6 ms·

If autopilot is slightly better than a human driver, why not use it? It's an incredibly low standard, sure, but more lives are saved, and it only to get better

by kiba 10y ago

If autopilot is slightly better than a human driver, why not use it? It's an incredibly low standard, sure, but more lives are saved, and it only to get better from then on. The only problem is the public's perceived unwillingness to accept such risks in using autopilot.

Tobani 10y ago
The actual problem is that autopilot doesn't remove the driver from the situation enough. It gives people the illusion that they can hand off driving to the computer, but in reality they still need to be doing 90% of what they would be doing anyway. They have to stay engaged in the act of driving. Giving people the illusion that they don't have to stay engaged is where I think the problem is. Until the car is driving itself entirely, I'm not sure there's much to be gained.
- drcross 10y ago
  >It gives people the illusion that they can hand off driving to the computer. The system gives multiple warnings that it's in Beta. >I'm not sure there's much to be gained. There's miles and miles of data to be gained.
  - kalleboo 10y ago
    > The system gives multiple warnings that it's in Beta. We're now in the world of software and users. "Multiple warnings" aren't enough. Chernobyl had multiple warnings. The car should quit if the driver doesn't have his hands on the wheel during a beta test.
  - Someone1234 10y ago
    If I am a driver next to someone being a guinea pig for this beta software, where did I agree to that? Tesla can get the driver to sign off on the risks, but that doesn't extend to other road users who are also at risk from bugs or defects.
falcolas 10y ago
Why not use it? IMO, it's because I don't think that the difference is enough. And the wrecks which are being reported are for stupid things. Not to mention, there's only one datapoint for Autopilot fatalities, which makes any attempt to use it as a statistically significant point useless. Just one more death in the coming months this year would make the number jump significantly above the averages reported in this article. And with the number of potentially deadly autopilot wrecks reported lately, it's inevitable.
- phaemon 10y ago
  (mostly copied from my comment on the other story) Given the number of Autopilot miles driven, you'd only expect one death so far, so that we have one doesn't really tell us anything. However, injuries from car crashes are (fortunately!) much more common - roughly 60 times as likely. So you'd expect to have about 60 Autopilot incidents resulting in injury by now. I don't really keep up with Tesla news but, based on the reaction to the previous story (where the guy wasn't even injured), I'm going to assume the actual number of injuries is tiny.
  - Tobani 10y ago
    You'd really have to look at the number of accidents per mile where people were driving in a single lane on the highway, not the overall number of traffic accidents.
    - phaemon 10y ago
      If you have better numbers, feel free to mention them. The deaths per mile is for highway driving, and I assumed the deaths:injuries ratio is the same, but that could be wrong. I disagree with the "single lane" though: if Autopilot enforces driving in a single lane, and that turns out to be safer, then that's all that counts.
      - gvb 10y ago
        We need "humans driving in a single lane on a highway" for a control group for the experiment - the number of miles driven by a human in equivalent situations as driven by the Tesla autopilot. It may be possible to "control for other factors" as they say, but I have not seen that done. If you don't have a proper control group, the Tesla autopilot will be significantly biased for success because it is dealing with much simpler driving conditions than "in all conditions" human driving.
SCAQTony 10y ago
Tesla over-promised: If they called the Tesla a 'driver-assisted vehicle' rather than an 'autopilot car' this would be a different argument. You can't claim you have a driverless car (autopilot) when it is expected of the driver to keep their eyes on the road, hands on the wheel and a foot near the brake. That is so over-promising.
- dawnerd 10y ago
  Exactly, my 2016 passat has some pretty slick lane assist and adaptive cruise control. Still requires hands on the wheel (although I've watched it take itself around a corner at a high speed and it was pretty incredible). Only thing it's missing is lane change and that's something I could see VW patching in if they were a company that cared about that sorta thing. I think the only thing tesla has over the competition is the willingness to upgrade their cars after they're sold. With everyone else it's basically always going to be how it was when you drove it off the lot. I imagine VW could also tighten up lane assist to not be as slot car and pick the middle of the lane rather than whatever side of the car is closest to the line.
- skykooler 10y ago
  Tesla has never claimed "driverless" - the driver has to be there (well, now - earlier firmware versions didn't check for weight in the driver's seat). What they did do is market it as "autopilot", which means different things to different people - in an airplane, "autopilot" merely flies the plane into the direction and altitude set by the pilot - it does not avoid other aircraft, take commands from ATC, or anything like that. So, technically, calling Tesla's auto-steer "autopilot" is pretty accurate. The problem is that most people being unfamiliar with aircraft systems think of autopilot as "something that flies the plane on its own" and analogizes that to driverless cars. You've never flown in a pilotless airplane. Autopilot is not AI.
cmurphycode 10y ago
Nearly everyone thinks they're an above-average driver. A slightly better than average autopilot will be viewed as riskier! There's also some sort of bias against negative events that are (more) out of your control, I think.
drcross 10y ago
What I wish people would understand is that you can be sure that all of the autopilot crashes which have happened so far will be analysed by teams of engineers to make sure those fringe cases never happen again. It's not as if anyone will ever die in the way ALL of the previous crashes have happened. Every Tesla crash you see is (ideally), the last time you will see a Tesla crash in that manner.
- mannykannot 10y ago
  The real issue is how many latent errors remain. You have described the software industry approach, but there is also the aerospace industry's way, which is more appropriate for cars (note that aerospace software, almost uniquely for software, is given the aerospace treatment.) If only the drivers of the test cars were at risk, I would have no objection to what Tesla is doing - in fact, I would try it myself, given the opportunity.
phkahler 10y ago
>> If autopilot is slightly better than a human driver, why not use it? Because it's not. comparing highway death rates to all-roads-and-conditions death rate is completely disingenuous.
ars 10y ago
> If autopilot is slightly better than a human driver, why not use it? Because it's not actually better than a human driver. It's worse. Much worse. The statistics cited in the article are for a HUMAN plus autopilot. Not autopilot alone. Autopilot alone has a very poor track record.

