I think people underestimate the long tail of most problems. Even if it’s 98% good, what do you do with that last 2%? The same applies for self-driving and other computer vision related technologies.
My plan when I finally purchase a self-driving car is to get 800 people from Utah to help me navigate for that last 2%