Couldn't agree more. Especially when it comes to city driving, which would obviously be necessary for robotaxis, when AI zealots promise "it'll be here in a year or two", I always wondered "Have these people ever driven in the city?" I mean, to drive in a city, you basically:

1. Need to understand all standard signage (seems possible with AI).

2. Need to understand all "unstandard" signage (not sure how possible).

3. Need to understand the cop with the thick NY accent yelling at you saying "Can't you see there's been an accident and the road is covered with glass you dufus? Turn the F around."

I can certainly see AI solving the problem of driving in specially designed limited access highways (which could also support normal human drivers), and that alone would be a huge benefit, but I never saw how so many were willing to make the leap to "robotaxis that can drive you anywhere in the city."

4. Need to understand the cop who is directing people into lanes by jutting his chin subtly in different directions when you make eye contact with him

5. Need to understand that occluded objects have not vanished from the universe never to be seen again

6. Need to reasonably predict what that human that just made eye contact with you would likely do next, and how that's different from what he might do when he doesn’t make eye contact with you. And all of that differs if you're in NYC or SF or small town, Indiana

7. Need to constantly read other drivers' body language (or "car language"?) and infer what their intensions are. I can constantly tell what a driver is literally thinking, by picking up on their subtle motions. Drifting over slightly (they're looking to change lanes but haven't yet turned on their blinker), they're slow to respond to green lights, or acceleration of forward traffic (e.g. they're on the phone, texting, etc.) I even routinely look at the eyes of the driver in front of me by looking in their rear-view mirror (yes this entirely possible and actually helpful). Are they looking down at their phone in their lap?

8. Generally understand the true context of things you're seeing. E.g a 4x8 sheet of plywood in the road, a squashed piece of road kill, a large metal pipe, a stopped vehicle, a baby carriage, a dishwasher, a tumbling large slab of styrofoam (all things I've encountered in the last year).

9. Is there a stopped firetruck in front of me (I list this because of the ludicrous fact that a Tesla plowed full-speed into such an obstacle).

7. Need to understand that drunk person staggering along the roadside has been repeatedly slipping off the sidewalk and there's a non-zero chance they trip and fall right in front of you.

8. Need to understand that sometimes the stripes you see aren't the real stripes, you're driving near sunset and are seeing the sun reflected off of old stripes that were painted over with glossy black (wtf WSDOT).

7.5. Need to understand that the drink person staggering _inside_ the robotaxi has just thrown up in the backseat. After they are dropped off, the robotaxi cannot pick up any new customers.

This is a dance move known as the "tenderloin lurch". Well, not really, the Tenderloin Lurch is when a person staggers up to a crosswalk, waits until they have a NO WALK sign and proceeds to cross, while ignoring all the drivers who about to run them over.

I sincerely hope you are not this unpleasant in social settings.

when I float that joke in social settings in SF, everybody nods sagely

I've been in situations like this. In SF. I grew up here.

What you describe is the classic west coast non-confrontationalism. You're probably annoying enough that it's simply not worth it to counter anything you say because you just dive into a petty and self-aggrandizing argumentative mode.

Brash and confused conservatives think everyone agrees with them, but in reality people don't want to get caught up in absolute bullshit by joining any kind of interaction with them. This is a common pattern by now. You would do well to recognize it.

Even then, even if you can solve every case involving actual roads with perfect markings and intact signs and functioning signals, I’ve found myself just this week:

- driving across an unmarked grassy mound to park a car at a store in their designated area.

- paying a fee with coins to enter and exit a toll road.

- stopping to move around roadworks based solely on hand signals from one of the workers.

These aren’t even scratching the surface in terms of edge cases that could be encountered regularly.

Agreed.

People point at statements like "it just needs to drive better than the average driver" and "well lots of idiots drive".

On the flip side, lots of intelligent people with other skills do not have the confidence & comfort to drive outside of their towns local roads, or at all.

There are problem sets for ML like quant trading or phone album image tagging where you just need to be right 51% of the time to make money / be novel & useful.

We've seen plenty of examples showing ML is not appropriate for life&death problems like being the primary reviewer of diagnostic imaging for cancer screening.

The bar for ML to drive a car is much much higher, and I'm not even sure we have the right sensor suites feeding into the models, not to mention the compute capacity to operate at low enough latency.