Google's vision of self-driving cars is wrong, says MIT professor
December 11, 2015 8:54 AM   Subscribe

David Mindell is Professor of Aeronautics and Astronautics at MIT. He has a new book: Our Robots, Ourselves: Robotics and the Myths of Autonomy. He has a one hour interview with Erik Davis on Expanding Mind podcast where he talks about deep sea submarine robots, drone airplanes (remote piloted aircraft) and google cars. Short CNET article: Google's vision of self-driving cars is wrong, says MIT professor.
posted by bukvich (68 comments total) 17 users marked this as a favorite
 
Read the cnet article - I feel like it's divorced from reality. As serfs (I'm expecting this word to make a comeback) come to own less and less material goods, including cars, a greater and greater amount of driving will be on demand, like Uber and its competitors. What is the motivation for the owners of such services to keep a human driver in each of those machines, if Google and its competitors can provide autonomous electronics to perform the same services, at less trouble and cost? Some kind of complicated moral argument? The creeping sense of dismay felt by some academics and pundits?
posted by newdaddy at 9:08 AM on December 11, 2015 [3 favorites]


From the CNET article:
Wouldn't it be terrible if HAL turned out merely to be Shallow HAL?

Screw self-driving cars; we need self-writing tech journalism.
posted by Atom Eyes at 9:15 AM on December 11, 2015 [8 favorites]


I trust Google more than the elderly, at least when it comes to driving a car.
posted by Artw at 9:19 AM on December 11, 2015 [15 favorites]


Screw self-driving cars; we need self-writing tech journalism.

Thinkpeice Bot could probably do it better.
posted by Artw at 9:20 AM on December 11, 2015 [1 favorite]


If only that was true. Still, as far as Google is concerned, any human pleasures associated with driving should be eliminated for the greater good.

I only read the article, but is he mainly critiquing the idea that we need Google's centrally-controlled automation to save us from our incompetence, not so much the idea of self-driving cars itself? Is he advocating that we should have control over our self driving cars to the point where we can still drive them ourselves if we choose (v.s. the Google's vision of cars without steering wheels)? It's sort of hard to tell.

But if I read it right, I think I agree.
posted by cosmic.osmo at 9:20 AM on December 11, 2015


No, I believe Milady's making the case not that humanity will come to find autonomy immoral but that Google is full of it if it thinks that full autonomy is even possible. The real world is too weird and complicated for algorithms to handle without human guidance.

Though if he's pointing to airline pilots as the counter, I dunno that it's a great one. There was an interesting podcast I listened to a while back -- I think 99% invisible? -- talking about how the increasing reliance on autopilot has contributed to human pilots performing badly when called upon to step in during a crisis. You can also make the case --- cynically, but with merit --- that the real function of the pilot is to serve as a "moral crumple zone," an agent to whom responsibility can be attributed when things go wrong even when it's the machines doing 99% of the deciding.

There's a practical question, too --- even if you put in a manual override that would allow for human control to avoid accidents, I don't know that that's really possible past say, 20 miles an hour. the sheer time'd it take to be alerted to a problem, engage the controls and attempt to avoid is probably far greater than time-to-collision. It'd be a Potemkin steering wheel.
posted by Diablevert at 9:22 AM on December 11, 2015 [16 favorites]


Stay off of unmanned public transportation! Wake up, sheeple!
posted by belarius at 9:22 AM on December 11, 2015 [2 favorites]


There was an interesting podcast I listened to a while back -- I think 99% invisible? -- talking about how the increasing reliance on autopilot has contributed to human pilots performing badly when called upon to step in during a crisis.

Yep.

Episode 170: Children of the Magenta (Automation Paradox, pt. 1)

Episode 171: Johnnycab (Automation Paradox, pt. 2)
posted by cosmic.osmo at 9:24 AM on December 11, 2015 [3 favorites]


At less trouble and cost? That seems to be the big point of contention. How does one answer the points about past attempts at full automation with earlier attempts - airplane flight, underwater and space exploration vehicles?

The infrastructure required by Google to make their vision work is pretty increadible.

Further, we've already got technologies that can handle the problems autonomous cars fix, namely rail. For whatever reason, US culture rejects rail. And somehow the associated costs for autonomous car infrastructure will be pallitable?
posted by Strange_Robinson at 9:25 AM on December 11, 2015 [4 favorites]


Still, as far as Google is concerned, any human pleasures associated with driving should be eliminated for the greater good.

Play videogames like the rest of us.
posted by grobstein at 9:26 AM on December 11, 2015 [1 favorite]


There's a case to be made that assisted driving rather than full automation is not an ideal solution either. The FAA issued guidance on the use of autopilots in aircraft (PDF) a few couple of years ago. One of their main findings was: "Autoflight systems are useful tools for pilots and have improved safety and workload management, and thus enabled more precise operations. However, continuous use of autoflight systems could lead to degradation of the pilot's ability to quickly recover the aircraft from an undesired state."

