October 27, 2015

"In reality, very little of this is true"

Self-driving cars, in the snow? Nope.

Parking lots also pose a problem.

It also has trouble spotting some people.

Nor can the car detect most road damage.

In summary, the MIT publication concludes that the car is barely able to do many of the things the public believes it can:

Colorado Technology Posted by JohnGalt at October 27, 2015 6:41 PM

Huh? The technology is imperfect?

Curious if my blog brother remembers tape library demos where the robotic arm convulsed on stage in a company meeting? WHAM! WHAM! WHAM!

I'm not going to shelve the whole project or my enthusiasm. I'll admit a likely disconnect between perception of progress and reality. I was rather stunned to hear that they have completed 100,000 miles in traffic. I have seen some very bad demos on TV (almost whamwhamwham) where the car is told to go park itself.

I'm not betting on dates so much as anticipating the safety, creative destruction and productivity increases it portends.

A post was going around Facebook about the ethics of software's choosing the least worst outcome in a crash: do you hit the baby, the dog, the Raiders fan, or plow into a crowd? I am going to put this in with the Policeman example. Police will need different tools to stop these cars than arm waving.

The ethical crash is pretty rich to me. Sheer stupidity kills tens of thousands every year. Now we are assuming that all these drivers are quick thinkers and beneficent philosophers with superb reflexes.

Many things to figure out and I am rather astonished that they have progressed as far as they have.

Posted by: jk at October 28, 2015 10:05 AM

You are a software engineer and I, a hardware engineer. I suppose I shouldn't be surprised that you think the software is perfectable and I think it's a fool's errand to try.

It's not the technology I'm worried about, it's the imperfect world in which the technology must coexist. It seems you're willing to accept a few dead pedestrians "for the greater good" of fewer drunk drivers. Well I can assure you the ambulance chasing lawyers are not. Those guys have even stopped tire stores from repairing tire tread punctures unless they're in the center of the tread. Perfectly good tires go in the waste bucket because one person had a blowout and was killed in the crash, and lawyers convinced a jury that a tire repair somewhere in the tire's history was to blame.

And you're going to put Siri behind the wheel? Not just policemen will need a tool to stop these cars. This may just be the justification for private ownership of bazookas. If the lawyers ever let it get that far.

Posted by: johngalt at October 28, 2015 2:18 PM

I have always thought that the tort bar threatened this dream far more than technical challenge. But we are not the only ones looking at that, and I've read a few articles written by people confident that it is superable. On a tangent, I wonder about the first death or serious injury when one of the auto-breaking Mercedes stops and creates a worse outcome -- that happened yet?

This software guy would suggest a phased release: keep them limited in geography, application and weather until they have proven safety, then gradually increase their domain.

Barring an incredible breakthrough, I see a transition first to Uber/Lyft, then autonomous cars taking over the easier routes. Lather. Rinse. Repeat.

Not certain I approve of the interpretation of my comment as "accepting a few dead passengers." Tens of thousands of completely innocent Americans are killed every year. This leads me to suggest that maybe these "tonomous" pilots are not really doing such a swell job. They can't handle their radio, cell phone, Starbucks and a cute member of their preferred gender on a nearby sidewalk without destruction. I rejected accepting that these same folks would make the right choice in a morally ambiguous situation and possess the skill to achieve the selected outcome ("I figured if I sped up, I could jump over the crowd of schoolchildren at the bus stop and miss the squirrel -- sorry!")

While we're pointlessly arguing .. . the tire outcome might be good. Thanks to global trade and comparative advantage, tires are becoming cheaper and cheaper while the monetary value of human life increases.

Like the mandatory left-turn-only-at-a-green-arrow lights across Boulder County. Ridiculous, but my 45 second delay to Starbucks versus a 1/10,000,000 chance of saving a life is unconvincing.

Now, if we had autonomous cars. . .

Posted by: jk at October 28, 2015 2:52 PM

Pedestrians. "...a few dead pedestrians."

Unless I missed something, no company has any liability for the stupidity of its product's users. But when the product itself does something stupid...

Posted by: johngalt at October 28, 2015 3:20 PM

Mea maxima culpa! I saw "pedestrians;" I mistyped.

I'd say in the "Runaway Toyota" contretemps, they were sued for owners' stupidity. All the warning labels "do not ride bike at night without light" imply stupidity is really not off-limits to the torts.

Posted by: jk at October 28, 2015 6:13 PM

Fascinating discussion! I agree with JK:

Many things to figure out and I am rather astonished that they have progressed as far as they have.
As far as the "ethical crash" scenario (as only a bunch of softies could conceive!), I think there are compromises that will be worked out - partly in court, partly by committees, and partly by drivers making choices - such that this CAN happen (the technology, not the crash) at some point.

First one is probably: these cars go 5-10 mph under the speed limit and feature a flashing amber light: "under computer control! Cross at your hazard!" Police might need to have a veto, but probably not... if it follows proper rules (stay on street, under limit, fixed destination)

Does system/software design require the study of the Hippocratic oath?

Posted by: nanobrewer at October 29, 2015 1:16 PM | What do you think? [6]