My sweet summer child.... if AVs were to rely on crowd sourced reported map corrections the results would be unreliable. There's a reason why wikipedia is mocked as a source, it can be excellent and it can vandalized.
After John Oliver called on the Internet to vandalize the Wikipedia biographies of US congressional representatives, Wikipedia's volunteer editors cleaned up the vandalism.
wikimediafoundation.org
Crowd-sourcing to train FSD is questionable in my mind (I note TMCers who think driving over 100mph could be important for 'safety' and routinely driving at 90mph is a god given and responsible right/action in the US) but at least it isn't as if tesla drivers made a point of driving badly in order to sabotage the NN training. If we ask AVs to rely more on mapping data that is crowd sourced, that crowd sourcing needs to be from disengagement triggered recorded driver reporting, with either AI or a human confirming the car's video aligns with the disengagement report from the driver and tesla making the report to the mapping software company.
I agree, reporting to Google Maps can work. In my neighbourhood, once realizing Google Maps routed beach traffic onto my no-exit street and through our apartment building's surface parking lot, our local traffic committee put in a successful concerted effort of reporting from committee members, and me asking the landlord, as property owner, to report the error to Google. Signs by the city and landlord were ignored but the volume of traffic on summer weekends on the no-exit part of the street has dropped considerably now that the route remains on city streets. I'm willing to bet, though, a concerted effort by bad-actors would reroute things back again.