4 ms·

"turned out"... it's been starkly obvious since day one that true self-driving is a problem that requires general intelligence to solve. All you have to do to k

by cmsj 6y ago

"turned out"... it's been starkly obvious since day one that true self-driving is a problem that requires general intelligence to solve. All you have to do to know that is drive a car and think about what's actually required to successfully drive from A to B in the real world.

howlgarnish 6y ago
Waymo is running self-driving cars with paying passengers and no backup drivers right now in Phoenix, and last I checked we don't have general AI yet. https://www.reuters.com/article/waymo-autonomous-phoenix/waymo-opens-driverless-robo-taxi-service-to-the-public-in-phoenix-idUKL1N2GY21V https://www.reuters.com/article/waymo-autonomous-phoenix/way... Sure, you can move the goalposts and say it's not "true" self-driving until it can handle all environments and weather conditions and whatnot, but self-driving cars are here. The question is how fast they'll go mainstream.
- jessaustin 6y ago
  Something something "self-driving-complete"...
- bo1024 6y ago
  This doesn't prove that the technology is ready or safe. It only proves that Waymo thinks it's a worthwhile gamble. This is the same place where Uber's self-driving car killed a person with no repercussions to the company.
  - nmfisher 6y ago
    "Ready" or "safe" are matters of opinion, though. It's ready/safe when governments ("the people") decide it's such.
  - howlgarnish 6y ago
    It's obviously ready and safe enough for Waymo to roll it out (and they've been a lot more conservative than Uber here). As for safety, all we really need is for them to be safer than the average human driver. As evinced by the ~40,000 road deaths every year in the US, that's really a pretty low bar. I'd even argue that full autonomous is/will be less risky than halfway houses like Tesla Autopilot where neither side is really in full control, but that's another story.
    - woeirua 6y ago
      That’s actually a high hurdle to clear. Looking at fatalities per hundreds of thousands of miles driven humans are insanely good at driving. Using the 2018 NHTSA number of 1.13 deaths per 100 million miles driven, avg speed of 60 mph, 20 predictions per second, that equates out to an autonomous system can make a fatal mistake on average about every 120 billion predictions.
      - sitkack 6y ago
        You make it sound like every mis-prediction could result in death.
  - berdario 6y ago
    The technology is safe: https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf https://storage.googleapis.com/sdc-prod/v1/safety-report/Way... They've been very careful about it, and they have not finished work on it, so they need to keep the same standard that they had until now. The technology is also ready for Phoenix. The company is not yet ready for the whole world (but I have no specific insight into what's holding them back). It's totally different from Uber.
- 2muchcoffeeman 6y ago
  The obvious question is, why does this work in Phoenix? Where I live, road conditions are changing all the time. There’s new construction, new roads opening, temporary (a few hours at a B time) changes in lanes where you have to drive against the normal direction of traffic. You can call it moving goal posts, but to deploy them generally, they either have to have general intelligence, or you need to input construction schedules, so that it won’t take certain roads during certain times. And then what? You can’t order a ride to certain locations?
  - howlgarnish 6y ago
    Waymo has been dealing with construction since day one. If the road is entirely blocked, Google Maps/Waze is already capable of changing routes on the fly. There will always be corner cases that automation can't deal with, but Waymo will have some human troubleshooters around for that. So do us humans, for that matter: when a car breaks down, most people call the mechanic instead of applying their own general intelligence to fixing it.
    - 2muchcoffeeman 6y ago
      What’s the mechanism? Eg A road is accessible from 5am to 8pm. Outside those times traffic conditions change. Humans setup cones and control the flow of traffic. 2 lanes becomes 1 lane for all traffic. And traffic is managed so that depending on what direction you’re going, your driving against the normal flow of traffic. This is a real scenario in a moderately busy area. Times may be wrong. The road is not blocked. But you have to take direction from a person. In fact this sort of temporary traffic change is fairly common especially due to road works.
- dTal 6y ago
  That's not moving the goalposts, those are the goalposts. Anyone can make a car that loops round a racetrack by itself, on a good day (well, not anyone, but it's relatively straightforward). The challenge is making it robust against "the real world".
DennisP 6y ago
Not human-level AI, though. Plenty of animals do fine with challenging navigational problems. Birds fly through forests, catch or evade other birds, cross continents, etc.

