term, it seems that Peak Car may have already occurred in America (or perhaps that’s just demographics + recession creating a short-term reduction). Perhaps families will own one (instead of 2+) vehicles that can transport everyone, deliver food and goods, and spend off time as a “shared car”, a taxi that earns money back to the car owner? That sounds like a plausible next-step beyond the drive-assist vehicles.

How would attention be spent in/around the driverless experience? (I don’t know.)

Sitting quietly in the back like a taxi passenger seems like a waste. I wonder how the user’s experience of the driverless car will change, and how priorities will change for manufacturers in terms of what aspects of the experience will be invested in beyond risk-reduction.

Perhaps the car will be event-focused (instead of just location-focused). So you’d tell it that you have work today, it knows that work is an 8 hour event at so-and-so location. It knows you like to eat right after work and suggests food en-route, or maybe it finds more ethereal events (sales, concerts, an impromptu congregation of friends) to consider. Driving is a quiet time for creating/scheduling/preparing for events. The user’s sense of location is not relevant (unless they want it to be). Just as planes create flyover regions between cities, maybe the autonomous vehicle completely erases people’s necessity to pay attention to location.

More likely, the car is more functional, with a capacity to project information from machines on the driver’s person (embedded/wearable computers or just tablets/phones), acting as a small impromptu entertainment center.