6 ms·

Yeah, that's a bit of an overstatement. We created nukes, landed on the moon, took sludge out of the ground and used it to power the world. We connected this w

Yeah, that's a bit of an overstatement. We created nukes, landed on the moon, took sludge out of the ground and used it to power the world. We connected this world with wires and glass fibers to build a real-time global communication system that also gathered all of the world's information in a singular and immediately accessible place. Building a self driving car is hard but really not as tough. How quickly do think we'd get self driving cars if the USA spent 4% of the federal budget on it like NASA received in the 60s. (That's about $40B a year for a decade).

paxys 7y ago
All great achievements, but none of these is in the same ballpark of difficulty as creating a general artificial intelligence (which is pretty much what self driving cars need to achieve true level 5 autonomy).
- dumbfoundded 7y ago
  Level 5 driving doesn't require AGI at all. It's an extremely narrow set of problems that have to be solved incredibly well. It mostly just comes down to creating an accurate 3d representation of the world from a bunch of sensors. You also have to correctly segment and label each object in that 3d representation. If you did those two things extremely well, the actual driving logic can be hardcoded. The problem is that each of these systems has problems so they all have to improve and compensate for each other.
  - Judgmentality 7y ago
    This is incorrect. The hardest part of developing a self-driving car is predicting the world around you in the immediate future. Knowing whether or not that object is a person is a lot easier than guessing whether or not that person is going to jump out in front of the car 1 second into the future. You have to know who is going to run stop signs, when cyclists are about to cut you off, when someone is about to back up into a parking spot. I don't know whether or not AGI needs to be developed to make a useful self-driving car, but as time goes on I'm beginning to believe that's the case.
    - outworlder 7y ago
      You can probabilistically predict those events with machines much better than humans can. You don't really know someone will decide to run a stop sign, but you do know when the vehicle is past the point it was supposed to start slowing down. That's relatively "easy", we have been predicting physical object motions with analog computers, even. As the parent says, accurate data from sensors is a much bigger problem. But once you have the data, you can model these objects with dumb algorithms. Computers can also have a much faster reaction time, so a human may need to predict one second ahead, but computers may be able to get away with less.
      - dymk 7y ago
        > You can probabilistically predict those events with machines much better than humans can. This is an assumption and has not been shown to be correct or even probable
    - whatshisface 7y ago
      I question the claim that you need to predict the immediate future. Human reaction times are pretty slow, to the point where when our feet hit a pedal in response to light that hit our eyes two seconds ago, something that is prediction for us could be reaction for a machine. A human has to live two seconds in the future because our appendages and lower faculties are lagging behind in the past.
      - notahacker 7y ago
        > I question the claim that you need to predict the immediate future. Alternatively you could develop braking technology which gives vehicles a stopping distance of 0m, but this might be a bigger technological advance than full self-driving AI, and I'm not sure it would be that comfortable for the passengers....
    - semi-extrinsic 7y ago
      For me, the definite proof that we won't have self-driving anytime soon is the massive fail that was encountered in the recent chatbot fad. One key part of driving is communicating - with pedestrians, cyclists, other drivers. This happens through body language and other fairly subtle cues. When you can't make AI work for responding to questions given in text form on an extremely limited problem domain, how on earth would it work for something that's orders of magnitude less well defined and more broad?
  - SmooL 7y ago
    "If you did those two things extremely well, the actual driving logic can be hardcoded." I mean, it _could_ be hardcoded, but there's millions of edge cases so it's pretty infeasible. I agree with parent comment that full level 5 requires something close to AGI - the difficult part in in getting a self driving car is giving the AI something along the lines of "common sense", the ability to reason about what to do in an unfamiliar situation.
