[Osmf-talk] Alternative Strategic Plan

Mon May 15 04:47:02 UTC 2023

*TL:DR: the first part is me explaining why I don't believe generic
building=yes are useful the way newcomers are urged to add them. The
comments focus only on "common good" PoIs (both because they are highly
reusable, less likely to be worth competition, easier even for not
OSM-survey focused apps and easier to find volunteers to moderate/accept
them) in a way that maybe more than an anonymous user even from a
completely different app could confirm seems a clear win-win. But if we're
going to use OSM Notes, they could also be machine parseable from the
start, and if data was added based on 1 or more Notes, then we help humans
to keep some metadata of previous steps (like how many different people
confirmed the type existed in the local area).*

> Mateusz Konieczny said
> mapping accurate building=yes is highly helpful (low quality armchair
mapping has dubious or negative utility, but mapping building=yes is not
indicator of this - rather inaccurate guessing of building=* value is a
bigger problem)

Let me give a real world example. Weeks ago on unofficial OpenStreetMap
Telegram channel  (message https://t.me/OpenStreetMapOrg/101372 ) a mapper
from Colombia complained that the buildings added in an area (near volcano
Nevado del Ruiz) was visited by local civil defense. The building=yes
(added by a human mapper, but geometry suggested
Microsoft/buildingFootprints) actually was a... tree. Guayacanes and
Yarumos to be more exact. He was pissed off on the chat, saying it already
complained in the past.

Who's to blame?

Note there's an entire idea of what could save people's lives in case of
disasters around OpenStreetMap, so the person who added the data likely was
with good intentions because he believed it was easy and fun. Also, the way
is typically advertised how anyone can help saving lives without any
previous skill required, when really a false-positive like this could
happen, I could imagine his frustration recommending the rescue team to
check such places, but then the data would be unreliable because of the
entire process itself.

But even if it wasn't by invalid meaning (e.g. not a tree), it is still
highly problematic, so let's not blame AI alone. The way is explained by
new volunteers, the armchair mapping of building=yes would actually be
someone else's home who might be in danger, and that imagery alone would be
sufficient to make a difference. However, for the same reason one
tourism=wilderness_hut (which by imagery could be described as
building=yes) in the mountains in European country may be just non
permanent home place, is not possible by aerial image alone (even if is
lucky that imagery not outdated and there's no error in classifying it as
potential building/house what is not) be sure these isolated places in
Africa may be not inhabited.

So I could still be okay with "inaccurate guessing of building=*" this is a
case where organized editing (and likely AI-assisted buildings which try to
simplify it, but even without it) actually are over promising that it is
used to help people in the ground, but it simply cannot. It's not the blame
of data users (let's assume those on local civil defense) because of what
external organized editing keeps saying is possible.

The building=yes without any further classification, even from its creation
would be like highway=road
https://wiki.openstreetmap.org/wiki/Tag:highway%3Droad , which editors
treat it differently to disencourage its use.

> John Whelan said
> If you have no census information then at least you can estimate how many
people live in an area from the number of buildings.

Ok, let's think it could be possible, however there's a problem with this
way of thinking: almost all countries actually have national censuses, (see
<https://en.m.wikipedia.org/wiki/Census>), often each 10 years, and
population is surveyed in their homes to get data from near every person
living in each country (e.g. not a well planned sampling, which can be
accurate). Census is not a "rich country" thing, might not be easy to find
on websites, but they exist, and often local emergency response use
national data on this. The UN World Population Prospects <
https://population.un.org/wpp/> also has data based on national censuses
and estimation between censuses without going to all administrative levels,
but this one is meant to make it comparable between countries. In the case
of places which UNOCHA is focusing, it also would have detailed Population
Statistics (the COD-PS) already aligned with COD-AB (administrative
boundaries) so this kind of calculation is likely to be better reused
already reviewed by professionals. My point is that the premise of no
census cannot be used to justify OSM data as replacement.

Extrapolations to distribute the baseline population statistics on smaller
areas than census are typically done based on **verified features** in the
ground (so, they could work similar to sampling strategy, which requires
very strict care to not introduce bias), but even professional use without
full census will come with caveats. Yes, sometimes they might be based on
some kind of sensors (such as how many cellphones are connected to
antennas, or more recently, number of users of apps like Facebook to
estimate migration workflows), but all these come with aid to explain what
could make them less complete. In the case of using OpenStreetMap building,
the mere wrong sampling (because if cannot map EVERY building in a strictly
consistent way) would introduce bias.

