Community Lightning Talk, Mike Nice – Transcription

Next up is oh, me. [ Laughter ] Okay. Let me just so I’m not Mike. On the program it says Mike, Mike Nice. He is the one that actually did all the work. So I’m just here talking on his behalf, really. But I did inject a few things of my own. So thanks to Unicode and the limitations across computers. The railroad icon is gone it’s not displaying yet. I need to mirror it. Just a moment. So now I gave it away. Okay. So there’s that’s what I meant. No railroad character even though it was in the printed program. It was in the Website. It actually shows up if you if you exit presentation mode. [ Laughter ] So I’m disappointed. But there we go. So who here knows Map Roulette? Quite a few. It’s a micro tasking or gamification tool to help fix small things in OpenStreetMap. And one thing Mike came up with trying to fix is railroad crossings. It’s important because railroad crossings, level crosses are dangerous. In 2015 alone, there were 200 fatalities and 800 injuries at railroad crossings. It’s preventable if people can see them coming. It’s important to have these in OpenStreetMap. Even the federal roadway administration doesn’t have a good location database for them. They have one, but it’s very old. This is what we started out with in Tiger. Back in 2007. So you can see kind of see the problem. Railroads are crossing with motorways. Are just randomly strewn all over the place. And people have done a good job fixing it up. But there were lots missing. Mike created a challenge to see where there are railroads crossing pedestrian paths and normal roads. But there’s no crossing defined yet and no properties at the crossing to defined. So he wrote a diary post about it. You can read that. I encourage you to read his diary post. And he came up with about 70,000 locations where it needed fixing still. And in ten months’ time we did it. I think it’s a testament to how well Map Roulette can work to fix, especially the small things. This is very easy to fix. Looking at the intersection, yes? Okay. I’ll add it. Taking maybe 15 seconds if you’re halfway experience. You can do this if you start mapping, really. So about 250,000 users did this, a few did a lot and the rest did a fair amount of. But you can see how crowd sourcing comes together in a tool that makes it simple to contribute to small things. Now OpenStreetMap is the only map that has all these crossings at the right locations because we all aligned them to current aerial imagery. So that’s really a pretty fantastic achievement. Who of you has contributed to this particular challenge? Do you remember? One, two, three, four yeah! [ Applause ] So for more challenges, look at Map Roulette. I have a few small updates after this. Map Roulette overall, we fixed about 1.2 million things since 2012, that’s, I think, a good contribution a good chunk of contributions to OSM overall. Especially if you take into account that it’s really focused challenges that really ask about very simple but important things that could be fixed in OpenStreetMap. And people like spending either two minutes or two hours or even two days or two weeks helping to fix these problems. And new ones are added pretty much every week or almost every day. The problem was, though, that they’re not very easy to find, always. It brings me to the last little topic. Map Roulette let’s call it Map Roulette 3. So we already had 2. The backend is pretty much the same. But we did a lot of work on improving the user experience. Right now it looks like this. It’s okay, but it’s confusing to a lot of people. I get a lot of feedback about how do we find the interesting things to map? Because it’s challenging. And there’s a search box, but it doesn’t work very well. So I’m going to show you a few screenshots of the work in progress how we’re going make that much easier. So this is going to be looking at a task. This is kind of the main new interface. It’s going to be a discover menu and results for metrics. If you’re not signed in, you can always see a lot of things, but you can’t work on it yet. You have to sign in through OpenStreetMap. You will see a splash screen. If you haven’t seen Map Roulette before, this will help you quickly locate a challenge. And then once you’re past that splash screen you see on the left a lot of challenges you can filter by starting typing in the search box there. Filtering by difficulty and the type of work you want to do. Roads and schools and whatever. There’s different categories. And the list will get smaller based on the selection you make there. There will be geographic selections. If you want to work in Canada or Ecuador or wherever, you can do that also. So you can filter by types. This is just examples. But there’s going to be a list of types that’s going to be created by all of you. By difficulty. And then you can kind of expand that a little bit to see a little bit more about the challenge. These are kind of early mockups. But we’re getting pretty close to getting this done. Then looking at a task to fix, there’s three simple buttons. Fix looks good and skip is pretty much the same ass it is now. And that’s that’s what it’s going to look like. I think if you have feedback, I’d love to hear that. And we are some ways along. But really this tool is built for you. And in the end, also by you, in the sense that, like, you can contribute to the code, of course. But also any feedback about challenges. If you don’t know how to create one, I can help you out. And if you want to have other people help fix your problem in OpenStreetMap, this is the tool to do it. And I hope it gets even better with Map Roulette 3. So I’m excited for all of you to start using it. And it will probably be online in about a month, a month and a half. I think. Have fun with it. Thanks so much. [ Applause ] This is kind of awkward that I announce the next speaker. Ingrid are you around? There you are. Yeah. Okay. Good afternoon. Ingrid. OpenStreetMap marketing adversity, currently looking at and today I’d like