You're probably right, but I feel you are arguing from too small a sample. OTOH, the sample publicly available is highly biased. I think the only reasonable choice is to say "Provisionally current SAE5 vehicles are safe enough to allow full use...but with strong continued monitoring.". And also to recognize that not all SAE5 vehicles are equivalent.

Yeah. That part of Asimov's predictions is developing.

I'm not sure it it, in general, much safer. It certainly appears to be so in this case, though. (And remember that "self-driving" is not a unitary term. There are multiple implementations out there. So it's not just that this is a special case...which every single data point is.)

OTOH, most of the stories are more about inconveniencing people rather than injuring people. So maybe is is much safer. It's hard to tell, because it definitely makes mistakes that people wouldn't make, like stopping in the middle of an intersection because the power went out.

Is that mean, median, or mode?

I accept what I believe is your point, but the expression is...less than optimal.

One story doesn't show reliability. It's just an indication. (A really GOOD indication, since the kid walked away, but just an indication.)

You slightly overstate the case for "woods", as the trees are generally at a slight distance from the road. (Of course, deer jump considerably faster than a kid does. It depends on the speed of the obstacle as well as your stopping distance inclusive of reaction time.)

Ligers or tigons are at least occasionally fertile. So the species are not completely separated. They're rare enough that the species are essentially separated, but it's a statistical argument, so there can still be gene flow.

N.B.: It's worth noting that fertile mules (or perhaps it was hinneys) are occasionally produced.

Species are only "completely separated" when the only gene flow between them is due to either human intervention or viral (or perhaps bacterial) infections. (FWIW, I've heard that the roots of the pea manufacture hemoglobin. If this is true it's probably an example of the viral transmission of genetic information.)

I'd been considering for a year or so whether I should get an Apple computer to supplement my Linux system. (Well, really to generate programs that could run on an IPad.)

This is so unpleasant that I'm even thinking MS might be a better choice...and I haven't dealt with MS since the turn of the century.

OTOH, what it's really going to turn on is which company comes up with a good voice controlled computers that can read novels to me with reasonable pacing and emphasis. But Apple seems to be lagging behind there, too. (With luck, something like that will show up for Linux.)

Let me know when you successfully crossbreed an ant and an earthworm.

Completion, here, means that the branches cannot ever produce fertile crossbreeds. In cases where there is geographic separation this often happens when an intermediate branch goes extinct.

But what constitutes a "species" isn't well defined. There are decent reasons for considering Cro-Magnon, Denisovian, and Neanderthal to be one species...but there are also decent reasons for considering them to be separate species. As long as two groups can interbreed if the happen to meet and be in the mood, you can't be sure they are separate species.

Speciation is really only visible when it's run to completion. And sometimes even then it's dubious. (Consider "ring species".)

You missed at least one finch that I've read of. Probably lots of others.

OTOH, species that carry tools around are a lot less common, and if you just quickly make a tool, use it, and then discard it, there's little benefit in refining it.

If you train it on the internet, it will be almost as reliable as the internet. (Almost. Not quite.) And if you train it to always give a responsive answer (and not say "I don't know") then it's going to generate hallucinations even when the internet doesn't lead it astray.

OK. Basically an AI can be trained to do anything that a person can do **IF THERE IS A WAY TO CHECK THE ANSWER**. If the AI can't get good feedback, it can't learn to handle the problem correctly.

This is why I think that AGI is going to require embodiment, and LOTS of sensors. And it still often won't reach the same conclusions people reach. (Consider the arguments that even men and women have, and their goals/sensoria/brains are extremely similar.)