[wordsarelies]

So it's funny, when they first started rolling these out folks were interested, intrigued even. Folks walking would give them deference, mostly not run in front of them if they saw them, very unlike how they treat regular drivers. But now folks are getting sick of them.

So they were modeling all their behavior and driving around cooperative pedestrians and cars (mostly) now they have to model for an adversarial public.

Good fucking luck.

[imperialdrive]

They are MUCH more aggressive now. It used to be that if you saw one, you could be 99% sure it would stop and wait for the path to be fully clear. Now, they will buzz right by you within a foot or two. Very alarming how quickly they progressed to being careless.

[JoshTko]

Google will optimize minimizing deaths/accidents vs maximizing trips/revenue. So it seems that buzzing closer to people is optimal for Google.

[what-no-tests]

That's the "AI Alignment" problem we've heard so much about recently.

[ajmurmann]

The AI alignment problem for a decade used to be that AGI doesn't end humanity as a side effect. Now that the mainstream has caught on, it's been diluted.

[what-no-tests]

Ha we don't even have "alignment" among humans. In the Kantian style of "humans are an end in themselves, not a means toward the ends of others" kind of way.

[ajmurmann]

Again, that's a much more subtle version of alignment than what the term originally was used for. It used to refer to things like the paper clip maximizer and the strawberry problem which are much worse and fundamental.

[__loam]

Do you have anything beyond anecdotal evidence to support this?

[valine]

Would be easier to change the paint job. Just disguise them as regular taxis.