Alexis Madrigal has written an interesting look at how Google’s self-driving car really works. Google has figured out that rather than building a really smart car, they could instead build a really rich digital representation of the roads on which the car will drive. Obviously there are big questions about whether this approach can be scaled to work for a larger geographic area than Mountain View, California, but I love this approach, which only software engineer would come up with. I agree with the article that collecting and storing large amounts of this kind of data is a problem we understand better than the problem of building really intelligent machines. If this approach to controlling self-driving cars takes off, it also puts Google in a great position to make money licensing data to any company that wants to build them, rather than building cars itself.