by cmsj 6y ago

howlgarnish 6y ago

Waymo is running self-driving cars with paying passengers and no backup drivers right now in Phoenix, and last I checked we don't have general AI yet. https://www.reuters.com/article/waymo-autonomous-phoenix/waymo-opens-driverless-robo-taxi-service-to-the-public-in-phoenix-idUKL1N2GY21V https://www.reuters.com/article/waymo-autonomous-phoenix/way... Sure, you can move the goalposts and say it's not "true" self-driving until it can handle all environments and weather conditions and whatnot, but self-driving cars are here. The question is how fast they'll go mainstream.

jessaustin 6y ago
Something something "self-driving-complete"...
bo1024 6y ago
This doesn't prove that the technology is ready or safe. It only proves that Waymo thinks it's a worthwhile gamble. This is the same place where Uber's self-driving car killed a person with no repercussions to the company.
- nmfisher 6y ago
  "Ready" or "safe" are matters of opinion, though. It's ready/safe when governments ("the people") decide it's such.
- howlgarnish 6y ago
  It's obviously ready and safe enough for Waymo to roll it out (and they've been a lot more conservative than Uber here). As for safety, all we really need is for them to be safer than the average human driver. As evinced by the ~40,000 road deaths every year in the US, that's really a pretty low bar. I'd even argue that full autonomous is/will be less risky than halfway houses like Tesla Autopilot where neither side is really in full control, but that's another story.
  - woeirua 6y ago
    That’s actually a high hurdle to clear. Looking at fatalities per hundreds of thousands of miles driven humans are insanely good at driving. Using the 2018 NHTSA number of 1.13 deaths per 100 million miles driven, avg speed of 60 mph, 20 predictions per second, that equates out to an autonomous system can make a fatal mistake on average about every 120 billion predictions.
    - sitkack 6y ago
      You make it sound like every mis-prediction could result in death.
- berdario 6y ago
  The technology is safe: https://storage.googleapis.com/sdc-prod/v1/safety-report/Waymo-Public-Road-Safety-Performance-Data.pdf https://storage.googleapis.com/sdc-prod/v1/safety-report/Way... They've been very careful about it, and they have not finished work on it, so they need to keep the same standard that they had until now. The technology is also ready for Phoenix. The company is not yet ready for the whole world (but I have no specific insight into what's holding them back). It's totally different from Uber.
2muchcoffeeman 6y ago
The obvious question is, why does this work in Phoenix? Where I live, road conditions are changing all the time. There’s new construction, new roads opening, temporary (a few hours at a B time) changes in lanes where you have to drive against the normal direction of traffic. You can call it moving goal posts, but to deploy them generally, they either have to have general intelligence, or you need to input construction schedules, so that it won’t take certain roads during certain times. And then what? You can’t order a ride to certain locations?
- howlgarnish 6y ago
  Waymo has been dealing with construction since day one. If the road is entirely blocked, Google Maps/Waze is already capable of changing routes on the fly. There will always be corner cases that automation can't deal with, but Waymo will have some human troubleshooters around for that. So do us humans, for that matter: when a car breaks down, most people call the mechanic instead of applying their own general intelligence to fixing it.
  - 2muchcoffeeman 6y ago
    What’s the mechanism? Eg A road is accessible from 5am to 8pm. Outside those times traffic conditions change. Humans setup cones and control the flow of traffic. 2 lanes becomes 1 lane for all traffic. And traffic is managed so that depending on what direction you’re going, your driving against the normal flow of traffic. This is a real scenario in a moderately busy area. Times may be wrong. The road is not blocked. But you have to take direction from a person. In fact this sort of temporary traffic change is fairly common especially due to road works.
dTal 6y ago
That's not moving the goalposts, those are the goalposts. Anyone can make a car that loops round a racetrack by itself, on a good day (well, not anyone, but it's relatively straightforward). The challenge is making it robust against "the real world".

DennisP 6y ago

Not human-level AI, though. Plenty of animals do fine with challenging navigational problems. Birds fly through forests, catch or evade other birds, cross continents, etc.