So if currently the recommended route is simply a function of journey time, could a map have a more sophisticated utility function that takes into account other variables like number of connections, mode of transport and pollution levels etc. Initially the weights of each variable would be unknown (100% to minimising transit time as a sane default) but over time would infer the weights by offering you a range of choices and learning from your preferred selections.
One major flaw with this is that perhaps the map system relies on caching precomputed partial solutions which would be impossible if, in general, every user has slightly different preferences. I've noticed that (AFAICT) Google maps doesn't take into account individual walking speeds when calculating routes for example.