Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
Besides the tech boosterism, another issue is that it's disingenuous if Facebook claims to be strongly supporting OSM, while continuing to keep their valuable user-provided data in a separate, proprietary database. Facebook and Google have the two best lists of POIs like shops and restaurants, and an extensive database of customer photos and reviews which they control. While Facebook has decided to use OSM for road, street and waterway data (which they couldn't easily have users add), they keep this data for themselves. Were Facebook interested in improving OSM, they could share their POI data, including when a feature was last visited and notes about which feature no longer exist. This could add millions more OSM contributors for features like shops and restaurants, which are not yet completely mapped even in well-developed OSM communities in Europe, and it would be revolutionary in Indonesia and Thailand. Only a few people will every become hobby mappers, adding waterways, highways, landuse and such for fun, but every business owner wants to see their shop or office on Facebook, so these POIs would be added and kept up-to-date by users. I don't expect Facebook to share this data for free, because a large part of their business model is recording your geodata and using this to maximize profit for their shareholders, but if they ever decide to really prove "we're not that evil", sharing their data could go a long way to changing Facebooks poor reputation for corporate responsibility and transparency. Joseph On 8/1/19, stevea wrote: > (I chose the wrong source email address; apologies if anybody gets this > twice). > > Thanks, Jóhannes. I did try FB's tool myself and was pleasantly surprised > it does a "looks OK for now" job of how Mikel put it earlier: "a balance > between turbocharged and exploitation." I hear you as you say that > mapwith.ai has, as I described, a comfortable workflow of "AI suggests, > human maps, human checks that what is acceptable can be uploaded, human > uploads." That's fine, it does indeed have "a human in the loop" and the > human checks for quality, the human is not just being there for the sake of > being there. This aspect of "humans, not AI, determine quality" is a > critical component of what I am saying. > > What I believe raised ire here was the BBC botching the "press announcement" > as a stilted and seemingly uninformed "cheerleading" piece that made AI > sound as if it were a "magic bullet" that was going to save mapping in OSM > somehow. It isn't (magic) and it won't (though AI is an important tool > going forward, especially as it is coupled with human wisdom and a hawkish > eye towards high quality). OSM is, and will always be, a > human-participating project, with all of the social and "get outdoors and > map" project as one (human) might like it to be. AI can and does help, > that's fine, as long as humans are always "in charge." > > Again, it sounds like there is a lot of agreement here. > > SteveA > ___ > talk mailing list > talk@openstreetmap.org > https://lists.openstreetmap.org/listinfo/talk > ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
[OSM-talk] Proposed removal of rendering support for natural=marsh in Openstreetmap-carto style
I've proposed removing the rendering of natural=marsh in Openstreetmap-carto, the rendering stylesheet used for the "standard" map layer on openstreetmap.org https://github.com/gravitystorm/openstreetmap-carto/pull/3829 The common tagging for a marsh is natural=wetland + wetland=marsh as with other types of wetland. According to the wiki, natural=marsh has not been recommended at least since January 2009: https://wiki.openstreetmap.org/w/index.php?title=Tag:natural%3Dmarsh&oldid=217946; and since 2016 the page has specified that this tag is deprecated: https://wiki.openstreetmap.org/wiki/Tag%3Anatural%3Dmarsh Since 2009, wetland=marsh has shown steadily increasing usage. It is now used 130,000 times, compared to less than 10k remaining uses of natural=marsh. Please comment at github if there are any objections to this change or other comments: https://github.com/gravitystorm/openstreetmap-carto/pull/3829 ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
(I chose the wrong source email address; apologies if anybody gets this twice). Thanks, Jóhannes. I did try FB's tool myself and was pleasantly surprised it does a "looks OK for now" job of how Mikel put it earlier: "a balance between turbocharged and exploitation." I hear you as you say that mapwith.ai has, as I described, a comfortable workflow of "AI suggests, human maps, human checks that what is acceptable can be uploaded, human uploads." That's fine, it does indeed have "a human in the loop" and the human checks for quality, the human is not just being there for the sake of being there. This aspect of "humans, not AI, determine quality" is a critical component of what I am saying. What I believe raised ire here was the BBC botching the "press announcement" as a stilted and seemingly uninformed "cheerleading" piece that made AI sound as if it were a "magic bullet" that was going to save mapping in OSM somehow. It isn't (magic) and it won't (though AI is an important tool going forward, especially as it is coupled with human wisdom and a hawkish eye towards high quality). OSM is, and will always be, a human-participating project, with all of the social and "get outdoors and map" project as one (human) might like it to be. AI can and does help, that's fine, as long as humans are always "in charge." Again, it sounds like there is a lot of agreement here. SteveA ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
