> I think the real problem is obvious. Yes. The human brain and visual systems aren't nearly so trivial to replicate as a lot of people in the tech industry s

> I think the real problem is obvious. Yes. The human brain and visual systems aren't nearly so trivial to replicate as a lot of people in the tech industry seem to think. Tesla is just one of many case studies in the paired tech industry arrogance seen so frequently: - "A human is just a couple really crappy cameras and a neural network, we know how to do better cameras and neural networks, how hard can it be?" - "We can do anything we dream with 99.995% reliability in the synthetic, computer-based world of the internet because we know code. Therefore, we can do anything we want in the physical reality with code!" Both are far from evident in practice, but the belief in them continues, despite it being increasingly obvious to everyone else that neither one is true. Human vision and world processing is quite impressive - and, as pointed out elsewhere in this thread, a two or three year old would have no trouble working out that the obstacles were some things on a truck. I've got a nearly three year old, and I guarantee he wouldn't confuse those for stoplights in the slightest. I also wouldn't let him out on the road, though he does well enough with a little Power Wheels type toy. But there is far more going on in the visual processing system than we even understand yet, much less have the slightest clue how to replicate. And while code may be fine on the internet (where you can retry failed API calls and things mostly make sense), the quote about how fiction is constrained by what's believable and reality sees no such restrictions is very true. Out on the roads, all sorts of absolutely insane things can and do happen on a regular basis - and you can't predict or plan for all of them. But the car has to handle them or it crashes. As a random example, a year or two ago, I was behind a car that had poorly strapped a chunk of plywood to their roofrack with a good chunk hanging forward, and the front end of it was starting to oscillate awfully hard. I had a good clue that it was going to come apart sometime in the very near future, so backed off from a normal following distance to quite a way back. Sure enough, half a mile later, it failed, went flying through the air, slammed into the road a good distance behind the car, and tumbled a bit. Had I been using a normal in town following distance, it would have either hit me or tumbled into me, but using a human visual system, it was obvious that my existing following distance stood a good chance of being a bad idea. Stuff like this happens on roads constantly. Meanwhile, state of the art self driving can't tell the difference between stoplights and some poles on a truck. You'll excuse me if I don't think the problem is anywhere remotely close to solved for a general case Level 4 purpose.