SeanR

hacker-pschorr

I'm not moving it.

Wisconsin Joe

Nu-Uhhhhhh.

That's a piece of granite.

And self driving cars are showing a far lower at-fault accident rate than human driven.

However, they have a slightly higher rate overall. Mainly because they can and will stop fast if needed. So they get rear-ended by idiots on their phones a lot.

Even if truly self driving cars become mainstream, I think there will still be enough enthusiasts that the idea of outlawing driving won't happen.

Tom. M

I for one welcome our new self - driving car overlords....
The way most people "drive" these days, it'll be a huge improvement.
I am absolutely fine with driving at the track and having the daily commute done by automation....

Otto Mechanic

I don't think I've seen the video, I may have, but I honestly don't believe it's important. It's impossible to answer the question you pose, whether a human may have been able to stop. The fact a human wasn't in control of the machine and it caused the death of another human is the only important issue, which you point out later when you begin "The #1 question..."

Not use them. Not buy them. Not participate. I don't think it will take much overt planning or activity, the industry seems bent on doing it to itself.

I'm sort of familiar with AI and the state of the art, we aren't within 1 generation (human generation) of general artificial intelligence. It isn't the first time the Yuga has been declared either; during the 80's many were convinced general AI was "just around the corner", that was almost 40 years ago. Interestingly enough, the technology of "deep learning" being touted today as the final solution was known at that time, but wasn't the subject of outrageous hyperbole aimed at drawing investor money. Nothing has actually changed and neural networks are no more "intelligent" now than they were then. Make no mistake, self driving autonomous vehicles will require general AI to really work, nothing less will do outside of highly controlled and formalized environments.

What do "we" do? What we've always done; we'll be cautious. When the typical suburban soccer mom is faced with being an "early adopter" of autonomous vehicles, she's unlikely to need convincing; she isn't going to load her 5 year old into a pod and wave as she disappears in the distance. I don't think that's going to happen for a long time, and before it does the devices will manage to fail in some predictable but spectacularly horrific fashion that will prevent it from ever happening.

Consider that we've had commercial "self flying" aircraft since the 75/76 generation, just about 40 years too, but we still require a flight crew, even on vehicles that are only carrying cargo. Yep, they spend most of their time reading or sleeping (or doing other things), but they're still there and they can take control at any time. If semi-autonomous ground vehicles are to be successful, removing the controls isn't the way to make it happen.

I can honestly believe we might see semi-autonomous cars on major urban and suburban arterial routes very soon, and later on urban capillary streets, but I expect the cockpit won't be eliminated and we won't be seeing fully autonomous vehicles in our lifetimes. I would seriously consider buying a car that I could pilot to a freeway on-ramp in San Jose, push a button to engage an auto-pilot, set my destination for El Paso de Robles (or Rock Springs Wyoming) and go to sleep for a few hours/days. You could sell me something that did that, and I think the tech to do it is available now. I don't think I or my kids (or even grand kids) will live to see that same vehicle make it's way from a Hwy. 85 off-ramp to my home in the Santa Cruz mountains without waking me up though. That's a very different level of technology.

In the case of a fully autonomous vehicle, the legality will be (and in most minds already is) obvious; liability lies with the manufacturer as long as the machine's owner adheres to all manufacturer instructions (maintenance & etc).

It won't be nearly as clear in "mixed use" or "semi-autonomous" operations, and that's likely to cause real trouble. We've already seen the Tesla example and that hasn't been a benchmark case. When is it the human owner/operator's responsibility to wrest control of the vehicle away from the auto-pilot? What level of diligence is expected? This is exactly the question that may prevent the deployment of anything short of a fully autonomous vehicle. Unless the division of liability can somehow be made unequivocal, it may be we can't deploy anything less. How comfortable would you be reading a book or sleeping on a freeway if you knew that a failure of the autopilot might render you destitute or maybe dead? How useful or attractive would the feature be then?

Well, I see a two-fold solution to that problem, the most obvious being to not buy a modern car. The second (admittedly less attractive solution) is to make it illegal for companies to sell devices that absolutely cannot, under any reasonable circumstance, be repaired and maintained by their owner. That's the iFixit approach I mentioned.

The need of a computer isn't an excuse. I don't know of anyone, with the possible exception of my 83 y/o mother, who doesn't own and use a computer. Mom also doesn't fix her car anymore. There will always be folks who neither want to or are able to fix things they've bought, however it should be possible for a typically "able" person to perform repairs on equipment they've purchased. Manuals, parts and tools should all be available. Anything less is a lease, and we already understand leases. When Apple Inc. makes a phone that can't be repaired by you and that you have no control over the operation of (i.e. it can be remotely controlled without your consent up to and including being "bricked") you don't own it, you lease it. That's not a problem, but we need to stop letting them call it a purchase, because it isn't. It's probably necessary to require a computer, but it isn't necessary to require a different one for every deveice you own. That's not reasonable.

