What does "drive better than humans" mean? Drive better in all possible scenarios? That's likely impossible now and probably forever. Maybe driver better in a majority of scenarios weighted by occurrence probability? Maybe, but we don't currently have the technology to validate this requirement.

Indeed it's interesting!

There are obviously mistakes that SDCs will make that humans never will (ignoring sci-fi shit, I'm talking about the current state of the art), and things they'll be worse at.

There's also mistakes they won't make: they never get tired, impatient or angry. They never turn around to yell at (or beat, yep I've seen that) their kids while driving. Being more widely trained they won't lack experience driving with a find misting of unexpected liquid on the roads (hi LA you suck at driving in the rain) etc. They won't text. Won't try to punish other road users. Won't get drunk or high. And so on. Well they shouldn't. If SDC companies think they can skirt or edge over the law to get a competitive advantage they will, and now you'll have law breaking at scale (well kinda like regular drivers!).

And they can (Waymo, not Tesla) have superhuman senses, e.g. lidar and radar.

But... well it can be measured to an extent: you can measure deaths per billion km, and see if SDCs do better or worse. Though that can hide a lot of things. For example if they are utterly unable to cope with unexpected situations and stall blocking the roads, then deaths caused by blocked emergency services won't be attributed correctly to them.

Certainly any comparison to the worst drivers is useless.

Depends: I would assume (though common sense often isn't so YMMV) that the worst drivers are responsible for a much higher proportion of problems than the best. So let's say the SDC apocalypse eliminate the worst X% of drivers, but also flatten the best 100-X% down to X. I'd imagine that there is some value of X where this is a net improvement.

Question is what is X. Probably middleish I'd imagine since the worst X% do the most damage.

But of course also SDCs will always make different mistakes which makes it hard to evaluate because the X% level SDC won't really be like an X% level human.

I don't think the cherry picked examples of "SDC does something dumbass" because (e.g. in the UK, where we have some of the safest roads in the world) 1600 people die per year due to motor vehicle collisions a substantial number of those are due to utterly fucking stupid people (nearly 50% are alcohol related).

It's an interesting question. That of course centres around the mechanics of SDCs driving compared to humans. There are likely to be other damaging failure modes when you have 3 companies controlling almost all cars, where for example certain kinds of traffic management become impossible because getting google to agree to code for it is impossible and they somehow won't pay fines. Or you might have the "cruise forever if you can't find a parking spot" button which will will make traffic even worse because the roads are free to use and there'll be a mechanism to abuse free resources for personal gain even further.

This seems like essentially a variant of the Waluigi effect hypothesized here https://www.lesswrong.com/post... which had the advantage of a pretty fun name for the situation, and is worth reading

I've not seen that before. It was an interesting read.

Huh I agree with you, pretty strongly, about something!

Personally I think the right to privacy stuff is a little broken. It was fine to have no particular expectation of privacy in a public place because when those laws were written the only practical way to violate it was to have 1 person essentially stalking one other, and that doesn't scale. It's really expensive, so a few people might have their privacy violated in an unreasonable way but as a population it was fine.

We now have the tech to scale cheaply, so a well funded organisation can follow huge numbers of people all the time.

Depending on your views of private corporations, that may or may not be acceptable.

However, the temptation for the government is always too great here, and if that data is available they will use it legally or not. I think the only way to stop them is to make it much harder for companies to do panopticon surveillance in the first place.

Putin was a KGB officer in the Soviet Union. That indicates he was quite the enthusiast for communism. And considers the demise of the communist system of the Soviet Union to be the greatest tragedy of the 20th century. Note the numerous armored vehicles invading Ukraine flying the communist flag of the Soviet Union.

The Soviet Union was the pre-eminent Communist state for a while. Putin considers its demise the greatest tragedy of the 20th century.

Nope. Sometimes its an objective observation. And what they are doing has always been contrary to the writer’s belief.

But they are the side with the battleships. :-)

It absolutely can make such a claim when generating code based on its training . It does not know what sources it was trained on. But to be fair the far more likely scenario would probably be it searched the internet and found some relevant code to copy.

That’s an appeal to authority fallacy. A 5 digit ID is not evidence of competence. There were plenty of low skill devs who participating in the earliest days of slashdot.

But what is the quality of that code? How easy is it to understand for the next human dev brought onto the project?

I’ve watched a coworker do as you describe. Use the AI not only for developing new code but to debug and update that code as bugs manifest. It was a bit of a mess. The AI routinely broke formerly working code. If lost the context on why it had written the code that way in the first place, and when fixing a bug in that area it failed to consider a special case it no longer was away off. Introducing a new bug. I watched fixing one bug generate many iterations of one step forward one step back. The code suffered greatly as a result.

When I assisted him on a couple of occasions I was constantly spotting things both he and the AI missed. AI needs rigorous oversight for non-boilerplate code. Its a useful tool, but it can easily be misused. Often through overuse.

Well they imagine their tech only used against their political opponents so they feel “moral”.

Well, criminal convictions are generally a public record.

Shouldn’t we solve the evil sociopath problem first, before we let them out of the digital sandbox and into the real world?

Entire replaceable, and likely improvable.

What about offsetting the economic advantage of a predatory trade practice confused you?

That seems reactive not preemptive. IE, the bans were earned.