3 ms·

'graybox' AI products (mostly automatic but periodically return emergency control to pesky human) are going to need to model driver alertness the car certainly

'graybox' AI products (mostly automatic but periodically return emergency control to pesky human) are going to need to model driver alertness the car certainly should warn you at lower danger threshold if it thinks you'll need more time to return to full attention (even in a zero-shame case, like for example you're checking your mirror but the car thinks there's a hazard in front) separate question of 4th / 5th amendment consequences of using this for crash liability

marcosdumay 6y ago
The human attentiveness model is very easy: `is_attentive _ = False`. Those grayboxes probably already take more lives than they save on airplanes, and will be orders of magnitude worse on cars.
- nickik 6y ago
  > Those grayboxes probably already take more lives than they save on airplanes, and will be orders of magnitude worse on cars. All evidence so far point in the opposite direction. Airplanes are different, you have a highly trained pilot that has strict requirements for flight. Outperforming your avg human is far easier. What your evidence that these system cause more harm then good?
  - marcosdumay 6y ago
    > What your evidence that these system cause more harm then good? I have stopped following the accident investigations a few years ago, but that was the (very evident) trend. Most accidents tend to happen at the hand-over from the machine to the pilots when some problem appears. Among those it's very common that the both the machines and the manual procedures deal correctly with the problem, but the pilots have no chance of selecting the correct procedures. (The MAX-8 may have changed that picture all by itself, as I said, I stopped following them.) And, yes, airplanes are different. You have pilots that are trained on getting hand-over in a bad situation, and usually have minutes to decide what to do. Car problems must be solved in a second or two, and the drivers never see a bad situation before it happens on real life.
    - nickik 6y ago
      Just looking at accidents that did happen, its not a reliable way of judging overall safety.

marcosdumay 6y ago

The human attentiveness model is very easy: `is_attentive _ = False`. Those grayboxes probably already take more lives than they save on airplanes, and will be orders of magnitude worse on cars.

nickik 6y ago
> Those grayboxes probably already take more lives than they save on airplanes, and will be orders of magnitude worse on cars. All evidence so far point in the opposite direction. Airplanes are different, you have a highly trained pilot that has strict requirements for flight. Outperforming your avg human is far easier. What your evidence that these system cause more harm then good?
- marcosdumay 6y ago
  > What your evidence that these system cause more harm then good? I have stopped following the accident investigations a few years ago, but that was the (very evident) trend. Most accidents tend to happen at the hand-over from the machine to the pilots when some problem appears. Among those it's very common that the both the machines and the manual procedures deal correctly with the problem, but the pilots have no chance of selecting the correct procedures. (The MAX-8 may have changed that picture all by itself, as I said, I stopped following them.) And, yes, airplanes are different. You have pilots that are trained on getting hand-over in a bad situation, and usually have minutes to decide what to do. Car problems must be solved in a second or two, and the drivers never see a bad situation before it happens on real life.
  - nickik 6y ago
    Just looking at accidents that did happen, its not a reliable way of judging overall safety.