undefined | Better HN

0 pointsBarrin923y ago0 comments

>Any reason why everyone seems to be stuck on this problem?

ML maximalism focused on the narrow problem of 'solving driving' while not recognizing that any task as complex as driving requires probably something closer to general intelligence, and theoretically the field has been impoverished in favor of "throw more graphics cards at everything".

0 comments

5 comments · 3 top-level

hn_throwaway_993y ago· 2 in thread

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."

dmd3y ago

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

naijaboiler3y ago

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

2 more replies

highwaylights3y ago

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.

steveBK1233y ago

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.

j / k navigate · click thread line to collapse