> What am I conflating? Everything I described is unsafe. Note that I'm not talking about failing to signal on an empty street or rolling a stop sign when there's good visibility and no traffic. I'm talking about failing to signal when there are lots of other cars around who could really use that information, or rolling a stop sign too fast to tell if you need to stop, or just without bothering to look.
You weren't quite that clear about it in the first post, the rolling stop sign with good visibility was my first mental picture while reading, so I apologize for the misunderstanding. That being said, all of these maneuvers you're talking about could easily cause an accident whether the other vehicle was AI driven or not, so I don't see how it's relevant to the topic at hand: that corner cuts and slights which are technically violations but not inherently unsafe need to be predicted and handled by the self-driving car.
> It's really weird that you use driving the speed limit in the left lane as an example of dangerous rule-abiding behavior. That's against the rules in most jurisdictions, and definitely should be in any where it's not.
Again this was my mental picture, the idea of a robot driving according to the letter of the law reminds me very much of the sort of people who cruise at speed limit in the passing lane, just for different reasons.
> I agree that it's possible to break the rules safely. Rolling a stop sign in a situation with good visibility or exceeding the speed limit by a bit is perfectly fine. I agree that humans sometimes break the rules safely.
Agree.
> But I disagree with the statement that humans violate the rules in a safe and principled way. When humans break the rules, it's almost never based on any principle besides sloth or selfishness, and while humans do break the rules safely sometimes, it's almost always safe by accident.
I would assert that the times they break them safely is not always safe by accident. Every given decision in driving is a balance of risk and reward, the reward more often than not arriving at one's destination faster, and that's where the majority good decisions come from. The bad ones are often not decisions of R&R but of distractions, as you point out like texting, or worse, inebriation. I'm not arguing human drivers are better: I'm merely trying to say that the article is, I feel, attempting to demonize people for driving like people drive, and saying self-driving tech needs to be able to handle the unpredictable nature of human drivers, just as I did on my commute earlier today, if it's going to be a mainstream technology.