31. júlí 2019 kl. 19:01, skrifaði "stevea" : > Just because, as you say (and I agree), that "human mappers have not been > able to produce high > quality maps worldwide" doesn't mean that we can't, we simply must strive to > do better. And we do. > And we should using available tools like AI, though if we do, we absolutely > must include a strong > component of human-oriented quality assurance right at the forefront of doing > so. For the sake of this discussion it should be pointed out that (except for initial Malaysia thing) the mapwith.ai website does just that. The AI has found possible roads and it is up to humans to confirm if it is a road and re-classify it if Residential (the default) is not correct. ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
Right (or nearly right, imo), Kathy: thank you for your reply. I didn't say OSM absolutely DOES have high quality. In my last decade of mapping here, I certainly have seen it get better (in pockets) as well as worse (in smaller pockets), so on the whole, it gets better / higher quality. What I did say is that OSM absolutely "CAN have high quality without AI." We must strive to do so, knowing that we sometimes miss the mark. But including AI without hawkish attention to quality is folly, Just because, as you say (and I agree), that "human mappers have not been able to produce high quality maps worldwide" doesn't mean that we can't, we simply must strive to do better. And we do. And we should using available tools like AI, though if we do, we absolutely must include a strong component of human-oriented quality assurance right at the forefront of doing so. It sounds like we largely agree. Provided we keep quality at the top of our consciousness as we do so, whether we use AI or not. I appreciate the opportunity to share dialog, SteveA > On Jul 31, 2019, at 11:48 AM, Kathleen Lu wrote: > > I agree that human wisdom is critical to high quality, and AI isn't useful > if, at the end of the process, it doesn't produce quality output, but I will > challenge this statement: "you can have high quality without AI." I don't > think that's definitively true for a global map. It's very difficult to keep > something at that scale that is constantly changing up to date, and while OSM > is very high quality in some areas, human mappers have not been able to > produce high quality maps worldwide. Corporations that use traditional survey > techniques also have a lot of difficulty even while spending $$$ (and many if > not all of them are also using AI). AI can augment human mapping in ways to > make scale more manageable, and I think both will be needed to make a > worldwide high quality map. > > > On Wed, Jul 31, 2019 at 10:26 AM stevea wrote: > Oops, "social conscience." (not conscious) > SteveA > > > ___ > talk mailing list > talk@openstreetmap.org > https://lists.openstreetmap.org/listinfo/talk ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
I agree that human wisdom is critical to high quality, and AI isn't useful if, at the end of the process, it doesn't produce quality output, but I will challenge this statement: "you can have high quality without AI." I don't think that's definitively true for a global map. It's very difficult to keep something at that scale that is constantly changing up to date, and while OSM is very high quality in some areas, human mappers have not been able to produce high quality maps worldwide. Corporations that use traditional survey techniques also have a lot of difficulty even while spending $$$ (and many if not all of them are also using AI). AI can augment human mapping in ways to make scale more manageable, and I think both will be needed to make a worldwide high quality map. On Wed, Jul 31, 2019 at 10:26 AM stevea wrote: > Oops, "social conscience." (not conscious) > SteveA > > > ___ > talk mailing list > talk@openstreetmap.org > https://lists.openstreetmap.org/listinfo/talk > ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
Oops, "social conscience." (not conscious) SteveA ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
I believe introducing into OSM technologies based in AI / machine learning REQUIRES a concomitant discussion about how the data WILL BE high quality, because they are quality assured (and perhaps here is a brief sketch of our QA process, or a pointer thereto). Anything less feels disingenuous to me, as well as logically appears to be a false choice. To say "quality issues are a seperate issue" (sic) seems an insult to OSM and indeed the very introduction of the AI technologies themselves to our project. It is early (well, "earlier") days for these technologies as they are being built and deployed today, and while many (myself included) agree they can be useful and have their place, they MUST be accompanied by a social conscious as we do so. OSM already has strong tenets like community-developed consensus to create such a social conscious, so deploying AI without eyes wide open and a firm hand on the tiller is nothing less than insanity doomed to failure. The least we can do is to strongly couple discussions of quality with AI deployments, rather than divorcing them by declaring them "simply announcement." I know that whenever I hear such "announcements" without any discussion of how quality will be assured that it is time to be immediately skeptical. Please, let's keep AI on track by coupling it with discussions of quality, not making them separate issues, because truly, they are not. Simply wishing that we can separate AI and quality will only more firmly entrench those of us who know to keep them together: you can have high quality without AI, but you really shouldn't have AI without high quality. Not as long as human wisdom is present and has something to say about it. Not to put it too dramatically: do we really want to hasten "the robots are taking over" by taking the throttle off, by ignoring or diminishing the importance of quality and its discernment by humans? Of course not. SteveA > On Jul 31, 2019, at 6:27 AM, Florian Lohoff wrote: > > On Mon, Jul 29, 2019 at 10:41:17AM +0200, Michael Kugelmann wrote: >> Am 25.07.2019 um 11:36 schrieb Florian Lohoff: >>> And IIRC it was about a >>> collaboration with the local community in Thailand which their first aim >>> was. >> I just remember that the "collaboration" in Thailand some time ago >> (might be years) was quite poor: by using AI generated data simply >> thrown into the database they destroyed a lot of craft-mapped data. But >> unfortunately I am not aware how this evolved and about the current >> situation. That's the background why I would be very cautious about such >> "collaboration statements". > > The point was not about quality but about announcement and speaking up > publicly about it. > > And Facebook did - loud and clear for everyone to hear - Quality > issues are a seperate issue. I am pretty shure that AI can not replace > human, on the ground, observation. It can help identify places to > visit. > > Flo > -- > Florian Lohoff f...@zz.de >UTF-8 Test: The 🐈 ran after a 🐁, but the 🐁 ran away > ___ > talk mailing list > talk@openstreetmap.org > https://lists.openstreetmap.org/listinfo/talk ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