by kiba 10y ago

Tobani 10y ago

The actual problem is that autopilot doesn't remove the driver from the situation enough. It gives people the illusion that they can hand off driving to the computer, but in reality they still need to be doing 90% of what they would be doing anyway. They have to stay engaged in the act of driving. Giving people the illusion that they don't have to stay engaged is where I think the problem is. Until the car is driving itself entirely, I'm not sure there's much to be gained.

drcross 10y ago
>It gives people the illusion that they can hand off driving to the computer. The system gives multiple warnings that it's in Beta. >I'm not sure there's much to be gained. There's miles and miles of data to be gained.
- kalleboo 10y ago
  > The system gives multiple warnings that it's in Beta. We're now in the world of software and users. "Multiple warnings" aren't enough. Chernobyl had multiple warnings. The car should quit if the driver doesn't have his hands on the wheel during a beta test.
- Someone1234 10y ago
  If I am a driver next to someone being a guinea pig for this beta software, where did I agree to that? Tesla can get the driver to sign off on the risks, but that doesn't extend to other road users who are also at risk from bugs or defects.

falcolas 10y ago

Why not use it? IMO, it's because I don't think that the difference is enough. And the wrecks which are being reported are for stupid things. Not to mention, there's only one datapoint for Autopilot fatalities, which makes any attempt to use it as a statistically significant point useless. Just one more death in the coming months this year would make the number jump significantly above the averages reported in this article. And with the number of potentially deadly autopilot wrecks reported lately, it's inevitable.

