If she stepped into traffic outside a crosswalk then I'm not sure what result anyone is expecting.
Self-driving Uber car kills Arizona woman crossing street
921–930 of 1001 posts
Re: Self-driving Uber car kills Arizona woman crossing street
#922This is what's going to happen. If you've ever seen a machine learning algorithm in action, this isn't surprising at all. Basically, they'll behave as expected some well known percentage of the time. But when they don't, the result will not be just a slight deviation from the normal algorithm, but a very unexpected one. So we will have overall a much smaller number of deaths caused by self driving cars, but ones that…
Very well put. It boggles my mind that a forum full of computer programmers can look at autonomous cars and think "this is a good idea". They are either delusional and think their code is a gift to humanity or they haven't put much thought into it.
We look at the alternative which is our co-workers, and people giving us our specs, and our marketing teams and think 'putting these people in charge of a large metal box travelling at 100kmh interacting with people just like them - that is a good idea'...
It is not that we think that software is particularly good, it is that we have a VERY dim view of humanity's ability to do better.
Re: Self-driving Uber car kills Arizona woman crossing street
#923To put this accident in perspective, Uber self-driving cars totaled about 2 to 3 millions miles, while the fatality rate on US roads is approximately 1.18 deaths per 100 millions miles [1]. [1] https://www.nhtsa.gov/press-releases/usdot-releases-2016-fat...
Waymo has driven 2.7 billion miles in simulation [1]. Is that enough to give them some idea of how many deaths to expect? [1] https://waymo.com/ontheroad/
It's impossible to know. The simulation needs to be very good with thousands and thousands of all the weird situations that can occur in the real world. Without knowing how sofisticated the simulation is and if they also are using generative algorithms to try to break the autonomous system you can't even ballpark it.
Re: Self-driving Uber car kills Arizona woman crossing street
#924It's fascinating that so many brilliant people in this thread can talk so casually about any single loss of life for the tech built by the very same people in this thread. I would be devastated if I learned my code ended even one person's life. You're in a bubble if you think a world hyper connected to social media that's in the same world scared that autonomy is about to kill millions of jobs in one jab and would NO…
>Either we go out into the public roads with very high certainty that OUR software and OUR hardware won't kill someone, or we wait till we get there.
As long as it kills fewer people than would otherwise have died due to negligent human drivers I'd say it's already there.
Re: Self-driving Uber car kills Arizona woman crossing street
#925Here we go. We'll now have the first traffic fatality trial where it's not drivers-trying-drivers but people-trying-a-megacorp. (Disclaimer: I'm a bike advocate, so I may have a different perspective on some of this than most.) Our car-based transportation system is far and away the most dangerous thing any of us accept doing on a daily basis. 40,000 die a year. But when cases come to court, everyone on the jury has…
There's a current trend for people to anthropomorphise cars, with the backing of the industry (two eyes/lights and a mount/grill on the front of "him/her"). Won't the industry go into overdrive on the anthropomorphism front to try and preserve this judicial empathy?
Re: Self-driving Uber car kills Arizona woman crossing street
#926Here we go. We'll now have the first traffic fatality trial where it's not drivers-trying-drivers but people-trying-a-megacorp. (Disclaimer: I'm a bike advocate, so I may have a different perspective on some of this than most.) Our car-based transportation system is far and away the most dangerous thing any of us accept doing on a daily basis. 40,000 die a year. But when cases come to court, everyone on the jury has…
There's a current trend for people to anthropomorphise cars, with the backing of the industry (two eyes/lights and a mount/grill on the front of "him/her"). Won't the industry go into overdrive on the anthropomorphism front to try and preserve this judicial empathy?
People can't be cars, so, no.
Re: Self-driving Uber car kills Arizona woman crossing street
#927My personal hunch is that fully autonomous self-driving tech is not theoretically possible under currently known computational models, because it implies many well-known NP-Hard problems. Self-driving companies are betting on the ability to find a heuristic/approximation that works "sufficiently well". But I strongly feel that the chasm that needs to be crossed to be "sufficiently good" is not one of magnitude (i.e.…
There's problem classes and problem instances.
What does NP-hardness look like for self-driving car tech? non-deterministic polynomial in: number of objects? number of lanes? time steps in the planning horizon? action/observation branching factor? These things are bounded in practice.
Not saying that the computational problems aren't hard. But ending the conversation at "NP Hard" throws away too many nuances.
Re: Self-driving Uber car kills Arizona woman crossing street
#928The people leading the development should demonstrate that they can stop for pedestrians by personally jumping out in front of them on a closed test road. If they're not able to demonstrate this, they shouldn't be putting them on public roads.
This reminds me of a story I heard in college... an owner of a company that builds table saws demonstrated his saw's safety feature - a killswitch that pulls the blade down into the machine as soon as it detects flesh - by putting his hand on it while it was spinning.
Video: https://youtu.be/eiYoBbEZwlk
Re: Self-driving Uber car kills Arizona woman crossing street
#929My personal hunch is that fully autonomous self-driving tech is not theoretically possible under currently known computational models, because it implies many well-known NP-Hard problems. Self-driving companies are betting on the ability to find a heuristic/approximation that works "sufficiently well". But I strongly feel that the chasm that needs to be crossed to be "sufficiently good" is not one of magnitude (i.e.…
There's problem classes and problem instances. What does NP-hardness look like for self-driving car tech? non-deterministic polynomial in: number of objects? number of lanes? time steps in the planning horizon? action/observation branching factor? These things are bounded in practice. Not saying that the computational problems aren't hard. But ending the conversation at "NP Hard" throws away too many nuances.
NP in the number of data points received from all its sensors.
Re: Self-driving Uber car kills Arizona woman crossing street
#930Here we go. We'll now have the first traffic fatality trial where it's not drivers-trying-drivers but people-trying-a-megacorp. (Disclaimer: I'm a bike advocate, so I may have a different perspective on some of this than most.) Our car-based transportation system is far and away the most dangerous thing any of us accept doing on a daily basis. 40,000 die a year. But when cases come to court, everyone on the jury has…
Most drivers avoid the most serious penalties, and more importantly most victims are denied restitution, not because of sympathetic juries but because of insolvency.
The chance someone will cause a grave accident and the chance someone is insolvent are not independent variables.
You get some real horror stories in law school about people trapped in burning vehicles with no remedy for the surviving family.
https://en.m.wikipedia.org/wiki/World-Wide_Volkswagen_Corp._...
One way you can solve for this is strict products liability. Normally we ask if there's a defect in design or negligence. That produces costly litigation where a lay panel reviews a bunch of technical engineering documents with no special training, with highly varying outcomes. Instead, just have the manufacturer pay some statutory compensation whenever their product causes harm. The social harm of the thing will be borne by its purchasers.
It's hard to do that with car accidents though, because driver error contributes so much to outcomes, it's often weird to punish the company.
But we could vastly simplify the auto insurance system and improve the safety of these vehicles by forcing them to just warranty against harm once drivers are out as a factor. (Insofar as a warranty is a promise not to do harm with a fixed penalty up front based on breach.)
It would be fitting if such a statutory regime were the earliest primary laws about AI, because that would officially make a requirement that ai does not harm humans into literally the first law of robotics.