On 20/03/2020 19:00, Mikel Maron wrote:
But this thread is from Facebook trying to change that. To side step imports.
No they're not. It's a couple sections in a blog post that is being wildly
misinterpreted.
It's perhaps worth remembering how we got here:
https://news.ycombinator.com/item?id=17856687
describes Facebook's initial engagement with OpenStreetMap. Apparently "
It ended up not being so bad however. The entirety of the bad edits were
reverted in a few minutes with oversight from the head of the OSM data
working group". I was part of the "bucket and shovel" reaction to the
pile of excrement that Facebook dropped on Egypt there, and those
evenings are ones that I won't get back.
It appears to be only because their previous undocumented import
attempts were reverted, and they received pushback from OSM mappers
after other edits elsewhere such as
https://forum.openstreetmap.org/viewtopic.php?id=63456 that they are
trying this new strategy.
That's not to say that using "many eyes" to sanity-check "AI"-detected
stuff and to prompt human mappers to add things that they can see but
otherwise might not is necessarily a bad thing - it isn't, provided the
human mappers are in control and familiar with the process* and likely
quality of the data** that it is suggested that they add, and familiar
with the area into which they are adding stuff so that they can avoid
adding (e.g.) walls as roads (in the Egypt example) or crop tramlines as
roads (via Rapid).
However it _is_ true that Facebook have a history of trying to "side
step imports" (actually, worse than that).
Best Regards,
Andy
(from OSM's Data Working Group, but writing here in a personal capacity)
* I've not seen it documented how anyone can understand "why xyz was
detected here" - what source data was used, what alorithm, etc.
** we've seen examples in the US where Rapid-added stuff has been
voluntarily deleted by the mappers who added it because the quality was
rubbish. Issues include offsets in the source data, poor quality of MS
building footprints in some areas in addition to the "misdetections"
already mentioned.
_______________________________________________
talk mailing list
talk@openstreetmap.org
https://lists.openstreetmap.org/listinfo/talk