phaemon 10y ago
(mostly copied from my comment on the other story) Given the number of Autopilot miles driven, you'd only expect one death so far, so that we have one doesn't really tell us anything. However, injuries from car crashes are (fortunately!) much more common - roughly 60 times as likely. So you'd expect to have about 60 Autopilot incidents resulting in injury by now. I don't really keep up with Tesla news but, based on the reaction to the previous story (where the guy wasn't even injured), I'm going to assume the actual number of injuries is tiny.
- Tobani 10y ago
  You'd really have to look at the number of accidents per mile where people were driving in a single lane on the highway, not the overall number of traffic accidents.
  - phaemon 10y ago
    If you have better numbers, feel free to mention them. The deaths per mile is for highway driving, and I assumed the deaths:injuries ratio is the same, but that could be wrong. I disagree with the "single lane" though: if Autopilot enforces driving in a single lane, and that turns out to be safer, then that's all that counts.
    - gvb 10y ago
      We need "humans driving in a single lane on a highway" for a control group for the experiment - the number of miles driven by a human in equivalent situations as driven by the Tesla autopilot. It may be possible to "control for other factors" as they say, but I have not seen that done. If you don't have a proper control group, the Tesla autopilot will be significantly biased for success because it is dealing with much simpler driving conditions than "in all conditions" human driving.

SCAQTony 10y ago

Tesla over-promised: If they called the Tesla a 'driver-assisted vehicle' rather than an 'autopilot car' this would be a different argument. You can't claim you have a driverless car (autopilot) when it is expected of the driver to keep their eyes on the road, hands on the wheel and a foot near the brake. That is so over-promising.

dawnerd 10y ago
Exactly, my 2016 passat has some pretty slick lane assist and adaptive cruise control. Still requires hands on the wheel (although I've watched it take itself around a corner at a high speed and it was pretty incredible). Only thing it's missing is lane change and that's something I could see VW patching in if they were a company that cared about that sorta thing. I think the only thing tesla has over the competition is the willingness to upgrade their cars after they're sold. With everyone else it's basically always going to be how it was when you drove it off the lot. I imagine VW could also tighten up lane assist to not be as slot car and pick the middle of the lane rather than whatever side of the car is closest to the line.
skykooler 10y ago
Tesla has never claimed "driverless" - the driver has to be there (well, now - earlier firmware versions didn't check for weight in the driver's seat). What they did do is market it as "autopilot", which means different things to different people - in an airplane, "autopilot" merely flies the plane into the direction and altitude set by the pilot - it does not avoid other aircraft, take commands from ATC, or anything like that. So, technically, calling Tesla's auto-steer "autopilot" is pretty accurate. The problem is that most people being unfamiliar with aircraft systems think of autopilot as "something that flies the plane on its own" and analogizes that to driverless cars. You've never flown in a pilotless airplane. Autopilot is not AI.

cmurphycode 10y ago

Nearly everyone thinks they're an above-average driver. A slightly better than average autopilot will be viewed as riskier! There's also some sort of bias against negative events that are (more) out of your control, I think.

drcross 10y ago

What I wish people would understand is that you can be sure that all of the autopilot crashes which have happened so far will be analysed by teams of engineers to make sure those fringe cases never happen again. It's not as if anyone will ever die in the way ALL of the previous crashes have happened. Every Tesla crash you see is (ideally), the last time you will see a Tesla crash in that manner.

mannykannot 10y ago
The real issue is how many latent errors remain. You have described the software industry approach, but there is also the aerospace industry's way, which is more appropriate for cars (note that aerospace software, almost uniquely for software, is given the aerospace treatment.) If only the drivers of the test cars were at risk, I would have no objection to what Tesla is doing - in fact, I would try it myself, given the opportunity.

phkahler 10y ago

>> If autopilot is slightly better than a human driver, why not use it? Because it's not. comparing highway death rates to all-roads-and-conditions death rate is completely disingenuous.

ars 10y ago

> If autopilot is slightly better than a human driver, why not use it? Because it's not actually better than a human driver. It's worse. Much worse. The statistics cited in the article are for a HUMAN plus autopilot. Not autopilot alone. Autopilot alone has a very poor track record.