In other words, an autopilot that suddenly beeps, then switches off leaving the pilots in the soup isn't a great thing either, especially if long use of the autopilot means that the human pilots' skills degrade.
posted by bonehead at 9:27 AM on December 11, 2015 [1 favorite]


Of course, redesigning cities to make mass transportation workable across the board is unthinkable....

The problem with self-driving cars is that they are another attempt to automate something (see also, online searching) that can be done better and more cheaply by training humans better. Of course, that doesn't produce products that can be sold, so it's not very attractive to business.
posted by GenjiandProust at 9:28 AM on December 11, 2015 [6 favorites]


Rail? You're joking, right? That's like saying Boeing 747s fulfill all the use cases for quadcopters that are small enough to be carried by hand. Why not say canals already handled all of the purposes autonomous cars might fulfill back in ancient Mesopotamia?
posted by XMLicious at 9:31 AM on December 11, 2015 [4 favorites]


Came here to post those 99PI episodes and saw that they are already here. If you want a good take on this - I highly recommend giving 40 minutes of your life over to those 2 episodes.
posted by sauril at 9:33 AM on December 11, 2015


Still, as far as Google is concerned, any human pleasures associated with driving should be eliminated for the greater good.

Humans have proven themselves to be unacceptably bad at driving to the tune of tens of thousands of deaths a year so, yeah? If you need to pilot a car for pleasure you're welcome to do so on a racetrack or other closed course.
posted by ghharr at 9:33 AM on December 11, 2015 [29 favorites]


The problem with self-driving cars is that they are another attempt to automate something that can be done better and more cheaply by training humans better.

We've had more than a century to get this right. Road accidents still kill in excess of 90 people a day in the US.

Humans are able to drive if they're alert, aware and well-trained. Humans, unfortunately, are not always capable (or interested) in one or all of those. We're foggy, distracted, short on sleep, drunk or high, or driving without really knowing what we're doing. Humans are not very good at being consistent.
posted by bonehead at 9:33 AM on December 11, 2015 [19 favorites]


The infrastructure required by Google to make their vision work is pretty increadible.

I'm confused by what you mean by this. We have roads. We have gas stations. We have parking garages. We have car manufacturing plants. What the Google cars needs that regular cars don't got is a bunch of cameras, sensors and software.


Further, we've already got technologies that can handle the problems autonomous cars fix, namely rail. For whatever reason, US culture rejects rail.

Population density has a lot more to do with the lack of rail than "culture". Outside of the Northeast corridor, there just aren't enough people living close enough together to make rail a viable solution for commuting. Bus rapid transit, maaaybe.
posted by Diablevert at 9:35 AM on December 11, 2015 [5 favorites]



At less trouble and cost? That seems to be the big point of contention. How does one answer the points about past attempts at full automation with earlier attempts - airplane flight, underwater and space exploration vehicles?


No this is a matter of patience only. Human abilities, to a first order of magnitude, are fixed, while the capabilities of machines keep improving at a faster rate. If you can't imagine a machine that drives better than the best human, you're just not looking at a long enough time horizon.
posted by newdaddy at 9:36 AM on December 11, 2015 [5 favorites]


Let's not forget the unspoken part of this. Airplane pilots are better trained and in general are in better physical shape than the majority of automobile pilots.

I for one welcome out new automated car overlords. I am tired of crappy drivers.
posted by twidget at 9:46 AM on December 11, 2015 [2 favorites]


> The infrastructure required by Google to make their vision work is pretty increadible.

What the Google cars needs that regular cars don't got is a bunch of cameras, sensors and software.

If I'm not mistaken, in the current incarnation they're working off of comparing all of their own sensor data with centralized, previously-made higher-resolution (and presumably indexed somehow?) scans of the area. Which is far from an incredible degree of infrastructure, it's simply the same kind of information that was necessary to construct and maintain a road in the first place. And there's no particular reason this approach is essential for autonomous cars in general as technology continues to improve.

Doing it without a road at all was worked out early in the last decade, before Google even got involved.
posted by XMLicious at 10:01 AM on December 11, 2015 [1 favorite]


Further, we've already got technologies that can handle the problems autonomous cars fix, namely rail. For whatever reason, US culture rejects rail. And somehow the associated costs for autonomous car infrastructure will be pallitable?

That's not true though. Automated cars/trucks are point A to point B transportation, both for freight and humans. Rail use always ads extra complication, work, and extra time to getting from A to B. Your always going to need a last-mile transportation.

