We expect human drivers to attempt to avoid accidents which are not their fault (see: legal duty to lookout/avoid), why shouldn't we do the same for non-human drivers? To not do this seems like a step backwards in safety expectations.
I was going to add that cones are just as important for human drivers but I thought it was obvious. I don't think human drivers can detect wet concrete (or other hidden hazards) any better than machines.
These sorts of safety issues have been a always been a problem, with lax markings, etc, but now are more acute with non-human drivers.
I think ultimately it raises the question of whether it's reasonable to offer some accommodations to non-human drivers where they don't perform so well. Do they have to be at the standard of human drivers for everything? That's probably impossible.
Why? The entire argument for self driving cars is that they are safer drivers than humans. If that claim breaks down whenever the driving conditions are anything other than perfect, then self driving cars lose their only supposed benefit. If we have to substantially change everything we do on or around the road to simply help them match human efficacy then (1) what is the point of a "self driving" car that can trivially encounter breaking conditions and (2) if these changes would be beneficial safety why haven't we already done them for human drivers?
Also the picture in the article shows both cones and workers, the presence of either would generally cause a human to alter how they drive.
What we see consistently is self driving cars that randomly stop in perfectly reasonable driving conditions, and similarly we see repeated cases of them driving through clearly unsafe areas. If a single human driver repeatedly made these same mistakes they would lose their license, but for some reason we allow it from these unsafe self driving systems. If a car manufacturer ships a car that is found to operate unsafely, the manufacturer is required to recall them, and isn't allowed to return the car to market until the problem is demonstrably fixed.
Ok, on what grounds is technology that performs a job that humans can generally do better at beneficial in the broad sense? That it's benefitting a handful of individuals who profit off eliminating jobs isn't in question, but where's the upside for the rest of us?
My mom (who is legally blind) wouldn't have to resort to driving herself to her doctors appointments when myself and my brother are unable to due to work.
For me it's safety (each problem solved gets applied to others, but it could be more profound than that, like chess AIs) and potential to decide on the fly whether to have something similar to a rental car or a much lower rate. For instance maybe I could know a car would be waiting for me and leave my backpack behind while I go on a hike. Plus environmental and city planning hopes.
Current performance definitely does not suggest otherwise! Over Waymo's first million miles, it was involved in two collisions, one of them at-fault. The average for human drivers is 5.3 collisions per million miles.
So you're saying a handful of randos having a little spare time to catch up on Reddit during their commute is a net plus trade for putting mediocre operators on the road? That's definitely a take.
Properly marking a construction zone isn't a change in practice - it's already required. But that doesn't mean it was done properly in this case.
If a construction zone is unmarked, human drivers are more likely to drive into it as well.
Anyway, AI cars should be smart enough to "see" a potential unmarked construction zone, just as most humans can. I'm only saying construction zones should be marked explicitly for the safety of all involved.