to talk to you about the youth mappers contribution to community resilience. I’m part of the youth mappers chapter which is in my current university. So we did this, he had a hackathon earlier in the year. We were requested to come and see how we could contribute to how we can address our refugee situation in Uganda. Currently Uganda is the fastest growing center of the world’s refugee crisis. We are having as many as 2,000 refugees coming into the country and fleeing from harsh conditions. They are fleeing away from the famine and the wars happening. Most of the refugees are coming from South Sudan. When they come into the country, they’re given food, nonfood items. They’re also given a 50X50 foot square of land. They can set up buildings and tents and they can grow crops for them to be able to sustain themselves. So my contribution, I decided to do a land cover modeling and prediction for that situation in Uganda. In order to do this, I wanted to use as much open data as I possibly could. I needed to get a bunch of land site images and also get OSM data. I used the dataset. I also used to get the extra views of the camps. So in order to do this I used the I used an open software. And I used the plugin. I got the images. And I tried some processing with them using the maximum likelihood classifier. I was able to classify the 2014 and 2017 images to do a projection into the future to 2020. So my aim of doing this was so see how the land cover is changing and how it might be in the future. And to do this I used the neural networks algorithms because of the multipoint intersection. Between 2014 and 2017 there was a very big decline in vegetation and reduction in the builtup area. So I did the same for all the settlements. And I also calculated the amount of vegetation that was lost per refugee. And in order to do this, I just got the vegetation the pictures attached with vegetation in 2014 and 2017. And I realized that there was only 109 square meters for them. And 200 for the other. And my other colleagues, they also are youth mappers. They did similar work. But in their findings they used different time periods as compared to what I did. Differences there. And they found people are actually growing crops. So they are able to sustain themselves. And also another person, she used a lot of OSM data. She got a dataset and put in settlements and all that. She also used geological data and stuff. She was able to come up with a map which will show the best locations that could be used to set up in these different areas. So on the right is the Imvepi district, the district with most of the refugees in Uganda. And we have the biggest settlement right now. So as youth mappers in Uganda, we don’t just do these different kinds of analysis. We try to make we have as many maps as we possibly can. Because we tend to be really busy. But we collaborate with different people on different faculties to come in and teach them how to use OSM. How to download their data. How they can use it to make all kinds of maps. And then still this year there was a hackathon which happened in Uganda. And mappers were called to assist in that training. We are training many people how to use that data. And they also did best management analysis in which they were able to make maps, suitability maps, for which areas would be set up for dumping sites in Uganda and Kampala. And they would be able to dump at the sites. So also in my university, the department of geometry is teaching OSM. How people can go out and pick data using OSM tracker. And get the data from OSM and use it with a bunch of analysis. And I come with greetings from the youth mappers chapter in Uganda. Thank you. That’s it. [ Applause ] Thank you very much, Ingrid. And all the speakers. This was a very engaging and very fun session, I think. We have some time for questions. So the speakers are pretty much all here still. So if you have any questions for any of them, I would encourage you to step raise your hand and speak. Is there someone with – here? Yeah. That’s right. Clifford has a question there. AUDIENCE: This question is for Marc. I’m wondering how you envision communicating with the community on this project. I haven’t really heard much about this. And this seems like a pretty big undertaking. We’re talking about artificial intelligence. Is somebody going to be in my area making changes? How do you envision this work? This is kind of sounds kind of troubling. Right. Okay. So like I was saying in my talk, there’s still humans as part of that process. You can’t just unleash AI on the map and start editing willy nilly. You already had the Facebook team talking yesterday about how they’re going about and they’re engaged with local communities. It would be troubling if we have people that are not part of the community that are going to start adding mass imports. But I think we’re all in the same room, we’re all talking to each other. Like I said yesterday, we had a conversation about validation and how we can apply this in an automated way. And, again, yesterday was talking about how we have the compute power. We have the data to monitor these things. So I guess it’s going to be a balance. The community can push back, of course. I think we should welcome new mappers. If they’re AI mappers, then that’s fine. If we can systemically review all of thinker edits. And that’s what I propose. It’s not really a different thing. Just a new type of mapper and we shouldn’t be afraid of these types of mappers. Does that answer your question? [ Away from microphone ] We’re doing the reviewing. The community is doing the reviewing. We’re building validation tools. We’re building linting tools. Since this is going to scale out, we need a better approach to validation. There are a ton of, you know, validation tools that are being built. So we had OSM chart talk, we had Map Roulette looking at systemic problems. The linting rules that Facebook are building. Apple has really some open source around validation. It’s not we’re not talking about something