And as mentioned, we have the most costly infrastructure installed already: roads. The "self-driving" part is mostly a cost paid by the car/fleet owner.
posted by mayonnaises at 10:03 AM on December 11, 2015 [4 favorites]


The question seems like: should we be aiming for a much, much better version of cruise control or a fully automated car? I don't see the problem in starting from the 'cruise control' end of things and building outward. For one thing, it allows you to think in terms of a more evolutionary/gradual approach rather than some grand rollout of robot cars, where iterations of the technology can be put into commercial vehicles before the full infrastructure is there.
posted by zipadee at 10:05 AM on December 11, 2015 [1 favorite]


As well as whatever sensors need to be installed in the roads, and cellular improvements that give more than a meager few nines reliability connectivity, and better maintained roads. I'm sure I'm missing other enhancements required outside of the car itself.
posted by Strange_Robinson at 10:06 AM on December 11, 2015


Can you link to anything that makes claims concerning 21st-century driverless car technology requiring sensors needing to be installed in roads? I've only ever come across things like that when I was a little kid reading about proposals written in the mid-1900s before I was born.
posted by XMLicious at 10:13 AM on December 11, 2015


OK, CNET article is horrible. The author is just like, woo! Hooray for one the nerds finally speaking up for the humans, amirite! It sounds like it was written by E! or something.

I don't suppose the podcast has a transcript?
posted by ignignokt at 10:15 AM on December 11, 2015 [3 favorites]


I have my doubts that fully automated cars are going to work in the wild soon, mainly based on a recent experience where I was trying to get somewhere and the route was blocked by a parade. Various attempts to get around it by going parallel just found more parade, and I finally ended up having to go back to the interstate. Automated cars are definitely smart enough to not run through the parade, obviously, but the sort of forward thinking of 'this blockage will last for hours' and 'this whole series of blockages indicate I have to make a giant loop'... I suspect there would be a frustrated passenger sitting for a long time.

But it's not a moral issue -- I do expect that autonomous features are going to become required safety measures in the same way belts and bags and safety glass became required. Because if you want use public roads, being unsafe because you like it don't fly. And machines already drive more safely that humans do, on average, because humans are sucky drivers.
posted by tavella at 10:17 AM on December 11, 2015 [1 favorite]


I suspect there would be a frustrated passenger sitting for a long time.

If a dude with a xbox controller in Virginia can bomb people in Pakistan, an OnStar operator can get your self-driving car around an unexpected parade.
posted by GuyZero at 10:21 AM on December 11, 2015 [1 favorite]


As I understand it, they'll rely heavily on other cars in the network. That's all the information that has been released as far as I know. Still, considering the state of the art for object detection, they'll need sensors for anything the on board map falls short on. Redundancy is everything.

If they manage this, my applause. I'm a skeptic whenever it comes to techno-utopian claims.
posted by Strange_Robinson at 10:44 AM on December 11, 2015


The latest self-driving clown car has the capability to have a remote operator take control. It's not in that particular article, but I think it's been mentioned somewhere in public.
posted by GuyZero at 10:51 AM on December 11, 2015 [1 favorite]


Well, yeah, but that's not fully autonomous. I think Google has it right (despite previous posts on the blue about people claiming they are going the wrong way.) If a car doesn't have an option for a driver to take over, and to get the full benefits of self-driving cars that needs to be true for at least some, it's going to have to work within a network. It's going to have to receive data from all around to know where it should go, not just pick up objects near it with its own sensors. And there's going to have to be someone human that can be called on to redirect the car as needed for those complicated situations that exceed the limits of the AI.

So there's really two branches; autopilot improvements to human-driven cars, and fleets of managed autonomous vehicles. They'll both share some technology needs, the ones that have to do with navigating safely on streets with their own local sensors, but one branch will have to deal with when to yield to a human -- if someone panicking tries to mis-steer the car in an accident, should the car let them when it knows that it is about to be moved into the path of another vehicle? And the other will require a certain amount of mass scale to be economically viable.
posted by tavella at 10:53 AM on December 11, 2015


As someone who needs to interact with distracted drivers on a daily basis while not in a car, your 'motoring pleasure' can fuck right off. Driving in urban environments is miserable and awful drivers make every other option miserable and dangerous too. I don't think anyone who was working on early versions cars who had a glimpse of the climate catastrophe levels of traffic we're seeing today would have continued with their work.
posted by mike_bling at 10:54 AM on December 11, 2015 [7 favorites]


Part of the point of self-driving cars is to solve mobility problems for disabled users - there's no point in having a steering wheel override for a blind user or an epileptic user or a user with limited dexterity, etc. So the notion that the driver always needs to assert control makes no sense in these situations. Even for cars with steering wheels for local override, once you have all the self-driving apparatus in place doing remote human operation is a pretty small incremental step.
posted by GuyZero at 11:01 AM on December 11, 2015 [3 favorites]


"The notion of ceding control of something as fundamental to life as driving to a big, opaque corporation -- people are not comfortable with that," Mindell said.
Oh come on. Are those of us who use public transport fundamentally dispowered?
posted by howfar at 11:15 AM on December 11, 2015 [5 favorites]


The problem with self-driving cars is that they are another attempt to automate something (see also, online searching) that can be done better and more cheaply by training humans better.

The idea of the perfectibility of people as opposed to tools reminds me a bit of this from Chesterton:
Mr. Shaw asks, not for a new kind of philosophy, but for a new kind of man. It is rather as if a nurse had tried a rather bitter food for some years on a baby, and on discovering that it was not suitable, should not throw away the food and ask for a new food, but throw the baby out of window, and ask for a new baby.
posted by howfar at 11:20 AM on December 11, 2015 [1 favorite]


Oh come on. Are those of us who use public transport fundamentally dispowered?

This is literally one of the axioms of American culture, yes.
posted by GuyZero at 11:23 AM on December 11, 2015 [5 favorites]


The engineering question of how to combine human and automated vehicle control is a tricky one. One smart person working on this is John Lee . He consults with Volvo (very conservative, automation only in controlled environments), and Ford (automation, assuming human "ready to take over" anytime), and compares them with Google (100% automatic philosophy), and Tesla (beta automation that hopefully won't kill you). Lots of classic human-automation-interaction problems from aviation are being reprised with much more varied humans and uncontrollable environments.
posted by anthill at 11:54 AM on December 11, 2015


To all who clicked on the cnet link hoping it was better than clickbait I'm sorry.

His strongest argument against the google car (he has many) is that there is an unsolved problem implicitly programmed in there: if push comes to shove do I suicide the operator or homicide an obstacle? And not only is it not yet solved it may never be solvable.

That is not news. This was news to me: the car owner's manuals and the airplane operator's manuals are now cleared through legal to protect the manufacturers from liability. This explains why so much of what is in there reads like slimy legal boilerplate--it is slimy legal boilerplate. The manuals are not actually for the user at all.

The interview is well worth listening to. This guy got into subsea robotics on the team of the guy who found the Titanic.
posted by bukvich at 11:55 AM on December 11, 2015


from a couple years ago...
U.S. Department of Transportation Announces Decision to Move Forward with Vehicle-to-Vehicle Communication Technology for Light Vehicles - "The U.S. Department of Transportation's (DOT) National Highway Traffic Safety Administration (NHTSA) announced today that it will begin taking steps to enable vehicle-to-vehicle (V2V) communication technology for light vehicles. This technology would improve safety by allowing vehicles to 'talk' to each other and ultimately avoid many crashes altogether by exchanging basic safety data, such as speed and position, ten times per second."

not sure what their security model is tho :P here's the safecar.gov/v2v/ site; the latest looks like they've starting testing!
posted by kliuless at 11:57 AM on December 11, 2015 [1 favorite]


I bet this MIT professor had a royal freakout as a kid when he realized that he wasn't really steering the Autopia cars in Disneyland.
posted by markkraft at 12:10 PM on December 11, 2015


My vision for a humane transportation future would be lots of intercity rail and abundant intra-city public transit, and then mostly small, relatively slow electric golf-cart type vehicles (self-driving or not) to solve the last-mile problem.
posted by sevenyearlurk at 12:21 PM on December 11, 2015


There's lots of people who feel safer taking a long road trip than flying, even though there are way more deaths per mile driven than per mile flown.

It's less scary to die because of a mistake you made, than from an error made by a complicated system you cannot control or maybe even understand. Because if it's your own damned fault, well you're dead, but at least your illusions of control and cosmic justice are intact.
posted by aubilenon at 12:24 PM on December 11, 2015 [4 favorites]


As someone who needs to interact with distracted drivers on a daily basis while not in a car, your 'motoring pleasure' can fuck right off.

I don't particularly think it's the people who take pleasure in motoring - which generally implies in itself a level of active attention - who represent the weight of the problem, but the rest of us who don't, at all, yet still need to get places.
posted by atoxyl at 12:25 PM on December 11, 2015 [1 favorite]


As I understand it, they'll rely heavily on other cars in the network. That's all the information that has been released as far as I know. Still, considering the state of the art for object detection, they'll need sensors for anything the on board map falls short on. Redundancy is everything.

If they manage this, my applause. I'm a skeptic whenever it comes to techno-utopian claims.


"all the information that has been released"? You mean besides all the video footage from the last half-decade of Google's cars and vehicles from many other companies driving on highways, in urban areas, and elsewhere in traffic on the normal unmodified roads that all the other cars are driving on?

Not bothering to inform yourself about something and then spreading half-assed guesses as misinformation doesn't make you a skeptic.
posted by XMLicious at 12:33 PM on December 11, 2015 [7 favorites]


Cool! The CNet article is, yes, terrible, but I'm looking forward to reading the book. Until then:

It's interesting that there's this sort of dangerous valley between no autopilot and complete autopilot. If the autopilot handles most but not all of the work, the human will feel like they can stop paying attention. Then if there's an emergency where they have to take over, they have to make a big, sloppy context-switch. That's why Google's total-automation strategy makes sense to me: take away the steering wheel and let people stop pretending to pay attention to the road. It's still unclear how possible this is, but I'm optimistic.

