Humans are actually pretty good drivers because they instinctively understand other humans.
Computers instinctively understand nothing about humans so the computer’s ability to predict what happens next is nil.
- A human is driving down a street and a bouncing ball appears rolling out into the street. As a driver, I am instantly predicting that a child will come racing after the ball even before the child “suddenly” appears.
- The left lane of traffic is slowing and stopping while the right lane is clear. As a human, I am predicting that something or someone is crossing the road and I slow down. Uber struck the car making a left turn.
- A car in front of me is slowing down and then speeding up repeatedly. As a human, my prediction is that the driver is looking for an address and I expect a sudden stop or quick turn. A computer program just sees a repeated pattern and attaches no possible future prediction to that behavior pattern.
- A human driver sees a child pulling away from his mother’s hand to race across the street and anticipates that he will have to stop. A computer doesn’t know that the child is going to be in the roadway until the child is already there.
- A human driver sees a bunch of cars at the bottom of a hill in snowy/icy weather. The human driver evaluates that the street in question is extremely icy and should not be driven on. The computer does not realize the road is icy until after starting down the hill.
- A human driver sees deer eyes reflecting the headlights and slows anticipating a herd of deer starting to cross the road from the surrounding forest. A computer sees the deer only when they are actually on the road.
Technologists keep on talking about autonomous driving as a sensor problem. The sensors are the easy part.
The examples are examples of a human driver using their ability to empathize with other humans’ motivations to predict and avoid accidents.
Self driving cars can only help in situations that are pure sensor problems… which are very, very rare.