3 ms·
The fundamental problem of AI cars--they have a set of rules rather than common sense. They will fare far worse than humans when given impossible directions.
by LorenPechtel 2y ago
The fundamental problem of AI cars--they have a set of rules rather than common sense. They will fare far worse than humans when given impossible directions.
- shawabawa3 2y agoHuman drivers do this sort of thing all the time In fact, literally today, I drove into the oncoming lane of traffic due to construction work. There wasn't signage or anything to do it, I just saw the car in front of me doing it and did it too. No idea if they did it for any sensible reason or not
- tgaj 2y agoYeah, human drivers make mistakes all the time and we are fine with that. We allow teenagers and elderly people drive, so why we can't accept that AI-steered cars will sometimes cause an accident.
- Chyzwar 2y agoBecause teenagers and elderly can go for a trial for vehicular manslaughter. If Waymo drive into opposite traffic and killed someone, who is responsible? CEO of Waymo that knew that this could happen? If this happens multiple times, what is the value of human life? Even if Waymo become a safer driver than human, we need to answer these questions in regulation and common law. For me, it is insane how many people need to die before anyone hold Tesla accountable for its autopilot [1]. https://en.wikipedia.org/wiki/List_of_Tesla_Autopilot_crashes https://en.wikipedia.org/wiki/List_of_Tesla_Autopilot_crashe...
- FateOfNations 2y agoCriminal charges are appropriate in situations where a human (or group of humans) acts with malicious or reckless intent that harms (or could harm) the community. Self-driving vehicles reduce (but do not eliminate) opportunities for malicious or reckless acts related to vehicles. Criminal charges would remain appropriate in situations where an operator or manufacturer (and/or their employees) are malicious or reckless concerning a vehicle's design, manufacturing, or operation. It's important to note that just because someone was injured or died doesn't mean that there are malicious or reckless acts involved. This isn't a novel issue: the courts and legal system regularly deal with defective products, including vehicles, that cause injury or death. If a defective product causes damage, the injured party is typically compensated financially for their loss. They even have frameworks for determining https://en.wikipedia.org/wiki/Value_of_life https://en.wikipedia.org/wiki/Value_of_life in monetary terms.
- LorenPechtel 2y agoI'm trying to imagine an incident probably 30 years ago. Pouring rain, I can see that up ahead everyone is sharply veering right, but I can't see what is prompting that behavior. I know *something's* up, but not what. I get to the spot where everyone veers right--and see a double yellow. On my *right*. We were going straight, the road zigged (45 mph road, at the time they were only widening it from two lanes to 4 when the property was developed. While the road was "straight" it zigged back and forth because of the patchwork development), the rain was so hard we couldn't see the lane markings.
- tedunangst 2y agoA lot of human drivers could do with a bit more adherence to the rules. Maybe if you take the wrong off ramp, you continue down it and loop around instead of reversing back onto the highway, etc. I think in their current state, AI cars are a bit of a menace, but presumably they will improve over time, and I think it's evident we've reached a local maximum for humans, possibly even a decline as everyone is in ever more of a hurry, maybe let me just bend the rules a little this one time, it'll be okay.
- sharken 2y agoThe future for self driving cars are closed roads where only driverless cars are allowed. The ethical issues when a driverless car kills someone cannot be solved in a reasonable way. The question is which city/country will be the first to implement roads for driverless cars only ?
- marssaxman 2y ago> The future for self driving cars are closed roads where only driverless cars are allowed. This is the present in Honolulu, as of last year: a fleet of self driving cars, operating on a closed road where only driverless cars are allowed, transports over 3000 people per day. The service is called "Skyline".
- jhbadger 2y agoI looked it up and it seems to just be a typical driverless train, more or less the same as is found in airports for changing terminals, except that this is longer than those. It might be a good service, but not quite as exciting as your description. https://en.wikipedia.org/wiki/Skyline_(Honolulu) https://en.wikipedia.org/wiki/Skyline_(Honolulu)
- marssaxman 2y agoWell, yes: it was a tongue-in-cheek way to observe that requiring closed roads for self-driving vehicles would render the whole exercise pointless. We already have a more efficient solution for that problem, with no need for any exciting new technology!
- jamesrr39 2y agoOn a larger scale: Copenhagen and Vancouver both have fully-automated metro systems (i.e. driverless systems). Presumably there are many other cities with such systems around the world, and they probably all work nicely. Fine for getting around different areas of the cities, but it's not going to drive you wherever you want to go though.
- 2y ago
- tgaj 2y ago"Common sense" is a set of rules. I see a lot of drivers doing stupid things and I will confess that I am one of them sometimes so it's not like 'common sense" is some kind of magical human ability that we are born with. It's set of rules that we automatically acquire when driving long enough (or not if we are not observant enough) and I believe AI cars can learn that as well.
- pylua 2y agoI feel often times that common sense violates rules. For instance, driving over a double yellow to give more space to something inside of the road. Maybe it is a set of rules some of which exist outside of common law.
- LorenPechtel 2y agoExactly. The car sees the rule "don't cross the double yellow". The human sees the more general "double yellow means there is insufficient visibility for a pass". You must not put yourself in a position of having to be across it, but that doesn't mean you can't be nice to a bike in the bike lane if visibility and traffic permit.
- tgaj 2y agoWe can add the exception for the car "if there is no another car approaching and you have to overtake a bike then you can go over double yellow but you need to reduce speed" - computers are fast enough to analyze thousands of rules and exceptions every second and choose the best option.
- dzhiurgis 2y ago> They will fare far worse than humans when given impossible directions. That's why you keep roads regulated, clean and orderly.
- LorenPechtel 2y agoBut that's not always reality, especially around construction zones. Humans don't always get it right. And things sometimes take out the temporary markings even if it was done correctly. I've had to evade one of those big barrel cones rolling across the freeway. I suspect it fell off a truck rather than blew from road work, but there's no way to know. And would an AI car know that a barrel in front of it is a windblown hazard rather than an actual indication of a closed lane? And I've clobbered one of the ordinary cones when a guy was confused by the construction zone and went straight from the left turn lane and I had to evade. Didn't see where it went and didn't think of trying to put it back. Again, is the AI going to be able to realize that smacking the cone was the right thing to do? (The cone was blocking off road in preparation for what was ahead, I didn't actually go into roadwork.) Humans are far better than AIs at balancing things when faced with something having gone wrong. AI--we put in a driverless bus on a short loop. It was in an accident within hours--any human driver would have understood to get out of the way of the truck or at least lay on the horn if that wasn't possible. The bus just sat there and let itself be hit by the backing truck. If everything is as it should be they probably already are better drivers than humans. It's just they aren't as good when confronted with things that are wrong.