John Deere has already figured that out and they don't even offer to sell equipment anymore I don't think. Last time I visited one of their showrooms was 2002 as I recall, in Idaho Falls. A 40 horse tractor started at about $178,000 back then, I expect they're more now. That was a base model. I couldn't afford it, but that was OK because they had a very attractive lease deal available, which is what they really do now. They don't sell equipment, they lease it.

I eventually did buy a JD tractor though, and I even bought it from a dealer. It was a lease return, a 1976 JD 4630. Great tractor and I had a shop manual so I could fix it myself, which was lucky since the only time I ever called the dealer for service it took him a week to show up (during which time my hay got rained on), then he installed $600+ fuel pump that didn't fix my problem (which turned out to be rust in the fuel tank. Moron).

Anyone who leases a tractor they can't repair is a fool and won't be in agriculture long. Pure Darwin.

I'm sure you're right about where they want to take us, and why. I don't plan to go there. You're going to have to decide for yourself.

hacker-pschorr

It's very important. The person was crossing an unlit street & could not be seen. As technology gets better this kind of situation the AI might actually be safer since infrared and other sensors might be able to notice things the human eye never will.

When talking about accidents, it's absolutely imperative to try and determine how avoidable the accident was to begin with. One of the first self-driving car vs pedestrian accidents I remember involved a person walking out from between two parked cars right in line with the AI car. Even if the "driver" (computer or human) had seen a glimpse of the person, there wasn't enough space to stop the car at the given speed limit.

Too late, look at how many cars have active braking systems, automatic cruise control, and some are so advanced they will stay between the lines without human interaction (not supposed to, people figured out a way to "hack" the system by taping something to the wheel so the computer things you are hands on).

Many consumer surveys say this is a big want by the public. Nothing drives innovation faster than demand.

However, the biggest hick-up I have yet to see a viable solution for is winter. This may be the #1 factor that keeps self-driving cars from fully taking over. If the AI cannot "see" the road......makes things tricky. The idea of lining every road with sensors is impractical IMO.

That's not a solution. That's you personally avoiding it, doesn't work on a mass scale very well when people want to buy new modern things.

Click the links I posted, such legislation has been on the books for a while now but the manufactures are keeping one step ahead and fighting tooth and nail to keep things locked up.

You don't understand the scope of this problem at all. I deal with this on a regular basis and later today I'm on my way to a local shop having issues with one of the 4 specialized computers, each one for a different brand of car and each one costs thousands of dollars up front + subscription. It's impossible to fix these cars without such equipment and they still cannot do 100% of what the dealership computers can do - and no, you cannot buy the dealership computers unless you are a dealership.
These are not all "new" cars anymore, going back 15+ years such devices are needed for even simply regular repairs.

Welcome to 2018 and it's only going to get worse. It's fun when a car comes in you've been servicing for a few years now without issues suddenly won't talk to your equipment because the dealership fixed a recall & "flashed" the ECU's with an update locking out the non-factory systems. The company who makes the aftermarket computer had to update their software and send it to their customer base (hence the subscription service).

A friend of mine rotated his own tires on his Toyota. He had to pay $100 for the dealership to re-set his TPMS since the sensors were now in the "wrong" place. Sure there are some aftermarket devices that may work, depending on the year of the Toyota, that's not the point.

GT6ixer

Actually in this case the cars sensors did detect the pedestrian but the automatic braking was disengaged.

"At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision. According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. The system is not designed to alert the operator.”

https://arstechnica.com/cars/2018/05...are-ntsb-says/

Otto Mechanic

It's very important? I don't believe it is in this example. That an IR enabled AI may have seen the pedestrian is theoretical, it apparently didn't. Would a human have seen the unlit pedestrian? Maybe, maybe not. A human may have been more cautious because the light was bad. What tells a machine to be more cautious when light is poor? Nothing; not using "deep learning". Ask one of the programmers of that system to tell you exactly where "caution in low light" is expressed by the machine's training, it almost certainly couldn't be done, and that assumes it even exists.

You sort of need to know a little about the way neural networks actually work, the instructions aren't specific, everything is learned by example. In this example, it's pretty difficult to teach since we don't have many "learnable moments" involving the death of pedestrians in low light conditions captured on film, so it's highly unlikely that behavior was trained in either a positive or negative way.

We'll likely find other examples of unequivocal liability in individual cases, yet still be unable to generalize. The first example you give, implicates the machine, the second exonerates it. How is liability to be decided in that environment?


Hacker, I'll answer your assertion with a challenge; if I turn on my automatic cruise control, then allow it to plow into the car in front of me, who's fault is it? The manufacturer or the operator?

This is a toy problem. It's the operator. No question.