> John Whelan said:
> Useful for getting vaccines to the right place, how many schools do we
need etc.

Going from armchair mapping to armchair government assistance is a big
jump. Anyway, let me explain the limitations of why this cannot be done
without people on the ground.

Vaccines need specific storage conditions (e.g. transport might have time
limitations from point to point) and money for creating new schools,
hospitals etc cannot be based on pure statistical guessing that there's
people in a place or not. Imagine for example the full planning of building
a physical hospital, just by aerial image, and only go into the place for
the first time when starting the construction! Even internacional aid will
either partner with local government which will provide additional
Information and/or have person's in the region, which in the worst case can
ask people around estimation of population nearby or... just go first to
each potential village and survey. My argument is that de facto
implementation of aid needs some minimal quality of metadata, and often the
local guides may know citizens by their names. For a generic overview,
national censuses will work (that's the reason they exist, to also help
with decision taking), and for a local response, the data must be
trustworthy, not merely statistical probability of a place existing or not.
And, anyway, in case of emergency response, they often train a lot before
disasters, and surveying the place often is simpler.

The "getting (something) into right places" is actually very relevant for
apps that do consume OpenStreetMap data. I ask anyone who believes mapatons
which adds no more than generic buildings (sometimes not even nearby roads)
to try to use the OSM main website functionality to plan a route (it's near
the search functionality) to the mapped place and do NOT use
latitude/longitude. If you cannot think how to do it then one of the main
utilities of OSM data, routing, will not work in that region. Rationale:
when roads and buildings lack addressing information, and maybe there's not
even points of interest around to use as hint, there's no way to write the
destination for a routing plan!

I personally don't like any idea of attempts of use cases from
OpenStreetMap which implicitly assume national countries are a failed
state, and then any desperate attempt to try drastic measures (ones which
not even would be implementable in the home country of someone helping
overseas) to save lives become acceptable. A good strategy to learn is to
be humble.

> John Whelan said:
> The other thing is smartphone mappers can add tags to buildings that
exist is easy but mapping a building is more complex with a low end
smartphone.
> Christopher Beddow said:
> In my view, apps like Organic Maps are one of the biggest opportunities.
(...) People who use the product can naturally give the feedback the
builders need. (...)
> Mateusz Konieczny said:
> building=yes geometries are highly useful for orientation. (...)

The thread stated by Steve was about lack of addresses (which I agree), and
buildings (either as nodes or closed ways) aren't mapped equally around the
world. As drastic as may seem de-emphasys on building=yes, how to bring
people to care about metadata essential to make more than map for the
render, if all the discussion, even on saving lives, is aerial image
mapping?

For years there's been some apps focused on this task like the
StreetComplete and I think they already require the geometry of the
building. (And obviously, there's Vespucci). I still have no idea why
organized maphatons with mappers which actually live in the regions do not
use these apps at least on what is feasible to reach on foot. A thing I'm
not sure is now all these apps deal with previous mapped points of interest
that might need review after some period of time, but I'm 100% in favor to
ones of general utility, such as hospitals, schools and some other
government amenities (I can compile a list based on most used ones by
emergency response), all developers agreed to suggest check with higher
priority.

To my knowledge, beyond what's done by the few places which do have
government data with addresses, the apps which allow surveys are unsung
heroes, and we should greatly expand it. So yes, the comment from
Christopher Beddow about other apps, while obviously don't make sense
they're become equal to each other, as long as they are mobile apps (so we
could assume person is near a place) strategies to recheck the metadata
could exist on everyone of them which already is capable to upload data.

I also think that even in a worst case scenario (where metadata about PoIs
could be a commercial advantage for some private companies) , revalidating
some kinds of PoIs like hospitals or a police station should be perceived
as common good. I mean, the same way most addresses today are from generic
(non humanitarian mapping) this is unlikely to change drastically, in
particular for apps that are generic but simply have so many people passing
around certain places. However, if some apps are only able to do it via OSM
Notes, ideally we should know upfront some regex-like way to make the text
machine parse-able, so we could make it easier for potential volunteers in
each country to review them. In the worst case scenario (think like an
unreliable source, provider did not give even a name, only the type) either
if uploaded to OSM main server, or as additional data repository to be hint
to be checked by at least another human (maybe another app) then the next
person could go there.

So, the entire comment of Christopher Beddow makes sense. The apps using
OpenStreetMap aren't the same, so ideally it would have conventions on how
different apps could add or confirm the specific set of focused PoIs. Some
could be far more complete, but the basics are worth it, because in remote
areas, you may never have someone with a more complete OSM app, but knowing
the place does not exist anymore becomes relevant. Something good enough to
be able to work even with simple prompts as mentioned by Beddow is a good
thing and we assume that the original note is not expected to reply to
questions, because it could be simple to just wait until the next person
passes there. I guess that from this list, obviously Matheuz and his
experience with StreetComplete could give extra help with how we could come
with metatags of check dates for the PoIs which could work with more
complex scenarios or at least one for how many different initial users
checked the initial PoI as existing.

Att.
Rocha

On Sat, May 13, 2023 at 12:47 PM Steve Coast <steve at stevecoast.com> wrote:
>
> Dear all
>
>
>
> We formed OSMF to take care of the map. Let’s reset and focus on that
with a strategic plan by the mappers for the mappers:
>
>
>
> 1.     The website will focus on completing the map. The map currently
shows the “best” view of the map, it will be changed via voluntary or paid
efforts to show the “worst” view of the map to encourage completion, like
in the beginning when it was blank, and it led to huge efforts to fill the
map in:
>
> a.     We will decide what the main thing to complete is. For example, if
we decide it’s address data then we will do the following:
>
>                                                i.     OSM will only
render roads with a new tag “addr:complete", where mappers manually will
have to mark roads as address complete. This will immediately make the map
go very blank and create a large global project to finish addressing, which
is the main thing missing in OSM.
>
>                                              ii.     Once complete, we
will take similar steps to map, for example, PoIs and only show roads will
all PoIs added.
>
> b.     OSM will only render features newer than 24 months to encourage
refreshing and revisiting. A tag “feature:verified_2023” or equivalent will
be used to do this, along with the last edit date of the feature.
>
> c.     Map notes will be turned on by default.
>
> d.     Social and map quality features will be built in to osm.org which
will drive engagement and mapping towards completion, for example alerting
users to edits or routing problems near where they edit.[2]
>
> e.     Leaderboards of editors, countries, states, regions and counties
will be front and center to encourage editing, for example percentage of
“addr:complete” roads per country.
>
> 2.     Funding will focus on completing the map:
>
> a.     By having a clear metrics-based plan above we can seek funding to
build specific tools and features towards map completion.
>
> b.     A paid “OpenStreetMap Approved” program will standardize corporate
membership by certifying a company uses OSM data in a way that respects the
license and community.
>
> c.     A certified version of OSM will be released quarterly that has
been semi-automatically checked for validity and correctness. Paid
corporate members can be involved in the process.
>
> d.     OSM conferences and local chapters will pay OSMF a small fee per
attendee or member and in exchange also be “OpenStreetMap Approved”, once
the board is shown to be effective.
>
> 3.     The board will focus on completing the map:
>
> a.     Reduce the board size and require regional representation.
>
> b.     Board members must make a public financial and time commitment.
>
> c.     The board will build completion metrics for the map, by using
existing tools and working with companies who already have many tools, and
engaging with the community. These will be the main agenda of the board
meetings.
>
> d.     All discretionary funding will go to projects and community
members who build credible plans to help complete the map.
>
>
>
> If this is interesting, I’d love feedback. We can run some BoF sessions
at SOTM US, EU and Africa in July, November and December. Then SOTM Asia
when it is confirmed.
>
>
>
> Best
>
>
>
> Steve
>
> _______________________________________________
> osmf-talk mailing list
> osmf-talk at openstreetmap.org
> https://lists.openstreetmap.org/listinfo/osmf-talk

-- 
Emerson Rocha
Full stack developer at Alligo
Transdisciplinary researcher at Etica.AI
Member of  The IEEE Special Interest Group on Humanitarian Technology (IEEE
SIGHT)
Member of The IEEE Global Initiative on Ethics of Autonomous and
Intelligent Systems
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/osmf-talk/attachments/20230515/8a5db768/attachment.htm>