I also have to point out the many differences between self-driving car research and previous robotics work (deep-sea and space exploration, drone piloting, aircraft piloting), which it seems like Mindell is drawing on. For one, machine learning has only gotten big in the last few years, and that's at the heart of driverless cars' object recognition. Then there's data: I'm sure Google has tons and tons of real-world data from the road, about what kinds of objects come up, how people behave, etc., which would be just impossible for something like deep-sea exploration. And you can iterate easily: just take your car out on the local street and see what goes wrong.

But there's also the scale of all this. In the US, ~30,000 people die in car accidents every year. And I'd guess tens of millions drive to work in miserable traffic. A driverless car system that really worked would be very valuable, both socially and to the companies developing. (Of course it would also put hundreds of thousands of cab drivers etc. out of work, which is pretty bad socially, but is also the reason companies like Uber are working on this too). So it's worthwhile for companies to make big investments in this research, in a way that I think has never been the case before in robotics.
posted by gold-in-green at 12:43 PM on December 11, 2015 [4 favorites]


The networked cars thing is definitely being embraced by major manufacturers. The problem with it is all the 99 Toyota Corrolas out there that still run great. If all the cars on the road can talk to each other, you can do a lot of safety improving things. If 40% of the cars, sometimes, in some places, can talk to each other that gets a lot harder.

Of course, Google owns Waze, now, so maybe they don't need the cars themselves, just the phones. The mapping constraint might be more of an issue but they're already working with the big mapping cos to do detailed 3d of major cities.

Of course, even that isn't quite good enough, you get lots of random unpredictable changes in road conditions that no map can account for (downed branch, pothole, construction site). What they basically use is a kind of advanced radar to ping the surrounding environment. The sensors are in the cars, not the roads.

The soft autonomy stuff, the advanced cruise control, the networked communications for collision avoidance --- all that's definitely happening already. People already have less control over their cars then they think. You can buy a Merc or an Audi right now that will automatically slow down to avoid rear ending someone on the highway and which can park itself.

But the last mile to full, real automation --- that's the key. Soft automation is incremental --- fewer accidents, perhaps somewhat better traffic flow. Full automation is revolutionary, perhaps as revolutionary as the car itself. Because of you can get a robot taxi to your door in ten minutes whenever, you probably don't need to actually own a car. You don't need to store a car. Whole cities no longer need to own or store cars --- 90 percent of the time, cars are sitting around in parking lots. Even building in excess capacity for rush hours and stuff, an auto car world would mean the number of cars needed to meet transport needs would be cut at least in half, maybe 2/3rd, 3/4ths? Half the parking lots. Half the parking garages. Around 30% of land in urban areas is dedicated to parking, IIRC --- freeing up that real estate could have massive impact on house prices, density, all sorts of stuff. The advent of the car changed cities from something that looks like New York to something that looks like LA; an auto-car world could reverse that.

But that all depends on full automation. If a human being is required to be in the drivers seat in order to operate the vehicle, even if only as an emergency backup system/liability sheild, all that goes away. Because you can't have empty cars serving themselves around to where they're needed, which means it's still be worthwhile for a given individual to own a car for personal use.
posted by Diablevert at 12:46 PM on December 11, 2015 [2 favorites]


Mindell actually has a coherent argument in the article, its just that regular people aren't trained to read it. If you're a scientist and engineer and you know some Marxian theory, the CNET article lays out the implications plain as day.

I'm not saying he's right, but this is one of the cases where if you don't have the background, his perspective won't make sense and just sounds like a crazy/ineffectual professor spouting ivory tower gibberish. There's quite a bit of stuff that can be unpacked but the article just quotes a few lines from him - its not representative or developed for those not already in the know.

This is one of the key lines:
The notion of ceding control of something as fundamental to life as driving to a big, opaque corporation -- people are not comfortable with that," Mindell said.

Which isn't literally true, because there exist variation and conflict between comfort and tension; rather, it expresses and summarises the problem he is trying to identify as a socially conscious engineer - is Google's future legitimate and desirable for humanity? And if such a question is left unchallenged, that's a symptom of a real problem.
posted by polymodus at 12:46 PM on December 11, 2015 [1 favorite]


People already have less control over their cars then they think.

...and that's one of the reasons I'm happy to keep paying the yearly repair bills on my '92 SUV instead of getting something newer: the machine is actually mine, and actually does what I tell it to, with no uppity robots hidden inside waiting to stage a coup on a signal from their corporate overlords.
posted by Mars Saxman at 1:17 PM on December 11, 2015


is Google's future legitimate and desirable for humanity?

So for the moment these are research projects and not actual products you can buy, so you're no more ceding control to Google than you are to Ford or GM. We have already seen people's car's self-reporting systems essentially turning them in for traffic crimes.
posted by GuyZero at 1:24 PM on December 11, 2015


How does one answer the points about past attempts at full automation with earlier attempts - airplane flight, underwater and space exploration vehicles?

