[OSM-talk] Attribution guideline status update

[Christoph Hormann]

I am strongly against this in the current form because it addresses none 
of the major issues about corporate attribution of OSM (or lack 
thereof).

1) It does not in any way address the problem of second rate attribution 
(i.e. someone else - usually the service provider of the map service or 
the media outlet publishing the map) is being attributed more 
prominently than OSM.  The '50 percent rule' you invented:

"If OpenStreetMap data accounts for a minority (less than 50%) part of 
the visible map rendering, attribution with other sources on a separate 
page that is visible after user interaction is acceptable."

is ridiculous because 50 percent of the map area being functionally 
empty is essentially a property of most maps, in particular at large 
scales or high zoom levels.  There is no basis in the ODbL for allowing 
attribution in a case where attribution is required that is 
not "reasonably calculated to make any person [...] aware".  Therefore 
i would consider that rule in clear violation of the license.

And frankly it also contradicts the fundamental self-image of the mapper 
community.  As has been discussed plenty of times the way geodata is 
generated in OSM is fundamentally different from other geodata sources.  
While elsewhere people generating geodata are almost always rewarded 
for their work also in other form (like salery) in OSM the only 
recognition mappers receive from external data users is the attribution 
required by the license.  Putting OSM on the same level as other data 
providers like you do above is totally inappropriate.

As previously said my suggestion for regulating this is:

"If anyone else is attributed in the context of a work based on OSM data 
(like other data providers, designers, service providers or publicists) 
the OpenStreetMap attribution needs to be at least on the same level of 
prominence and visibility as those."

2) Also beyond that you formulate more exceptions than actual 
requirements and where you formulate requirements they are put in 
obviously weasely terms or are tightly limited to very specific 
situations:

* "you may omit the word "contributors" if space is limited" - since 
space is always limited obviously this is a bogus requirement with no 
practical effect.  So you essentially say "© OpenStreetMap" is always 
sufficient.

A suitable rule would be:

"if space is so limited that printing '© OpenStreetMap contributors' at 
a legible text size would take an unreasonable amount of space you can 
shorten this to '© OpenStreetMap'"

* "Except for small images, attribution must be visible [...]" - being 
vague here while being precise with the 480 pixel in case of mobile 
applications is remarkable.  But even more remarkable is that there is 
no attribution requirement given for these "small images" - which can 
be interpreted as if no attribution is required for small images at 
all!

* Naturally the section on "Geocoding - Search" would be generic on any 
non-visual interactive applications using OSM data.  Limiting these 
requirements strictly to geocoding is questionable.

* Declaring printing the URL as the only and a sufficient method "to 
make any Person [...] aware that [...] is available under this License" 
in non-digital/non-interactive applications does not seem a good way to 
implement the idea of the license.  Mentioning the license directly (© 
OpenStreetMap - source data available under ODbL) seems a more suitable 
and should at least be an equally allowable method of attribution in 
such cases.

3) Your paragraph about "Machine learning models" is essentially out of 
place in an attribution guideline.  The whole idea of a produced work 
becoming a derivative database is extremely delicate and with various 
issues.  The concept of derivative databases and produced works depends 
on an uninterrupted chain of responsibility from the original database 
via derivative database to produced work.  Interrupting this chain by 
allowing a produced work to be turned back into a derivative database 
essentially breaks the license.

The very purpose of a machine learning system is to generate semantic 
data and a common property of such systems is that when run on the 
training scenario they more or less reproduce the training data.  
Considering this an exceptional use case is highly questionable.

Sneaking this into an attribution guideline is ill-advised IMO.  
It seems this has been looked at purely from the perspective of 
corporate OSM data users and not from the perspective of hobby mappers.  
I see no reason other than corporate greed why machine learning models 
trained with OSM data should not be considered derivative databases.

4) The most obvious practical guideline to fulfill the "reasonably 
calculated" would be that the attribution would need to be designed in 
a way that at least 50 percent of the map users could, when asked about 
the origin of the map they are looking at, quickly and without much 
difficulty point to the attribution.  But you don't say anything in 
that direction.

Overall i think this is totally unacceptable and looks pretty much like 
being written by corporate representatives as how they would like 
attribution to be handled with very little regard to the interests of 
the hobby mapper community and the mission of the OSMF.  I formulate 
this so strongly because i have on many occasions in the past pointed 
out that we have to formulate clear requirements to data users for what 
we expect from them - yet i can find hardly any of this in the draft.  
This is very disappointing.  As i have shown above with various 
formulation suggestion it is not actually that difficult to put clear 
requirements into words which makes me think this draft explicitly did 
not want to do so.

If the OSMF is not able to create an attribution guideline that 
safeguards the interests of the OSM community we will have to create 
our own guideline that lives up to the promise of being a 
real "community guideline".

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