Experience, knowledge of the area, crashed cars, skid tracks. Eyeballs are good too.
And you think AI wont be able to any of that? It already does most of it. For example Ford already have databases of accident hot spots,
3d mapped roads and AI can currently, through mapping of terrain and correlating it with temperature data, predict where black ice might form and react accordingly.
But lets pick your comment apart a bit more shall we.
What if you're a new driver or inexperienced, are you saying you wouldn't be able to predict where black ice might be? AI would have the accumulated experience of millions of miles travveled by millions of cars over millions of hours. A human couldn't compete with that level of experience.
What if you're new to the area? It wouldn't matter to AI whether it had been to the area or not, as the system shares this knowledge with every vehicle.
A computer database will hold more information of recorded accidents than any human ever could.
AI can quite easily identify tyre marks. Pattern recognition has bee around for decades.
But AI has access to cameras many magnitudes better than your eyes and has the potential to be faster at interpreting that information and reacting to it than you as well.
My view on autonomous vehicles before I go to sleep:-
We already have hardware that can exceed every human parameter, what we're lacking is the programming conceptualisation to make it a reality, but it's getting there at an astonishing pace and it's widely recognised that widespread adoption of autonomous vehicles isn't just a possibility, it's an inevitability. And it's not even the AI that's standing in the way, it's humans and how they deal with the legislative challenges.