  - paxys 7y ago
    Let's say you do sense and label the entire world (which in itself is an impossible problem). Do you think every single action that a driver takes on the road can be hardcoded? What happens when a street is temporarily closed but doesn't have the correct signage? What if there's a police office or road worker signalling instructions by waving his hands? What if the lights at an intersection stop working? What happens if there's a car burning on the side of the highway and drivers need to change lanes to go around it? And these are just some of the problems in a large American city. Think about rural areas, places with more aggressive traffic, places with wildly different written and unwritten traffic rules.
  - paganel 7y ago
    > Level 5 driving doesn't require AGI at all. > It's an extremely narrow set of problems that have to be solved incredibly well. It does require AGI, at least if you're planning to drive on most of the world's roads, and not only on some "pampered" streets in the middle of the desert or on heavily-regulated and very well maintained streets like in Norway or Switzerland. As a human a I have a quite "accurate 3d representation of the world" but even so, many times I'm left dumb-founded by what the people driving on the same streets as me are doing. And even if you do manage to replace all those other people with self-driven cars, how do you account of cows ending up in the middle of an interstate road (it happened to me at least once), with wrong street markings or no street markings at all or with drunken bicyclists whom you can't see at night?
gowld 7y ago
Waymo has a far better safety records than NASA, and it self driving cars. NASA is allowed to kill people. Waymo isn't. The Internet is allowed to be full of malware. Waymo isn't.
obernaru 7y ago
Depends what you mean by "self driving car". IMO, building a fully self driving car requires artificial general intelligence, which is so hard I'm not sure humanity will ever achieve it. If we ever manage to create the kind of AI that fully self driving cars require, self driving cars will be one of the most boring and trivial things that is done with that AI.
foobiekr 7y ago
We won’t know how tough it is until it gets done once. I don’t think we’ve demonstrated anything autonomous beyond the most trivial kinds of autonomy (e.g. the V2) in all of our technology history.
erfgh 7y ago
It's way tougher.
alkonaut 7y ago
My guess is that it (full 100% self driving) has around the same technological difficulty as putting a man on Mars. It also depends on what constitutes "100%" of course. > How quickly do think we'd get self driving cars if the USA spent 4% of the federal budget on it like NASA received in the 60s. (That's about $40B a year for a decade). At that cost we could adapt all infrastructure to suit self driving cars, instead of developing self driving cars to adapt to human infrastructure. But I think that kind of cost is always going to be beyond what's acceptable. I think the discussion is mostly pointless because of diminishing returns: if you can have "99.9% full self driving" for a tiny fraction of the cost, who would want to pay to go from 99.9% to 100%? Initially human remote drivers will take care of the rest. And then there is a very slow commercial race towards using fewer humans that drives the very slow march to 99.9% and 99.99% self driving and so on. Driving the last second of the last edge case route is basically something that requires AGI (as long as we don't adapt infrastructure).
- Miraste 7y ago
  I agree with most of this, but remote drivers are completely unfeasible with current or planned network infrastructure. You could drive with acceptable levels of sensor bandwidth and latency right next to an unobstructed low-utilization 5G tower, and that's about it. That's unlikely to correspond with the locations one would need remote drivers.
  - alkonaut 7y ago
    Luckily the hardest situations at least occur in cities and not on highways, and cities have good broadband. But yes, it's still "100%, but only here" a.k.a. not 100% I think remote drivers will probably have to "rescue" cars without piloting them, usually just assessing a situation and overriding something (driving through an obstacle etc). A passenger (if there is one) could do the same. But sometimes actual remote driving would be required of course.

