It’s worth noting that many of us consider self driving to he essentially analogous to full AI in its complexity and difficulty, and thus believe that we are literally nowhere near anything that fits this description.

No amount of “data” is going to solve the problem. Driving a car requires making a full mental model of the immediate world you’re in and creating accurate predictions of what everything else in that world is about to do, many of those things being sentient beings with whom you’re communicating through your own actions.

Nothing that Tesla is doing is getting much closer to that. The next major landmark on this timeline is a team successfully passing the Turing test, not a car moving across a parking lot.

If you share that opinion, that makes the Tesla a lethal toy.

What’s a “full mental model of the immediate world”? I’d argue that nobody has anything approaching “full”. You have “enough to get by” and that’s all computers need as well. This isn’t a full sentient AI, it’s just sensing the world and navigating a path, at an industrial scale.

No, it’s not path finding. The DARPA races were pathfinding and getting from point A to point B hasn’t been the problem for some time.

Driving around other people requires theory of mind in order to make decisions based on what the people around us expect to happen

I’m teaching a friend to drive right now - it’s all about eye contact, being waved through, stopping and starting predictably and without surprises, you’re always looking around for pedestrians and checking if they see you and predicting, according to everything you’ve learned about humans so far, whether they’re about to step into the street

Deep learning on photographs is not going to cut it.

Sure, but that's why there are armies of programmers writing rules for self driving cars. The "theory of mind" is being developed in the form of control code. It's not easy, but it is accelerated by the data collected in iterative development just like Tesla is doing.

It’s just not. How does it tell when the other car is waiting for them to move? What does it do when two people go into an alley at opposite directions and one has to back out? What happens when there aren’t any markings in the lot, or when people ignored the markings and parked in a different pattern? And so on.

These aren’t “edge cases” that “data” will sort out. These are common everyday occurrences in parking lots that require something like consciousness to navigate and involve subtle communication with other participants.

You could also argue that humans create such models in a way that is automated in confidence of predictability and familiarity. Familiarity and predictability being a byproduct of how the human chooses to make certain actions and judgements routine. Given this, humans exhibit quite a lot of overconfidence when it comes to driving: see phone use. Over time a human will make many more mistakes that any amount of paying attention would of prevented them.

If it isn’t fully functional it shouldn’t be in a consumer product which is capable of killing people when things go wrong.