Interesting Links for 05-06-2021

[andrewducker]

Following a truck full of traffic lights turn the highway will really freak out your Tesla

(tags:driving automation fail )

UK approves Pfizer jab for 12 to 15-year-olds

(tags:UK vaccine pandemic children )

Looks like Scottish exam grades are going to be a clusterfuck again this year

(tags:Scotland exams Doom )

Why mathematics can never be complete or certain

(tags:mathematics video viaSwampers )

Covid variants: The Greek alphabet names explained, and which strains Alpha, Beta, Gamma and Delta are

(tags:pandemic names )

Covid patient leaves hospital after Scotland's longest ICU stay (167 days)

(tags:pandemic hospital Scotland )

The Broken Earth gets a movie deal. Author N.K. Jemisin to adapt. (Very excited about this!)

(tags:nkjemisin books movies )

19 million years ago: sharkapocalypse

(tags:sharks extinction prehistory )

Denmark parliament approves artificial island off Copenhagen

(tags:Netherlands )

Why companies supporting Pride gives LGBT people conflicting feelings

(tags:LGBT business pride )

Italian Artist Sells Nonexistent Sculpture for More Than $18,000

(tags:art funny )

Comments

[mtbc]

As an unsuspecting driver unfamiliar with London I was once briefly flummoxed by the Traffic Light Tree, back before they moved it, and have since regarded it as a public menace, up there along with the radio stations that play noises like car horns without warning. In general, I regard driving autonomously as an AI problem that remains rather beyond our reach, others' optimism notwithstanding.

[birguslatro]

Those flying traffic lights in the video could be considered an AI optical illusion, I guess. And the car didn't appear to be attempting to avoid them, which was the sensible thing to do. So while there was a glitch, it was hardly a total fail on the Tesla's part.

Autonomous vehicle effectiveness should be measured statistically and not based on some apparently inexplicable (to humans) accidents. They don't think like humans and so we shouldn't expect them to behave like humans. Do all humans, for instance, stop using cellphones while driving when they're told they're supposed to? If not, why not? And should they all be taken off the road until they've been fixed?

[andrewducker]

There are definitely things that are unforeseen, but there are multiple examples I've seen in discussions which they should be able to cope with better.

Speed signs as stickers on the rear of trucks, for instance. And stop signs likewise. A simple rule of "signs which are moving should be ignored" would be useful.

[mtbc]

Indeed. However, sometimes fairly complex and non-obvious situation awareness is required to drive well, and the car was clearly rather clueless about the situation. The current Teslas wisely don't even pretend to have the autonomy I referenced, though I can't help but be curious as to how even that one's sensor suite would cope with, say, a ladder (so not at all high after it's landed) falling off the truck ahead across the same lane (I've had to safely dodge one myself). Though, to me, the current model feels flawed anyway: if something troublesome happens and it beeps and wants the human to take over, because they weren't already driving it's hardly a safe bet that they can gain situation awareness quickly enough to handle the situation well.

But, once manufacturers get more ambition toward autonomy, it'll still be quite a challenge for them to gracefully deal with situations that I do as a driver. A couple of examples: (a) road signage that is clearly inapplicable because it's been left at the roadside by a construction crew who were careless in tidying up after their work; (b) when I don't immediately pull up to the stop line in waiting to turn onto a road because I can see there's a long vehicle also approaching the intersection from another road and it might need the room to pull onto the small road I'm on. And a fair bit of that kind of thing won't show up as easily in statistics: there's rather more to driving decently than not colliding with things.

No, an accident needn't require pulling them all off the road. Personally, I think that driving infractions by humans should be punished more severely and drivers should be trained better and tested frequently and harder. Bad driving is mostly why I've not bicycled as much as I'd have preferred. But, practically, tougher driving requirements for humans also requires investment in affordable, useful public transportation, autonomous or not, so that driving becomes less of a practical necessity for many. Sure, it makes sense to compare the statistics, but if good driving is a hard AI problem then don't give up on improving the humans too.

[birguslatro]

"The current Teslas wisely don't even pretend to have the autonomy I referenced"

FSD Beta Teslas near enough do, despite requiring the human driver to be able to take over at all times.

This owner's testing of FSD Beta is the best I've seen for showing how close (and how far away) Teslas are from being fully autonomous.

https://www.youtube.com/user/bimbels/videos

The next FSD Beta update is supposed to be totally camera-based, (no Lidar), and they're claiming it's much better than the current version, so it should be interesting to see if it is! Kim at that link will be the one to watch for an honest opinion about it.

Italian Sculpture

[danieldwilliam]

I wonder how much you could sell limited edition prints of the Iralian scuplture's certificate of authenticity for?

Re: Italian Sculpture

[andrewducker]

He should certainly at least try that!

[movingfinger]

That sculptor is on a good wheeze! Has he also sold NFTs of his invisible/nonexistent works?

[andrewducker]

If not he should!

[doubtingmichael]

I like your title for the maths talk better than the clickbaity "Math has a fatal flaw". It's worth bearing in mind that most maths is completely unaffected by these gaps. I imagine it was still pretty traumatic at the time though...

(My qualification: I am a mathematician in the same sense as Rincewind is a wizard.)

[andrewducker]

Makes you more of a mathematician than me!

And yes, the fact you can't prove everything doesn't mean there aren't an awfully large number of things you can prove.