I fear it will end up more like nuclear energy: a clearly better, objectively safer technology that nevertheless is very easy to fear-monger against. Development won’t be pushed to the Third World, but rather those political systems less susceptible to being swayed by nonsense public pressure like that.

I feel like nuclear energy is a bad example, because when it goes wrong it's "one big scary thing" as opposed to the papercuts that would add up with dangerous robotaxis. Also the benchmark for car danger is already so incredible low. You can see in this thread the argument that "cars are already killing people, so we as a society just have to decide how many it's ok to have die while we figure out to make them work."

As far as "objectively safer" - not sure how that works (or matters). Since when has safety actually mattered in terms of public opinion? If safety mattered we wouldn't have cars on city streets at all, all last mile would be slow & public and we'd have built our towns & cities to support that.

The road conditions are not ideal though, poor road markings, road conditions, other drivers not following the rules of the road. Hardly a place to test cars.

Counter take is that it's the perfect environment from a learning perspective: if you succeed there, you can succeed in easier conditions.

Not really, hectic traffic and nonexistent roads aren't a superset of organized traffic and good roads. If you take a human driver from a country with hectic traffic and put them in one with organized traffic they'll struggle and vice versa, the same would hold here. Being skilled in one doesn't really help you out in the other.

They're both a superset of the problem space: safely navigate from A to B.

And broadly speaking, with more varied training data you can achieve better out of sample results.

This sounds like your opinion and doesn’t seem accurate to me at all.

I think the parent is right. Every country has different traffic laws anyway - but driving in a developing country is often fundamentally different.

In Vietnam the effective priority of vehicles is "biggest first". A neural net trained in that environment won't be very useful in the UK, where priority is defined by road markings and street signs.

Absolutely, when they need to test on hard-mode.

I agree. The hard part of autonomous driving is when people and things behave unpredictably and there aren't many context clues.

However, as someone who regularly visits and drives in a 3rd world country, there's a "vibe" to it if you will, and once you learn the vibe it becomes natural. (It seems to be that people respect the laws of physics over the laws of the land, if you will) If you train the cars on a 3rd world country's driving patterns, further refinement will be necessary to fit in better in a more developed country. But early on it will deal with stuff like driving on a road with no lines at the edges or between lanes, one lane bridges, sharing the road with pedestrians, cyclists, motorcyclists, and animals, etc.

I'm not sure this holds up to scrutiny. So far we have numerous companies testing autonomous vehicles in the states and personally I don't know any that test in developing countries.

I'm not saying it's happening now, but a shift might occur.

You're correct when you say

>we have numerous companies testing autonomous vehicles in the states

The article we are responding is is about increasing political pressure against that practice.

thanks for sharing that messed up thought with us I guess

Companies are already testing self-driving cars in SF, Phoenix, and LA. So much for this theory.

Yes, and the article to which we're replying is a story about political pressure to STOP that testing.