On Mon, Jul 29, 2019 at 10:41:17AM +0200, Michael Kugelmann wrote: > Am 25.07.2019 um 11:36 schrieb Florian Lohoff: > > And IIRC it was about a > > collaboration with the local community in Thailand which their first aim > > was. > I just remember that the "collaboration" in Thailand some time ago > (might be years) was quite poor: by using AI generated data simply > thrown into the database they destroyed a lot of craft-mapped data. But > unfortunately I am not aware how this evolved and about the current > situation. That's the background why I would be very cautious about such > "collaboration statements". The point was not about quality but about announcement and speaking up publicly about it. And Facebook did - loud and clear for everyone to hear - Quality issues are a seperate issue. I am pretty shure that AI can not replace human, on the ground, observation. It can help identify places to visit. Flo -- Florian Lohoff f...@zz.de UTF-8 Test: The 🐈 ran after a 🐁, but the 🐁 ran away signature.asc Description: PGP signature ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
John, Kathleen, thank you for this perspective I did not have. Yves Le 29 juillet 2019 19:25:34 GMT+02:00, john whelan a écrit : >I agree with Kathleen. Given that smartphones are more common than >internet connected computers and it is easier to add or change tags on >a >smartphone than add a long highway at least the locals stand more >chance >this way. > >Cheerio John > >On Mon, Jul 29, 2019, 1:00 PM Kathleen Lu via talk, > >wrote: > >> On the other hand, if the map of your area is completely blank, it >looks >> very daunting to a new mapper, who may be discouraged and abandon OSM >> (either as too difficult to improve and as too poor quality to use). >> The map is constantly changing because roads and other things on the >map >> are changing in the real world. A city might close off a road and >then it >> will become a "bad" street. It's easier to delete a bad street than >to add >> a bunch of streets, especially when you are surveying on foot and >don't >> have a mouse. >> I personally would much rather have a 101% map than a 1% map. >> >> On Mon, Jul 29, 2019 at 9:21 AM Joseph Eisenberg < >> joseph.eisenb...@gmail.com> wrote: >> >>> Re: "OSM map with a one percent of roads is far worse than having >101% >>> of the roads mapped with the help of AI with 1% of extras, because >>> fixing that 1% is far less work than adding 99% by hand" >>> >>> I'm not certain this is true. It might be very difficult to find the >>> 1% of incorrectly mapped roads; you don't know where to look, and >you >>> must survey on the ground with GPS, and check each road segment to >>> find the 1% that actually are blocked by a fence or gate or don't >>> really go through that clump of trees. >>> >>> In contrast, when 99% are missing it's very obvious when looking at >>> the map data. You still have to survey and add the streets, but it >may >>> actually be faster to get to a complete map of your home >neighborhood, >>> than trying to find 10 bad streets out of 1000 segments in your >>> neighborhood. >>> >>> Finally, when you look at the map and it looks 100% complete, you >>> won't see the need to start mapping and become a totally addicted >>> OSMer like you will if your village is only 1% mapped, so we may not >>> get the new contributors that we need to actual maintain the data >that >>> our robot mappers have added. >>> >>> Joseph >>> >>> ___ >>> talk mailing list >>> talk@openstreetmap.org >>> https://lists.openstreetmap.org/listinfo/talk >>> >> ___ >> talk mailing list >> talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/talk >> ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
No need to argue that much about it: I think everyone will agree that we should not, at any case, add a track in OSM that doesn't exist. It can be dangerous in any emergency situation anywhere in the world. Yves ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk
Re: [OSM-talk] Facebook mapping highways using AI in collaboration with OpenStreetMap
On Mon, 29 Jul 2019 10:53:07 -0400 Yuri Astrakhan wrote: > On Mon, Jul 29, 2019 at 6:19 AM Martin Koppenhoefer > wrote: > > > speaking about risks, having an incomplete network of verified, > > correct roads is probably more useful and less troublesome than an > > "overcomplete" one which also contains non-existent roads (e.g. > > waterways interpreted as roads) or shows connections that aren't > > there in reality. > > I think this position should be a bit more nuanced. Taken to > absurdity, OSM map with a one percent of roads is far worse than > having 101% of the roads mapped with the help of AI with 1% of > extras, because fixing that 1% is far less work than adding 99% by > hand. I'm sure we can find a good balance between both positions. Having hiked in areas with 1% maps, and having hiked in areas with 101% maps, I have to say that I prefer the 1% map. With the 1% map, there's at least no question that you're off the map and on your own for route-finding. With the 101% map, it's very easy to get in trouble because the connecting trail you were counting on doesn't exist. -- Mark ___ talk mailing list talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk