I am not sure either. The authors (more correctly, the author of the review of the the parent paper) imply that the constant challenges to adapt to complex spatial navigational tasks is the most likely explanation. Familiarity with routes is not the key variable (no bus driver benefit). It is active navigational adjustment or perhaps just ascertainment. Controlling for ascertainment bias would be possible but hard because you would face secular confounds (e,g., impact of COVID-19 on taxi drivers vs other professions).
By the way, the original l”London taxi drivers have larger hippocampii paper” was NOT replicated. So sad since this is such a fun story.
From a programmer's and former taxi driver's perspectice, finding a fuzzy path through something chaotic and shifting like a city optimizes for very different logic than finding a path through a maze or a jam of vehicles. It's reducing from a big picture versus trying to iterate solutions from a small picture. And I can't tell you how many times I get in a lyft these days and know better than the GPS how to get around a traffic snarl, because pathfinding generally doesn't consider what might happen to traffic 30 minutes from now given the current conditions (if everyone is headed northeast to a concert, for example).
By the way, the original l”London taxi drivers have larger hippocampii paper” was NOT replicated. So sad since this is such a fun story.