The moon trips were 1969-72. I would not recommend trying to build an autonomous car with 1960s technology. But an awful lot of modern space exploration is autonomous - Curiosity on Mars, for example. And modern aviation is hugely autonomous at this point.

As to other people's suggestions that cars should be easy for humans to manually control, that would undo a great deal of the safety that autonomous vehicles can provide, which is the reduction of unpredictable behavior nearby as well as things like mesh networks that will allow the cars to coordinate for safety and efficiency.
posted by Candleman at 1:27 PM on December 11, 2015


Doing it without a road at all was worked out early in the last decade, before Google even got involved.

XMLicious, TerraMax is one of my favorites from the DARPA Grand & Urban challenges. In the 2004 Grand Challenge, it was by far the largest and most imposing vehicle, but it had been programmed so cautiously that it only made it through 1.2 miles of the course before it began backing up because it saw a bush in its way, then saw another bush and backed up again, and so on for half a mile of backing up before they stopped it. It did OK in the 2007 Urban Challenge until it became confused and tried to plow through a building, necessitating a quick E-stop.

I have the same point of view as newdaddy: Robots will eventually be able to do anything we can do, and better, if we want them to and we spend the effort.

Re: parades, I googled "road closed parade" and ran the top image through Google's cloud vision API. It labeled it with [vehicle, road, truck, sport utility vehicle, signage] and recognized the text ["CLOSED", "DETOUR", "48"]. And that's at night.

It's helpful to remember that just about all the planes, trains, cars and spacecraft these days are at least partially autonomous. Even the "manual" mode on the Apollo Lunar Lander still had software involved in making decisions.
posted by jjwiseman at 1:31 PM on December 11, 2015 [1 favorite]


Ted Talk about how google self driving cars "see" the road.

This is from earlier this year. The presentation was really eye opening for me and made me realize that the Google way of approaching this is so much more sophisticated and very well researched and understood that most people understand. They are also pioneering some technologies that will eventually be critical in the development of augmented perception for humans, which will be very interesting to watch.

Many of the skeptics about self-driving cars remind me very strongly of one of the protagonists in Vernor Vinge's "Rainbows End". It was also an amazing insight into one of the main cultural (and yes, it is entirely cultural) motivations against automated anything. When something is essentially a black-box, and you have no foundation of any of the underlying principals or technologies that make that black-box work, the first human instinct is to fear, then hate that technology.

I fight with this everyday, doing a lot of IT related work. Invariably, because of many factors, people are afraid of technology not because it is inherently scary, but because there is some critical piece of information, way early on in the cognitive chain of understanding, that they are missing, and they do not know how to ask for the information (or even that there is something there to understand).

Also, self-driving cars will be a major boon for equality and equitability in society. I love public transportation. I used to live somewhere that required a 45 minute to an hour commute to work (one way). At the time, I loved the feeling of control while driving, and getting time to myself in the car before and after work to decompress and what have you. Now, I very much enjoy being able to live very close to where I work, so my commute by car is all of 10 minutes, maybe 15 with bad traffic. But what I would enjoy even more is being able to travel in the way I travel on public transportation (reading Metafilter, natch) or zoning out, or thinking about what I need to do for the day, but on my terms and on my own schedule. The freedom that a car gives a person is the ability to travel on their own schedule, not the schedule of a bus or train. The pressure of having to make sure you get out of the house at a certain time is such a massive frustration. If you drive yourself, you leave when you are ready, and are not reliant upon some larger system in which you cannot control. Self-driving cars gives all the freedom of public transit AND the freedom of travel autonomy. Once people realize this, it will likely gain much wider acceptance.
posted by daq at 1:33 PM on December 11, 2015 [7 favorites]


Self-driving cars may give you the illusion of freedom, but it's clear already that they will be end-to-end 0wned from the factory by the megacorps that build them, and they will therefore inevitably be used as mechanisms of social control. That's not the kind of freedom I care about.
posted by Mars Saxman at 1:56 PM on December 11, 2015 [3 favorites]


I wonder how suburbanites & citydwellers break down on pro/con self-driving cars, setting aside factors like age, income, etc.

Purely anecdotally, among my tiny group of acquaintances those who live in large cities are most strongly in favor of self-driving vehicles, while those who live in the suburbs or smaller towns are most strongly against. Interestingly (to me, anyway), the one rural guy I know is also strongly pro-, although his reason was amusing -- it's because he could get hammered at the bar and not have to wait before driving home. The bigger bar in his area has a half-dozen crappy overnight rooms specifically for the over-imbibing.

