Robot Cars
| Tue Oct. 14, 2008 9:14 AM PDT
ROBOT CARS....Matt Yglesias, riffing off a Tim Lee piece, says that self-driving cars could free up lots of parking spaces. Which is true, I guess, but seems sort of like saying that cold fusion would be great because it would allow us to build smaller cooling towers. If we ever do build a genuinely self-driving car, it means we're only a stone's throw away from nearly human-level artificial intelligence. More efficient parking will be the least of our worries at that point.
Continues Below
Continued From Above
Comments
If we develop technologies advanced enough to make self-driving cars, and yet we're still riding around in cars, we are a stupid, stupid species of animal that deserves to go extinct. The shame will be the damage we're doing on our way out.
If we ever do build a genuinely self-driving car, it means we're only a stone's throw away from nearly human-level artificial intelligence.
I don't think so. Navigating between two points on a relatively smooth surface with a high level of standardization and flow control, without smashing into anything, isn't really a very tough problem, compared to, say, behaving at a human level in verbal or musical or emotional terms.
Anon is correct. Self-driving cars would not require artificial intelligence, provided you got all the human drivers off the roads. Self-driving cars would only have to be intelligent when interacting on the road with other human drivers. If the whole system was computerized, it wouldn't have to be intelligent, it would merely have to obey a simple set of rules.
If and when we get self-driving cars, we may as well all just ride buses. If I just wanted to sit and be a passive participant in the driving process, that's what I'd get. The point these people forget is that a lot of us actually ENJOY driving.
...it means we're only a stone's throw away from nearly human-level artificial intelligence
I dunno, that would be say that remote conrol toyes have that kind of capacity already. We're
over-thinking the technology here as I believe there are ways to simplify some types of technology if we stop to really think how we might achieve either a way to consolidate space for parking or cars that return home - without the high cost of gasoline that we all are currently paying.
...it means we're only a stone's throw away from nearly human-level artificial intelligence.
What Anon said. Kevin, I love your blog, but you'd best stick to politics. Fortunately, you're aware of your limitations in the technology department, but still you go and toss off ridiculous statements like that.
Simple way to reduce the complexity of self-driving cars: put them on rails.
Posted by: Boronx on 10/14/08 at 1:16 PM Respond
The complexity of your rail system would be 100 times as difficult of a problem to solve as the self-driving car problem because it would be that much more difficult for cars to get out of each others' way.
As Atrios points out, you get the same parking benefits with a robust car-sharing setup, which doesn't require any new technology.
My guess is self-driving cars would only do their own driving part of the time, when they get into high-density environments like urban spaces. In tight conditions, robot cars could pack themselves onto the road very tightly since all traffic would be under unified control.
Given that humans have not built a downtown in the past 70 years or so, self-driving cars may be a solution looking for a problem. Most of the benefits (density, efficient parking) could be achieved by other means, like car sharing, increased urban density, and trains.
So let's say we get self driving cars. You stumble out of bed at 5:00 am, and roll into your self driving car. You sleep three more hours, then get up, do a towel bath, brush your teeth, eat your breakfast, finishing just as you reach work. You leave the office at 3:00pm, open up your computer and cell phone and get three more hours work done. Read the paper and you are home at 6:30, having commuted from say, DC to western West Virginia, where houses cost $1.57. It's hard to even get my head around the tectonic shifts involved in such a way of commuting.
I fully expect to see self-driving cars within 15 to 20 years. They will be little comuter cocoons. While riding, you will be able to read, eat meals, use a computer, surf the web, watch TV, nap, maybe even work out. After it drops you off, your car will will go park itself until you summon it to take you home.
Combining Google Maps, GPS, and grand-theft-auto level intelligence, I feel like we're getting pretty close in the next decade or two. Especially if you consider that computer cars would have a pretty high acceptable margin for error (as long as there were less than 50,000 road deaths per year and fewer than, oh, 2 million accidents per year, we're still coming out ahead.) The demand is also likely to be huge from suburban commuters, but it doesn't solve that pesky gasoline problem.
Maybe the cars could drive to a sunny place and recharge their batteries when not in use.
Submitted by reason (not verified) on Tue Oct. 14, 2008 1:30 PM PDT.
mmm Some people won't get into planes or trains (much safer than cars) because of the lack of sense of control. And I wonder personally whether engineers won't do the equivalent of what derivatives have done in finance, reduce daily risk but increase catastrophic risk (one out of control robot on a packed highway can do a lot of damage).
"computer cars would have a pretty high acceptable margin for error"
Wrong. People believe computers and robots should be perfect. They'll accept millions of highway deaths at the hands of other drivers, but the first time a robot car kills someone the experiment will end.
Submitted by Legacy on Tue Oct. 14, 2008 8:27 PM PDT.
A bunch of you guys are underthinking the problem. We aren't going to get all humans off the highway and we aren't going to build a comprehensive, nationwide grid of computer-controlled roads. Genuine robot cars will have to be ordinary cars that can drive on ordinary streets surrounded by ordinary human drivers.
And when we have that, we're not too far away from artificial intelligence. These robots won't compose sonnets, but they won't be too many years away from it.
Depending how you define the problem it could be hard or easy. In a walled garden you could probably have self driving cars today. In effect we already have this in various airports where automated trains carry people from terminal to terminal. If you completely open the problem where lane definition is fuzzy as we often see in construction zones and where there are lots of human drivers and pedestrians and dogs and such then we are a long way from a solution. The limiting factory would be litigation. Any computer driven car let loose in a city would be liable for any damage it caused. Currently the individual driver is liable. If they cause millions of dollars in damage they can go bankrupt after the insurance runs out. GM or Toyota is not liable for every traffic accident. On the other hand, if Apple made such a product the lawyers would look at their cash hoard and lick their chops and start suing no matter how small the infraction.
Historically, intelligence computers have always been just over the horizon. I don't see how that has changed. Engineers have gotten amazingly better at digital control, but that is not the same as saying that they have made the conceptual leaps needed to emulate human thinking. I'm not trying to say that is impossible. Just that it is still a problem we haven't cracked.
I think it would be easier if they just built 'sleeping cages' in the basement of where you work. Someone would roust you at 6AM and they could play lulabye music at 11:30 when you returned. That's pretty efficient.
Submitted by marc sobel (not verified) on Tue Oct. 14, 2008 9:33 PM PDT.
It clearly doesn't take intelligence to drive. Artificial or otherwise.
A lot of design work on "Personal Rapid Transit" (PRT) was done in the 1960s and 1970s, some by large aerospace companies. Some projects have continued up to this day. A lot of money was spent in the US and Europe for designs and prototypes, but all major efforts have been abandoned as too expensive or ineffective.
The basic idea can be seen here. And there is a Wikipedia article on it.
Kevin, you're still wrong about the AI involved. No matter how hard you think driving on roads with other humans is, it's still not nearly as hard as other problems, such as having an awareness of self and falling in love.
We're still a loooong way away from genuine "AI"
