Computer programs are just human-dictated logic, so we can program the same decision-making processes into computers. A logical way to do this would be to train

Computer programs are just human-dictated logic, so we can program the same decision-making processes into computers. A logical way to do this would be to train mapping computers the same way we do now, by allowing them to learn from human drivers. Just like how Google hires drivers with cameras to take pictures of streets, we can start with semi-autonomous cars which compare their programmed decisions with the decisions of the drivers and analyze differences. Afterward, they could ask the driver questions about why their behavior was different from what was expected in a programmatic way. Trends could be analyzed, and suggestions for altering the driving logic could be incorporated by programmers.