4 ms·

Common in early car and aircraft cultures as well. Always a tension between innovation and safety. I don't mean that as a cop-out. Rather that there's no singl

Common in early car and aircraft cultures as well. Always a tension between innovation and safety. I don't mean that as a cop-out. Rather that there's no single "right answer". We are choosing something on a spectrum of risk vs reward. Personally, although I have no direct experience, it's starting to look like Tesla is being too risky. Yes, they're safer than human-only drivers, but the mistakes they're making are too algorithmic. They're not enough safer to call Autopilot a success. Contrast that with Waymo, who are paranoid-ly safe. They seem to be taking the long view that building trust (and not taking any chances to lose that trust) is as important as technical achievements. Edit: Or, at some other extreme of poor citizenship, Uber, who seem to have gone off the deep end of "Keep up appearances of having a self-driving research program even though it's a disaster internally."

TAForObvReasons 9y ago
I firmly believe that a technology deployed in a potentially life-threatening context like a car or plane or pacemaker deserves a much more conservative development approach. "Move fast and break things" isn't appropriate when human lives are at stake. Waymo is applying the appropriate conservatism; Tesla and Uber are playing fast and loose by comparison.
- snthd 9y ago
  The dilemma is that slow development costs lives as well :-/
  - chaboud 9y ago
    Unfortunately, that sentiment, wielded as a shield by the ambitious, papers a veneer of virtue over the callous disregard of cutting corners with other people's lives. The reality is that careful development is expensive. If Uber and Tesla were racing to save lives, they'd open up their work to the public. They're racing to make the most money possible in the end. Part of that equation is cost. Another is maximizing the utility of available resources. They only need to be safe enough to not get shut down or lose substantial market share (ongoing capital).
    - saalweachter 9y ago
      I think the exact logical flaw is that saying "slowing down development for testing costs lives" assumes the solution will be validated as correct. Waiting a year to deploy the correct solution for want of testing costs the lives that solution could have saved in that year, yes, but pushing one correct and four incorrect solutions immediately into production - when the incorrect solutions each cost twice as many lives as could have otherwise been saved - and re-evaluating after a year kills far more people than are saved.
watsocd 9y ago
32,000 people die every year in car accidents in the US alone. If automation reduces that by only half, that's 80,000 lives saved. The real problem in the press is that these crashes are like plane crashes; rare but catastrophic.

TAForObvReasons 9y ago

I firmly believe that a technology deployed in a potentially life-threatening context like a car or plane or pacemaker deserves a much more conservative development approach. "Move fast and break things" isn't appropriate when human lives are at stake. Waymo is applying the appropriate conservatism; Tesla and Uber are playing fast and loose by comparison.

snthd 9y ago
The dilemma is that slow development costs lives as well :-/
- chaboud 9y ago
  Unfortunately, that sentiment, wielded as a shield by the ambitious, papers a veneer of virtue over the callous disregard of cutting corners with other people's lives. The reality is that careful development is expensive. If Uber and Tesla were racing to save lives, they'd open up their work to the public. They're racing to make the most money possible in the end. Part of that equation is cost. Another is maximizing the utility of available resources. They only need to be safe enough to not get shut down or lose substantial market share (ongoing capital).
  - saalweachter 9y ago
    I think the exact logical flaw is that saying "slowing down development for testing costs lives" assumes the solution will be validated as correct. Waiting a year to deploy the correct solution for want of testing costs the lives that solution could have saved in that year, yes, but pushing one correct and four incorrect solutions immediately into production - when the incorrect solutions each cost twice as many lives as could have otherwise been saved - and re-evaluating after a year kills far more people than are saved.

watsocd 9y ago

32,000 people die every year in car accidents in the US alone. If automation reduces that by only half, that's 80,000 lives saved. The real problem in the press is that these crashes are like plane crashes; rare but catastrophic.