...which I actually find to be a pretty compelling argument in favor. I mean, heck, if I owned a reasonably-sized drinking establishment in rural Wisconsin or the like, I might buy a self-driving car just for the sake of offering rides to/from the bar for my customer base. Yeah, there's the one guy with a "cab" but I think I've seen him like twice in all the times I've visited so he doesn't seem like a reliable source of transport.
posted by aramaic at 1:57 PM on December 11, 2015 [2 favorites]


Is it just me, or does anyone else feel it is more reasonable to hold an ambivalent perspective in the face of a complex sociotechnological development?
posted by polymodus at 2:00 PM on December 11, 2015 [1 favorite]


Regardless of the philosophy of the debate, it seems to me that the reality is quite straightforward. Within 20 years, urban driving will be a task that is much more safely conducted by driverless cars than by distractable, irascible, exhaustible humans. Insurance premiums will reflect this. Driving your car manually will become prohibitively expensive.

I, for one, welcome our new automotive insurance overlords

Edit: It may already be safer, but within 20 years it will be acknowledged to be safer
posted by Jakey at 2:31 PM on December 11, 2015 [3 favorites]


Purely anecdotally, among my tiny group of acquaintances those who live in large cities are most strongly in favor of self-driving vehicles, while those who live in the suburbs or smaller towns are most strongly against. Interestingly (to me, anyway), the one rural guy I know is also strongly pro-, although his reason was amusing -- it's because he could get hammered at the bar and not have to wait before driving home.

I haven't done a survey of my own acquaintances but this seems like the opposite of what I'd expect. I would have thought that people who live in places where a car is absolutely necessary and you can really be screwed if yours breaks down, you become unable to drive, or you simply don't have one, would be most interested in their neighbor or relative or a car rental place being able to send a car as needed, or in being able to share a car amongst people who live in completely different places.

Where I live I'm technically not completely screwed if I lose my car or it breaks down, because if I walk for half an hour I can reach a nursing home that is visited exactly once per week by a bus from a nearby city, which goes to a supermarket where the return trip leaves from an hour later. So that lets me either get better-than-convenience-store food once per week, as much as I can carry by myself, or take a one-way trip into the city because from the supermarket I can reach other bus routes.
posted by XMLicious at 2:47 PM on December 11, 2015


My father had to give up his keys recently because his vision is too compromised. My mother lost all her peripheral and then some to glaucoma years ago. This guy is welcome to explain to my parents why a driverless car is too much of a restriction on their freedom. Maybe he can do that while he drives them home from the grocery store.
posted by Hildegarde at 2:59 PM on December 11, 2015 [6 favorites]


Yeah, I know there's an anti-corporate and anti-data-collection set of people out there, but watch the video of a blind man driving solo to run errands and tell me if he's lost any of his freedom. ok, he's not solo, there's a google person in the passenger seat
posted by GuyZero at 3:07 PM on December 11, 2015


Self-driving cars may give you the illusion of freedom, but it's clear already that they will be end-to-end 0wned from the factory by the megacorps that build them, and they will therefore inevitably be used as mechanisms of social control. That's not the kind of freedom I care about.

No, but it's the same sort of freedom that cars already give you. It's called consumer capitalism and somewhere back in the Cold War it was rebranded "democracy".
posted by howfar at 3:08 PM on December 11, 2015


We have already seen people's car's self-reporting systems essentially turning them in for traffic crimes.

Except that as the story says, the woman was the one who authorized it -- it's an option on recent Fords to have the car automatically call 911 on your cell phone if your airbag deploys, but you have to set it up. So it just did as she had instructed after she smacked into someone and then tried to drive away. I'm not really finding that a real horror of automation.
posted by tavella at 3:57 PM on December 11, 2015 [3 favorites]


I have my doubts that fully automated cars are going to work in the wild soon, mainly based on a recent experience where I was trying to get somewhere and the route was blocked by a parade. Various attempts to get around it by going parallel just found more parade, and I finally ended up having to go back to the interstate. Automated cars are definitely smart enough to not run through the parade, obviously, but the sort of forward thinking of 'this blockage will last for hours' and 'this whole series of blockages indicate I have to make a giant loop'... I suspect there would be a frustrated passenger sitting for a long time.

This might be less of a problem than it seems. In addition to driver override where the person in the car takes over control, there are (at least) two other possibilities for human intervention in an automated car-driving system: human-curated road data broadcast over the network, and remote human control. The first could be used to send up to the minute reports of road blockages such as parades, traffic jams, roadworks, accidents and so on to cars in the system, as well as do stuff like clear the road (or a single lane) when emergency vehicles need to go through. All it would take would be for the location of the parade to be broadcast to all the cars and they would route themselves around it. Some of this kind of data is already sent to ordinary vehicle navigation apps already. The possibility for remote human operation could also allow for specialist professional drivers to take over control of self-driving cars in the event of an accident or other extreme event. Imagine a cadre of traffic officers whose job it is to take remote control of all cars within a 200m radius whenever a traffic accident (or near accident) is detected - they would be dealing with extreme driving situations every day and wouldn't suffer from the normal skill atrophy of drivers who are using autodrive 99% of the time. Perhaps the law will even be changed to give the police the legal right to remotely commandeer all self-driving vehicles at any time in the interests of public safety, with officers immune to all liabilities.
posted by L.P. Hatecraft at 4:39 PM on December 11, 2015


I used to drive to the local interurban, then take a train to work... did that for 15 years. I never really liked driving into the city, as the other drivers tended to make me angry. I had a 45 minute block of time to play computer, work with photos, or socialize with friends... each way, every day.

