Those cameras are backed by processing that's so far ahead of anything silicon can do it's not even funny. It's not inherently infeasible, but it seems computers will need a huge leap in certain capabilities before they will be able to act as a drop-in replacement for human drivers.

I find that argument being thrown around once in a while, but let me ask you this. Why are we spending billions of dollars in research that we can vaguely hope to reach human perception levels in one distant day. This whole thing is just not worth it. That's why I said especially for dense urban traffic driving, we need extra sensing, and need from the environment so it can help the car. Imagine if the car knew there is nothing coming from a junction we don't need to stop all the way etc.

I don’t object to providing extra data to self driving cars, I’m just a teeny bit confused (as I think many people outside of ML are) about why certain problems are much harder than others for AI.

I think there is more senses being used in driving then just eyes and ears. I am not the best driver in the world, but I use a lot of senses while driving, abnormal vibrations that are felt cause me to be more alert. Constant awareness of hills and road conditions are felt.

Saying it's just cameras and microphones vastly over simplifies what driving involves for a human.

It's interesting that you talk about using more senses while driving. I always crack the window to listen to the road noise as it helps me guage what the other cars are doing and has helped several times in avoiding collisions with other drivers that weren't paying attention since I could hear the change in noise within the standard noise of the road.