I don’t think their current testing has too great a potential for disaster, after all there is a driver sitting behind the wheel waiting to take control if the car shows any sign of danger. I also imagine this as a gradual process, starting with increasingly sophisticated cruise control—i.e. I wouldn’t expect physical steering wheels to disappear until several years of confidence had been established with cruise control systems so smart and reliable that they demonstrated a clear ability to dramatically reduce the number of accidents.

And I think it’ll show up first as either farm or military equipment, then find it’s way into the trucking industry and the luxury car cruise control systems, then trickle down into the more general consumer market.

I agree it would be less efficient than mass transit, but I don’t think mass transit can really compete with cars in large portions of America due to our lower population density. It also ought to be more efficient than normal cars, since the computer can optimize acceleration & braking tasks more accurately than humans.

There are some people who think automated cars could also become a form of public transport, where you called automated cabs, which would reduce the need for parking which turns out to be an enormous waste of resources. And then there is the potential to automate public transportation itself, like buses, since they too suffer occasionally from human failings. A few years ago a train operator in Boston collided with another train because he was texting—the radar based cruise control systems of many luxury cars could have easily prevented that, which is why I think we’ll see our society do this sort of thing.

And with the size and power of computers, why not just have multiple redundant systems? The Space Shuttle four redundant computers all checking one another with a fifth backup holding emergency landing procedures on top of that—the size power & cost of computing is making that trivial, you could double up the radars even and power everything separately, though none of this is really that important if you still have someone sitting behind the wheel who will be responsible for handling the car in the event of a catastrophic failure, the way we treat current cruise control systems.

Your skepticism is good, but I think we still probably disagree on a lot here, though I’m sure some is in part due to my naivety. Thank you for engaging in this discussion with me, and civilly no less, I hope I’ve been equally courteous.

“Did you see how it has no problem identifying cyclists and pedestrians and side walks and signals and even understands what sort of ‘body language’ to express at a multi-stop intersection to negotiate right of way?”

Hey, what you saw was a sales pitch. I have no doubt such systems are impressive, but what we see in that video is how these cars behave under ideal conditions. I’d like to see a demo of how they respond when something goes wrong with any or all of their many subsystems. Automated machinery can simply shut itself down when something goes wrong. An automobile traveling at highway speeds in heavy traffic cannot.

It’s possible that at some future date this technology might be sufficiently fail-safe for general use, but the presentation I saw was clearly based on how it would operate under ideal conditions, which tells me the promoters haven’t really thought through all the many possible situations under which drastic failure, with serious consequences, could occur. Nevertheless, they are currently testing these vehicles under actual highway and city conditions, which as I see it is a disaster waiting to happen.

I can see this technology as useful for the military, under combat conditions, as a control system for a drone tank, for example. And I can see it as possibly useful in strictly controlled settings, as for example in special highways set aside exclusively for such vehicles. Also I can see it piloting airplanes, where the environment will always be extremely predictable.

But I can’t see this as the future of land-based transportation. Even if it worked as advertized, it would be a huge waste of resources compared with mass transit.

Victor, you’re missing the point: there is no database. These cars use databases for addresses, but they don’t need to “pre-know” anything about the roads’ directions, speed limits, etc., they learn it all as they go, just like us. And yes, of course they have to be able to handle all the hazards you raise, just like a human driver, did you watch that video? Did you see how it has no problem identifying cyclists and pedestrians and side walks and signals and even understands what sort of ‘body language’ to express at a multi-stop intersection to negotiate right of way?

Some types of AI are far off, yes, but why should driving be one of those types? It doesn’t require a lot of the more challenging aspects of higher thought, it operates in a very limited realm, and many of the technical problems have been under intense technological development over the last several years due to the DARPA challenges.

And failsafe operation is an obvious given—like a Segway it’ll have to operate in a safe way even under massive multiple failures. I was kind of thinking even home computers have sort of overcome crashing, I don’t remember the last time I had a crash (though I have a mac and reboot probably monthly), and my old roommate who has a PC ran his computer for many months without rebooting or anything.

Watch the video, it is making a reasonably high resolution 3d model of it’s entire surroundings at all times, it definitely has a better eye on the car’s relationship to the world within 50 feet than any human can possibly have. (360° continuous 3d mapping and highlighting all moving objects, estimating their trajectories, detecting signals and signs, looking for any interference with it’s planned route.) I really think you should check out the video I linked before you argue these points—I am just as concerned with safety as anyone can be, I’m convinced it won’t be more than 10 years, maybe only 5, before someone is selling automated driving systems somewhere.

I honestly think they can be made so reliable and sophisticated that when they do inevitably get involved in accidents they’ll have detailed 3d models of the local environment leading up to the accidents that will both vindicate them and produce greater trust in them.

Here is the video again: http://youtu.be/YXylqtEQ0tk

It’s a fourteen and a half minute presentation about some of the technology behind the car, it’s functioning and the challenges they’ve faced thus far. I’m not saying they’re ready, but I find it hard to think they won’t be within a few to several years.

Also, stop thinking in terms of a program and supporting hardware that is programmed to do one thing rigidly and think instead of a system that can record a comparable level of visual detail, a 3d scanner providing a 3d model with matching detail, and the well developed software for facial recognition applied to various categories of obstacles. You know facial recognition software is now better than humans? And HD cameras fit in cell phones?

The potential for automated cars to reduce accidents is really mind blowing. And that is why it will ultimately take over.

We’ve all heard of sudden acceleration problems in cars equipped with computerized driving controls. The designers are in denial, but no one knows for sure what causes such problems. My guess is that these computers crash from time to time. Usually, however, the driver can take control and override the computer, which is why such events are relatively rare. A driverless car would be completely out of control if its computer crashed. Even if there were some relatively “minor” software glitch, that could have serious consequences.

The sort of optical recognition that would be required to take all possibilities into account, including obliterated or oddly positioned signs, power outages that affect traffic lights, cars that suddenly pull out of their lane, etc., requires a sophistication far beyond what is now possible for any AI system.

I must say I’m also disturbed by the many posts here expressing concern for the legal ramifications, but completely indifferent to basic safety issues. Brave New World we live in!

As long as this holds true, it may be that the driver will continue to hold responsibility and liability on the automation. After all they can intervene at any time and rescue the system.

What if the driver is asleep or distracted? That could be addressed by simply legislating that the driver cannot sleep or be distracted. Although I have to admit, the latter severely limits the appeal of driverless vehicles.

Given a choice people will instinctively only accept a high safety level before they would give up control.

Animals that didn’t mind putting themselves in dangerous situations where they have no control and their genes play a lesser part in their survival prospects probably died out long ago.

Whereas if the animals were in dangerous scenarios where their genes have greater influence over their survival odds (climbing cliffs etc), the survivors may go on to contribute potentially good traits to the population’s genetic pool.