I now drive every day to a job that isn't amenable to the train. I've gotten (mostly) over the rage at other drivers... but I've taken to calling the commute by a far more accurate name...

"Pay attention or Die", which I play for an hour each way, each work day. It sucks... I for one would welcome the chance to just sleep while Google does the driving even if it's only 30 mph.
posted by MikeWarot at 5:26 PM on December 11, 2015 [3 favorites]


No, but it's the same sort of freedom that cars already give you. It's called consumer capitalism and somewhere back in the Cold War it was rebranded "democracy".


But that argument just admits these are newfangled technophilisitic objects that don't change anything in sum. It's almost like a conservation law.

So either argue that it does improve things globally, or argue that nothing changes.

And that's the problem, if you try to articulate something more than just heartstring-anecdata as to why people need more and more advanced tech. There's a burden of proof since specific technology is not neutral, so the question is how is it biased, what are the implications, how is it being framed, etc.
posted by polymodus at 5:46 PM on December 11, 2015


My father had to give up his keys recently because his vision is too compromised. My mother lost all her peripheral and then some to glaucoma years ago

The average person lives about 10 years longer than he/she is able to drive. So, we're all going there.

Unfortunately, most of us don't like to spend much time thinking about that version of the future us, so solutions that would help everybody out during that period of their lives--and a lot more people besides, like people with disabilities--usually don't get much shrift.
posted by flug at 6:07 PM on December 11, 2015 [1 favorite]


Just as something different, Erik Davis (whom I slightly know) also wrote the libretto to the Burning Man Opera, and appeared in the production (that's him in the bunny suit. He was really good.)
posted by emmet at 7:29 AM on December 12, 2015


Self-driving cars are all well and good, but they don't get around the constraint that 1) cars are the least space-efficient form of transport ever invented and 2) spaces designed for car travel are shitty for people walking, because wide roads and lack of demand push density down. Self-driving cars or not, a three lane freeway is going to carry just 5,000 people per hour per direction. That same person-moving density could be done with a half-lane wide bike path.

Right now the only thing stopping baby boomers like my parents from continuing to use their societal influence to push for (or passively condone) car-dependent sprawl is the realization that one day they might want to be able to walk (or roll) to visit their neighbors.

As an urbanist, the potential benefits of self-driving cars (less danger and less parking demand) aren't clearly a net benefit if self-driving cars increase the demand for driving (kneecapping demand for dense, walkable urban form and public transit).
posted by anthill at 8:02 AM on December 12, 2015 [3 favorites]


There are still people who have to work and live out their lives in the vast expanses that those dense, walkable urban places have to extract resources from and evacuate waste into to keep going, though.

It's well and good if city-dwellers can walk or take public transit to everywhere they need to go, and I'm all for cities consolidating themselves into those high-density walkable regions and preventing the pseudopods of low-density sprawl creeping out into the countryside around them. But until we have the technology to build giant closed-loop hermetically-sealed archologies every last person can't live in cities and the rest of us need to be able to get around and get to doctors' appointments when we're elderly or injured or disabled, too.

One thing that always occurs to me trying to envision the future is the way that in Europe streets and roads are often much narrower, and European drivers will often put up with an incredible frequency of annoying stopping and turn-taking thanks to two-way traffic passing through essentially one-way chokepoints absolutely everywhere.

Self-driving cars won't get annoyed with that sort of thing, so I think there will come a time in North America where municipalities large and small will be tripping over themselves to sell off strips of their streets and roads because all the people who no longer drive won't care too much or even notice if all the roads get narrower. So it seems to me that right now is the time to start putting laws in place that will favor converting that kind of municipality-owned real estate for pedestrian or cycling use, over selling it into private hands for quick cash.
posted by XMLicious at 11:32 AM on December 12, 2015 [2 favorites]


I have come up with an abstraction that roughly approximates the Mindell objection.

Replace car with $Weapon_of_Moderate_Destruction

Replace google with $Multi_Billion_Multinational_Corporation

Ask yourself: do I want M_B_M_C deploying millions of M_o_M_D?

Holds for any values of M & M. He does not object to robot cars; he objects to google making and selling us robot cars. I bet he would say the exact same thing about Amazon drone freight delivery. They need to be supervised closely by the Feds just as if they were building antitank missiles; and they will be if they screw up as you might predict they will when profits mean everything and safety is one design feature.
posted by bukvich at 12:15 PM on December 12, 2015


« Older I'm Not Sleeping Because I'm Happy   |   Recite 'The Fatiha,' Laugh Your Muslim Ass Off Newer »


This thread has been archived and is closed to new comments