3 ms·

The wright brothers made the first heavier than air human flight 113 years ago. Look what we can do now! First mobile phone call was 43 years ago and now we hav

by pmx 11y ago

The wright brothers made the first heavier than air human flight 113 years ago. Look what we can do now! First mobile phone call was 43 years ago and now we have such advanced tech that making calls is almost an afterthought. Once tech gets a grip its momentum is huge.

gambiting 11y ago
40 years ago we thought(well,some of us did) that accurate image recognition is a matter of few months of work, at most. It's 2016, and our most advanced image recognition software can't tell a zebra and a sofa in a zebra print apart. It's an insanely difficult problem. We can't even do voice recognition completely right. Automatic cars rely on that exact type of challenge - recognising patters accurately. I'm sure in certain settings they work exceedingly well. Spectacularly even. But to be safe for human transportation, they need to be 100% accurate. Not 90% or 95% accurate. They need to work in snow, rain, when a huge sinkhole appears in the ground or an infant walks into the road. They need to consider legal, moral and ethical implications lest they be allowed on the road - is it allowed to hit a pedestrian to save 4 passengers? Is it allowed to hit 4 pedestrians to save 1 passenger? Can you at any time take manual control? If yes, how do you regulate insurance? If no, how do you tell the car that you want it in that particular spot and not any other, be it your garage or a middle of a clover field? Yes, there already is software that can drive safely on lit, dry roads, while there is a human behind a steering wheel tracking its every move. But I still believe we are at the very least a few decades away from fully autonomous vehicles.
- JoeAltmaier 11y ago
  Its not at all clear that the bar is that high. Humans that fail to meet that bar spectacularly are let on the road every day by the millions. Its more an economic issue than a moral/technical one.
  - gambiting 11y ago
    I actually met plenty of people who say "oh, they just need to have less accidents than humans and we are good". No, that's absolutely not true. For the most recent example - someone made a raspberry Pi controlled insulin pump. Insulin is actually incredibly dangerous to humans if you get the dose wrong, so making an insulin pump based on hardware that does not conform with highest safety standards is just not acceptable. You know what the person behind it said when it was pointed out to them? That it doesn't matter, because raspberry Pi is still going to kill less people than the number of those who die through incorrect injections due to tiredness or simple mistakes. That's absolutely incorrect - even if such machine lowered the overall number of deaths due to incorrect insulin injections, no one would ever allow it on any market ever. No company would ever get its way out of "poor hardware choice lead to death of Mr. Smith" by saying "hey, but actually, our machine kills less people than would die naturally due to similar causes each year, so you can't hold us accountable, right??". To me, it's the same with automatic cars - they cannot merely "have less accidents than humans". They need to have 0 accidents or they won't be acceptable. That's why the bar is high. If you are in a situation where a choice is between hitting a pedestrian or running under a semi and possibly killing everyone in your car, no one is going to blame you for doing either - our primitive brains probably are going to go with whatever seems most logical at the moment, you can blame anything on adrenaline. Computers don't have that luxury. They need to make a calculated choice - and then whoever makes them(the computers) has to live with that choice. The computer chose to hit the pedestrian - now the company who wrote its code is being sued for millions - no matter how they frame it, that's not a situation anyone wants to be in. Of course I'm going off into theoreticals here, since we don't actually have this problem yet. But I am sure it will become an actual problem and it will need to be solved one way or another before widespread adoption.
  - sandworm101 11y ago
    And when those humans fail, we punish them. They pay fines or have their license suspended. When they are really bad, we lock them up. I'd be all for Teslas if, should they every brake a law or hurt someone, the car was physically locked in a box for a year or two.

by pmx 11y ago

gambiting 11y ago

40 years ago we thought(well,some of us did) that accurate image recognition is a matter of few months of work, at most. It's 2016, and our most advanced image recognition software can't tell a zebra and a sofa in a zebra print apart. It's an insanely difficult problem. We can't even do voice recognition completely right. Automatic cars rely on that exact type of challenge - recognising patters accurately. I'm sure in certain settings they work exceedingly well. Spectacularly even. But to be safe for human transportation, they need to be 100% accurate. Not 90% or 95% accurate. They need to work in snow, rain, when a huge sinkhole appears in the ground or an infant walks into the road. They need to consider legal, moral and ethical implications lest they be allowed on the road - is it allowed to hit a pedestrian to save 4 passengers? Is it allowed to hit 4 pedestrians to save 1 passenger? Can you at any time take manual control? If yes, how do you regulate insurance? If no, how do you tell the car that you want it in that particular spot and not any other, be it your garage or a middle of a clover field? Yes, there already is software that can drive safely on lit, dry roads, while there is a human behind a steering wheel tracking its every move. But I still believe we are at the very least a few decades away from fully autonomous vehicles.

JoeAltmaier 11y ago
Its not at all clear that the bar is that high. Humans that fail to meet that bar spectacularly are let on the road every day by the millions. Its more an economic issue than a moral/technical one.
- gambiting 11y ago
  I actually met plenty of people who say "oh, they just need to have less accidents than humans and we are good". No, that's absolutely not true. For the most recent example - someone made a raspberry Pi controlled insulin pump. Insulin is actually incredibly dangerous to humans if you get the dose wrong, so making an insulin pump based on hardware that does not conform with highest safety standards is just not acceptable. You know what the person behind it said when it was pointed out to them? That it doesn't matter, because raspberry Pi is still going to kill less people than the number of those who die through incorrect injections due to tiredness or simple mistakes. That's absolutely incorrect - even if such machine lowered the overall number of deaths due to incorrect insulin injections, no one would ever allow it on any market ever. No company would ever get its way out of "poor hardware choice lead to death of Mr. Smith" by saying "hey, but actually, our machine kills less people than would die naturally due to similar causes each year, so you can't hold us accountable, right??". To me, it's the same with automatic cars - they cannot merely "have less accidents than humans". They need to have 0 accidents or they won't be acceptable. That's why the bar is high. If you are in a situation where a choice is between hitting a pedestrian or running under a semi and possibly killing everyone in your car, no one is going to blame you for doing either - our primitive brains probably are going to go with whatever seems most logical at the moment, you can blame anything on adrenaline. Computers don't have that luxury. They need to make a calculated choice - and then whoever makes them(the computers) has to live with that choice. The computer chose to hit the pedestrian - now the company who wrote its code is being sued for millions - no matter how they frame it, that's not a situation anyone wants to be in. Of course I'm going off into theoreticals here, since we don't actually have this problem yet. But I am sure it will become an actual problem and it will need to be solved one way or another before widespread adoption.
- sandworm101 11y ago
  And when those humans fail, we punish them. They pay fines or have their license suspended. When they are really bad, we lock them up. I'd be all for Teslas if, should they every brake a law or hurt someone, the car was physically locked in a box for a year or two.