> these systems greatly improve some situations, but it’s not acceptable to do so at the expense of worsening others
Can you elaborate please? E.g. do you mean that such tradeoffs are not acceptable, even if they result in a net benefit?
Can you elaborate please? E.g. do you mean that such tradeoffs are not acceptable, even if they result in a net benefit?
The camera/radar/sensor things are going to "see" and react to different things than humans while driving, and while I would never left the ball of dust in the middle of the room, in plain sight, the robot would not clean the floor under the sofa any less efficiently than the other parts of the floor. And perhaps this carries over to self driving, for now it will make certain mistakes that humans never will do, but as others have mentioned here, it would also not tire, not look at cell phone, not get distracted.
So I wonder if we are treating these 5%-cases differently just because we associate "cleaning" with "not leaving dust balls clearly visible" and driving with "not crashing into pedestrians in situation x,y,z" while somewhat accepting that people in situations r,s,t do kill others in traffic to a certain extent.
It might not be strictly more capable than a single ideal human always paying attention, but that is neither here nor there when comparing if software assisted driving is better than non software assisted driving.
The easiest way to analyze this is to see if auto insurance companies offer a discount for using the software assist, as they are most directly impacted by population wide changes in the metric we are interested in (although I don’t think Tesla shares sufficient data about when and if FSD was used for auto insurers to be able to discern this).
Let me counter your collective auto collision/injury rate. Let's suppose the only injuries sustained in a year are the deaths of school children exiting a school bus. There's no rhyme or reason to why sometimes the driver assist mows down children. BUT, collectively, there are far, far fewer deaths. Say 1,000 school children are killed per year, but those are the only deaths. That's far less than the 40,000 Americans killed per year in auto crashes. So that's good, right? No. Of course not.
We want these systems to not make the same mistakes humans make and to not make mistakes humans wouldn't make. Do that, and your fatalities will decrease in an acceptable manner.
Although, note that the US government has long provided better medical care to old people (via Medicare’s higher reimbursement to healthcare providers) than to [poor] children (because Medicaid pays less).
In the 1990s, it was funny seeing my 80+ year old immigrant grandparents get tons of healthcare while my dad would tell me to play carefully because we couldn’t afford the doctor if I broke an arm or leg, or we couldn’t afford a dentist and braces (small business owner so Medicaid disqualified due to assets, yet insufficient cash flow to pay doctors).
> There's no rhyme or reason to why sometimes the driver assist mows down children.
If you are claiming a software engineer is throwing in a random kill/maim function in the driver software, then that would be worse as it could be implemented at scale (rather than individual drivers choosing to kill/maim).
Otherwise, I would classify injury caused by driver assist mechanisms as technical issues due to hardware/software, directly comparable to injury caused by human drivers due to say choosing to look at their phone or drive drunk. Or being 95 and lacking physical/cognitive capacity.
>We want these systems to not make the same mistakes humans make and to not make mistakes humans wouldn't make. Do that, and your fatalities will decrease in an acceptable manner.
I don't think anyone is going to claim 1000 kids getting mowed down by automated cars is "good", but it's far preferable to 40k people getting mowed down normally. They are however, willing to accept the deaths of 1k kids if it meant saving the lives of 40k people.
I bet they're not. People instinctively reject utilitarian trade-offs, especially when it involves children. The idea of sacrificing a few for the many might make sense on paper, but in practice, it’s emotionally and politically untenable.
More importantly, utilitarian arguments rarely persuade lawmakers. Policy decisions are driven by public sentiment, political incentives, and moral framing - not abstract cost-benefit calculations.
Comparing mortality/morbidity data is the opposite of abstract. It’s about as defined as you can get in a discussion about safety.