paxys 7y ago

All great achievements, but none of these is in the same ballpark of difficulty as creating a general artificial intelligence (which is pretty much what self driving cars need to achieve true level 5 autonomy).

dumbfoundded 7y ago
Level 5 driving doesn't require AGI at all. It's an extremely narrow set of problems that have to be solved incredibly well. It mostly just comes down to creating an accurate 3d representation of the world from a bunch of sensors. You also have to correctly segment and label each object in that 3d representation. If you did those two things extremely well, the actual driving logic can be hardcoded. The problem is that each of these systems has problems so they all have to improve and compensate for each other.
- Judgmentality 7y ago
  This is incorrect. The hardest part of developing a self-driving car is predicting the world around you in the immediate future. Knowing whether or not that object is a person is a lot easier than guessing whether or not that person is going to jump out in front of the car 1 second into the future. You have to know who is going to run stop signs, when cyclists are about to cut you off, when someone is about to back up into a parking spot. I don't know whether or not AGI needs to be developed to make a useful self-driving car, but as time goes on I'm beginning to believe that's the case.
  - outworlder 7y ago
    You can probabilistically predict those events with machines much better than humans can. You don't really know someone will decide to run a stop sign, but you do know when the vehicle is past the point it was supposed to start slowing down. That's relatively "easy", we have been predicting physical object motions with analog computers, even. As the parent says, accurate data from sensors is a much bigger problem. But once you have the data, you can model these objects with dumb algorithms. Computers can also have a much faster reaction time, so a human may need to predict one second ahead, but computers may be able to get away with less.
    - dymk 7y ago
      > You can probabilistically predict those events with machines much better than humans can. This is an assumption and has not been shown to be correct or even probable
  - whatshisface 7y ago
    I question the claim that you need to predict the immediate future. Human reaction times are pretty slow, to the point where when our feet hit a pedal in response to light that hit our eyes two seconds ago, something that is prediction for us could be reaction for a machine. A human has to live two seconds in the future because our appendages and lower faculties are lagging behind in the past.
    - notahacker 7y ago
      > I question the claim that you need to predict the immediate future. Alternatively you could develop braking technology which gives vehicles a stopping distance of 0m, but this might be a bigger technological advance than full self-driving AI, and I'm not sure it would be that comfortable for the passengers....
  - semi-extrinsic 7y ago
    For me, the definite proof that we won't have self-driving anytime soon is the massive fail that was encountered in the recent chatbot fad. One key part of driving is communicating - with pedestrians, cyclists, other drivers. This happens through body language and other fairly subtle cues. When you can't make AI work for responding to questions given in text form on an extremely limited problem domain, how on earth would it work for something that's orders of magnitude less well defined and more broad?
- SmooL 7y ago
  "If you did those two things extremely well, the actual driving logic can be hardcoded." I mean, it _could_ be hardcoded, but there's millions of edge cases so it's pretty infeasible. I agree with parent comment that full level 5 requires something close to AGI - the difficult part in in getting a self driving car is giving the AI something along the lines of "common sense", the ability to reason about what to do in an unfamiliar situation.
- paxys 7y ago
  Let's say you do sense and label the entire world (which in itself is an impossible problem). Do you think every single action that a driver takes on the road can be hardcoded? What happens when a street is temporarily closed but doesn't have the correct signage? What if there's a police office or road worker signalling instructions by waving his hands? What if the lights at an intersection stop working? What happens if there's a car burning on the side of the highway and drivers need to change lanes to go around it? And these are just some of the problems in a large American city. Think about rural areas, places with more aggressive traffic, places with wildly different written and unwritten traffic rules.
- paganel 7y ago
  > Level 5 driving doesn't require AGI at all. > It's an extremely narrow set of problems that have to be solved incredibly well. It does require AGI, at least if you're planning to drive on most of the world's roads, and not only on some "pampered" streets in the middle of the desert or on heavily-regulated and very well maintained streets like in Norway or Switzerland. As a human a I have a quite "accurate 3d representation of the world" but even so, many times I'm left dumb-founded by what the people driving on the same streets as me are doing. And even if you do manage to replace all those other people with self-driven cars, how do you account of cows ending up in the middle of an interstate road (it happened to me at least once), with wrong street markings or no street markings at all or with drunken bicyclists whom you can't see at night?

gowld 7y ago

Waymo has a far better safety records than NASA, and it self driving cars. NASA is allowed to kill people. Waymo isn't. The Internet is allowed to be full of malware. Waymo isn't.

obernaru 7y ago

Depends what you mean by "self driving car". IMO, building a fully self driving car requires artificial general intelligence, which is so hard I'm not sure humanity will ever achieve it. If we ever manage to create the kind of AI that fully self driving cars require, self driving cars will be one of the most boring and trivial things that is done with that AI.

foobiekr 7y ago

We won’t know how tough it is until it gets done once. I don’t think we’ve demonstrated anything autonomous beyond the most trivial kinds of autonomy (e.g. the V2) in all of our technology history.

erfgh 7y ago

It's way tougher.

alkonaut 7y ago

My guess is that it (full 100% self driving) has around the same technological difficulty as putting a man on Mars. It also depends on what constitutes "100%" of course. > How quickly do think we'd get self driving cars if the USA spent 4% of the federal budget on it like NASA received in the 60s. (That's about $40B a year for a decade). At that cost we could adapt all infrastructure to suit self driving cars, instead of developing self driving cars to adapt to human infrastructure. But I think that kind of cost is always going to be beyond what's acceptable. I think the discussion is mostly pointless because of diminishing returns: if you can have "99.9% full self driving" for a tiny fraction of the cost, who would want to pay to go from 99.9% to 100%? Initially human remote drivers will take care of the rest. And then there is a very slow commercial race towards using fewer humans that drives the very slow march to 99.9% and 99.99% self driving and so on. Driving the last second of the last edge case route is basically something that requires AGI (as long as we don't adapt infrastructure).

Miraste 7y ago
I agree with most of this, but remote drivers are completely unfeasible with current or planned network infrastructure. You could drive with acceptable levels of sensor bandwidth and latency right next to an unobstructed low-utilization 5G tower, and that's about it. That's unlikely to correspond with the locations one would need remote drivers.
- alkonaut 7y ago
  Luckily the hardest situations at least occur in cities and not on highways, and cities have good broadband. But yes, it's still "100%, but only here" a.k.a. not 100% I think remote drivers will probably have to "rescue" cars without piloting them, usually just assessing a situation and overriding something (driving through an obstacle etc). A passenger (if there is one) could do the same. But sometimes actual remote driving would be required of course.