It's an interesting idea, but I'm not sure they need to go to the trouble of crowd sourcing votes. You'd think that they could integrate data from Street View to determine, say, the number of trees, parks, or bodies of water along a route, and cross-ref that with Google Traffic data to filter by routes that have little traffic on them. Might be more difficult form a software perspective, but it would certainly be more expeditious when all is said and done.
Coding up a heuristic for things like "beauty" is hard. Training a classifier for beautiful images given enough images labeled beautiful or not would be relatively easy, but flickr tags are easier, and flickr tags worked out just as well for them in Boston as their crowdsourced voting in London.