[skuhn]

It's not just taking traffic into account, which it can (sort of) already do.

They need to actually use the accident data. Use a diverse set of routes to get to the same place, so that there isn't one path for every car on the road. Understand stoplight patterns and where it's hard or easy to make a turn. Not focus obsessively on shortest path rather than most tolerable path.

Fundamentally what a good cab driver (which are not that common) offers you is knowledge of the best route to a destination from experience. Google Maps doesn't have that. And if it did, and it controlled a significant portion of the cars, it would no longer be the best route.

[TeMPOraL]

You know that you're describing things that machines are inherently better at than humans? And the problems you described stem mostly from the fact that maps also serve informative function (to learn the route in advance), and apparently Google didn't decide yet to compute routes in their navapp based on other active users of said navapp in the neighbourhood. But the data, infrastructure and algorithms are there.

> And if it did, and it controlled a significant portion of the cars, it would no longer be the best route.

Of course it would, because it would be able to run a global route optimization on all the cars simultaneously, thus outperforming even the most experienced human drivers by orders of magnitude. Getting above human level seems like a college-level exercise.

[skuhn]

Well then, I look forward to it happening any day now.