[OSM-talk] Learning to Use Machine Learning - A learn along for folks who want to be using ML in their work.

[Christoph Hormann]

As a quick reminder to any mapper who wants to use algorithmically 
generated data as a source for mapping work:

If you upload such data without manually verifying the individual 
features against local knowledge or suitable primary data you are doing 
a mechanical edit or import and must follow the rules we have for 
those:

https://wiki.openstreetmap.org/wiki/Automated_Edits_code_of_conduct
https://wiki.openstreetmap.org/wiki/Import/Guidelines

Practically this would for example be the case if - when being asked 
about the validity of your mapping by a fellow mapper - you'd be 
inclined to answer "I don't know, that's what the algorithm generated". 
You would then be in the mechanical edit/import domain.  

This is not a new topic, we have had this kind of problem in the past on 
several occasions, for example with use of automated tracing tools like 
scanaerial - which can be used both productively and responsibly for 
manual mapping as well as for doing bad quality mechanical 
edits/imports.  And in particular with algorithms advertised with the 
terms 'learning' and 'intelligence' implying human like capability and 
thereby a lack of need for human control and verification this is 
important to keep in mind.

If you are not controlling the algorithm yourself but are being given 
pre-generated data by others for the purpose of uploading it to OSM - 
with or without manual verification - you are always doing an import 
and need to follow the guidelines.

Side note:  It would be a responsible thing to include a reminder like 
what i wrote above with a message like the one i reply to here or in 
the welcome messages/FAQs etc. of dedicated communication channels.

-- 
Christoph Hormann
http://www.imagico.de/