How does that analysis change when the "cruise control" is supposed to know better? That's the question, and you're right to say it's already being asked, but its far from settled and so it's hardly "too late". The other shoe has yet to drop. The results aren't in. The fat lady hasn't sung.

True, but many surveys also predicted a landslide win by Hillary in 2016. I don't think this is over.

That's a trivial problem. Yes, embedding markers in the roads and sensors in the cars are both cheap and effective. It doesn't take winter conditions to make that necessary.

I assure you I do, and with respect to an environment similar to the one you describe, with the difference being I have a background in artificial intelligence and its application to self-diagnosing computer systems. I have at least a limitted grasp of the subject, limitted enough to be confident saying the state of the art isn't close to approaching general AI anytime soon, and that nothing less will suffice.

Otto Mechanic

Good memory Nate, I read that Ars article too. Apparently Uber deliberately disengaged the native braking system because they wanted their final system to be portable, independent of the car's sensors? At least that's the explanation I think I read.

Regards,

docmirror

So let it be written, so let it be done. I will not muck up this thread any more.

Wait, what were we talking about? Oh yeah, money. It's a hit - don't give me that do goody-good bullspit. Money, it's a gas - grab that stash and dash.

dr bob

I write what I like to call intelligent software to run power plants. But its based on user responses and sharing of things that happen, spread over a disparate fleet of units and equipment. I wish there was an easy way (read: I was smart enough) to build some direct AI learning capability into the critical decision-making sections. As it is things adapt to wear and environmental conditions, but the list of failure and protection conditions and actions is static. Until I or someone tells the system that something that happened was wrong, it will continue to make the same mistakes or correct actions blindly until it's told otherwise.

Meanwhile... All drivers are above-average when you ask them. We get to see some pretty dumb-sh!p driving just by watching our own car-cams or (gasp!) looking out the car windows. Somebody might be reading a message on their phone or dash display or GPS or a billboard, sipping coffee or munching a taco, distracted enough to add half a second or more to their reaction time. Soccer parent yells at the kids in the back seats. The self-driving car of the future doesn't need to be perfect, it just needs to be better than that average. Like we are. Look for a minute at the driver aids that have come to us just in the last decade or two. ABS was hugely controversial when it first arrived in two-channel form, merely releasing brake pressure when it though a wheel or two was slowing faster than it/they should. Thirty-plus years later, the systems and owners' expectations have turned to blind reliance on the systems. The systems have expanded to include various stability and traction-control capabilities, all integrated with driveline management. Now (I mean currently) with collision avoidance, adaptive cruise and lane-guidance technology, drivers need only aim the car down the road once, push the cruise button at whatever max speed they want. Systems meant to supplement the driver's shortcomings are taking over more and more, so drivers depend on them more and more so they can do less and less while behind the wheel.

A look at the Model T might be a good snap-back. You get to adjust mixture and spark advance using column levers. Throttle might be another lever or might be a "foot feed". Pedals change directions and operate cable-actuated brake mechanisms on drive-shafts. You sit out in the weather. You also shared the unimproved road with horses and buggies. Drivers were those brave enough to risk their lives for the extra speed and freedom, and those folks also learned quickly the limitations and mechanical demands that made driving such an adventure. What's that handle folded in front of the radiator, and where's the keyless start button?

Once the modern cars' systems learn to communicate with other cars, human drivers will be a dangerous exception to the ordered flow of traffic. Eventually human operation will be restricted or even outlawed, along with those stinky combustion-engined tanks that are 20th centure automobiles. Modern self-driving vehicles will have no need for any of that heavy and unsightly crash protection structure. No need for headlights or tail lights. Pods will link together electronically and maybe even mechanically for auto-caravaning in groups. Pods will have just enough motive capacity on-board to transit between controlled auto-travel segments without disturbing the comfort of the passengers inside. Human drivers will be obsolete, banished to the nether regions where they will gather in groups to celebrate "the old days", driving carefully around small maintained roadbeds sections. There will be a replica section of the Santa Monica Freeway intersection with the 405, another of the Van Wyk Expressway and some Manhattan streets. On those sections, the small and very expensive petrol supply will be sold to support these activities, with costs so high that only a small slice of the population will be able to afford "driving". Tires will be a rationed and expensive factor for folks who "drive". There will be additional and very punitive taxes assessed for the toxic tailpipe pollution that will poison the otherwise-pure air we breathe.

Think about it...

Otto Mechanic

And I'm not moving until someone puts bacon on it.

dr bob

We take stuff like that for granite. I truly marble at the apparent ability of some to make sense out of a picture of a rock.

Crumpler

I'm secretly optimistic that Sky-Net will become self aware before this happens to me...

Dan87951

Munk's is top notch! The fellas over at Munk's know the 928 very well. As a person who is very detailed oriented, i'm very hard to please, with that being said, Munk's is one of the few places I would trust to work on my car. Sure they may cost a little more, but they will do the job right. Good luck!