[sudhirj]

The missing possibility is that no one wins generalized self-driving.

Self-driving is not a solvable or a winnable problem, because the problem or question isn’t defined. It’s a probabilistic space, where the world presents scenarios, and a computer must make a “correct” or “good” decision as often as possible, for the widest range of inputs.

Because there isn’t a grand unified formula for self-driving, I’m not even sure how we can talk about it being “solved”. Maybe Tesla might be able to self-drive on US or EU or first world clearly marked streets with 99.99% automation, assuming training data exists for all road signs and weather conditions? Is that what we mean by self-driving being solved? That might be possible in the next few years.

Other than that, doesn’t seem like there’s going to be much more than driver assist. Driving is a general intelligence problem, not something machine learning can solve - by definition ML can only work on problems it has many known solutions to.

[twelve40]

Why is this so complicated, to me self-driving being solved is when I get into a car to get to work and don't have to touch any controls other than the AC maybe. Pretty well-defined, no?

[xnx]

Self-driving can still be a huge value even if it isn't totally automated between every two drivable points. I'd pay a lot of money for a car that could safely drive itself on highways alone. Get on the highway, fall-asleep, wake up at a rest stop on the other side of the country. In some ways, even better than having a private jet.

[cipher_system]

It would depend on your route to work. There are also billions of routes to work and other places that should work as well.

Maybe if enough routes work that people find it worth to buy.

[6gvONxR4sf7o]

How's that a missing possibility? It's in number eight.