[thrmsforbfast]

> Self driving cars will have their AI-winter time soon

I don't see this happening any time soon. There might be a contraction in highly speculative investments (i.e., VC cash) aimed at autonomous driving systems. However, even if tech never gets beyond level 2, this will still be a high-growth and high-wage field for the next couple of decades as automobile manufacturers and their tier one suppliers incorporate existing ADAS across the lineup.

And that's to say nothing of the many limited domains where level 4 is definitely doable, including closed-environment mining and manufacturing sites. As well as adjacent industries (e.g., maritime) where even levels 1 and 2 could help a lot.

Self-driving is a scorching hot fireball right now. It might cool down as people realize that level 4-5 is not happening any time soon, but a prolonged winter is hard to believe.

[coldtea]

>And that's to say nothing of the many limited domains where level 4 is definitely doable, including closed-environment mining and manufacturing sites. As well as adjacent industries (e.g., maritime) where even levels 1 and 2 could help a lot.

Yeah, I have no doubt about special purpose level 4 in limited environments. My comment was about seeing general purpose 4 and 5.

[zepearl]

I just had a look for the first time at the description of levels 4 and 5 here (https://en.wikipedia.org/wiki/Self-driving_car#Levels_of_dri...).

The definition of level 5 being "under all roadway and environmental conditions that can be managed by a human driver" means that whatever mechanism being used would have a better "brain" (at least in the area of "driving") than the human one, correct?

If yes: kind of scary, as such capability would probably have to include the comprehension/understanding of e.g. "context", the ability to "abstract", be able to generalize from indirecly related informations, etc.. to understand potential upcoming dangers.

For example when driving behind a truck which has its trailer that wobbles continuously left and right (but which can still keep its lane) because of a partially flat tire would probably need a very advanced AI to recognize the potential danger.

[krapp]

Yes... the problem of autonomous cars is likely that of strong, general purpose AI.

Unfortunately, the politics around autonomous cars insists that human beings are basically idiots behind the wheel, so the problem seems more trivial than it is. After all, how hard can it be to design an AI smarter than an idiot?

[coldtea]

>For example when driving behind a truck which has its trailer that wobbles continuously left and right (but which can still keep its lane) because of a partially flat tire would probably need a very advanced AI to recognize the potential danger.

Or, say, driving in Mumbai or Beijing.