abstract. This work is happening because there’s a lot of interest in working with OSM data. And if the database is going to scale, we have to do these formal processes for checking. And I don’t know if it’s a problem with communication. Maybe we should push this as part of the Foundation. I don’t know if there’s an actual validation working group. And talk to them. Or if there isn’t, we should create one. Because this is an important time in OpenStreetMap history where we’re not just filling the base map. We have to start, you know, reviewing and creating reputation systems, you know, talking about tags and filling out and correcting information as well. So I can’t give you a straight answer that this is the person that is working on this. I’ve seen it. I have been to three State of the Map conferences, like the past three. And this year was very much focused on validation. And that’s also in conjunction with ML. So I don’t know. We’ll have a document up. At least the people that talked about validation yesterday in the birds of a feather. We’ll try to make it as visible as possible. Other questions? We have some more time. Don’t be afraid. Well, yeah. Back there. AUDIENCE: So I don’t really remember who said there was going to be a stats talk in birds of a feather, but what kind of stats are we talking about here? [ Laughter ]

Hi. Marc. [ Laughter ] So so we built well, I can talk about a bit of the work that about the missing map stats. We built user pages and leader boards. You know, those stats that number of buildings. Added per user. We could talk about map density. Really what we want at that birds of a feather is we want to hear use cases for analytics. If we’re going to build the next version of the stats, next version of any stats by applying that is going to start working on the entire planet history. I mean, I know I’m working on that. I know that a few groups are working on that. I’d like to share use cases so that we’re not repeating work. We had this talk this birds of a feather last year. It was very useful. I think we should host it every year. If you have any ideas of use cases for stats on the entire planet history, come talk. One interesting thing and because we said that validation is very important, especially this year, is how do we do validation on the entire planet history. Maybe we build a reputation this isn’t my idea but go back in history and run the Map Roulette checks on the entire user’s contribution history and maybe we can build a reputation system around that. Does that answer I don’t know who asked the question. Does that answer your question? AUDIENCE: Hi. So I’ve built these kinds of machine learning mixing with crowd sourcing pieces before. And gone out into contributing communities and helped them make the transition into becoming a validating force and a force for editing to sort of scale out the their existing contributions to some kind of effort like OpenStreetMap before. And this sort of thinking about the algorithms is sort of just at annotator on the map. Another human sort of robots and human’s perspective. And because I’m asking this question because you just mentioned this about taking those algorithms, running it over the contributor’s source. That if you contributed in the past and coming up with these sort of reputation metrics. Yep. AUDIENCE: And just like today when there’s a lot of, you know many of the edits come from a very small number of contributors and those kinds of dynamics is kind of an 80/20 dynamics. Turns out that when you apply it to reputation, you get similar 80/20 dynamics. Some people are really good at making edits and some aren’t. And what ends up happening in the community is you find some people are very good at some kinds of tasks or some kinds of regions. And these incorporating machine learning into the annotation system brings that to light in a way that hadn’t actually previously been available. Positively or negatively. AUDIENCE: Positively or negatively comes up to whoever is leading the groups, right? It definitely can be a force for a lot of positivity. Right? It transitions the conversation into aligning what contributing members are good at with what they enjoy. Right? And that’s a live question. That’s a dynamic question that evolves as the machine learning evolves. I think there’s a lot of depth and nuance to that conversation. But it can be very positive. And it can often end up satisfying or completing whatever the original founding vision of OSM was for. Right. And like I answered before, I think that going to be communitydriven. Your experience and your voice should be written down and shared with the rest of the validation working group or just group that is thinking about these things. When measuring AI behavior on the map. So I don’t think we have anything formal used now. If you can I’d be happy to talk to you after the talks. Does anyone have a question for another speaker? [ Laughter ]

Marc, not a question, but one of the outcomes of that birds of a feather was a validation Slack channel. So if you’re interested in participating in that, talk to myself. I’m Josh. Or you can talk to Marc also and we can get you included in that. Cool. I can stay up here. I don’t have a I mean, we can run the hour. Should I sit down? [ Laughter ]

AUDIENCE: So I had a question for whoever did the thing on the districts in Virginia. [ Laughter ] Got you covered. So I was wondering, since you have the census data, if since you’re trying to make all the districts, if you could go a step further and also kind of put in the neighborhoods and the households with the addresses and all that. Would that be a possibility? So since you’re making the districts of Virginia, could you take it a step forward since you had the census data and also make like individual neighborhoods with addresses for households and all that good stuff? Sure. AUDIENCE: I mean [ Laughter ] It would make it a little more exciting for you, right? A little more detailed. For one thing, we don’t have all the houses or buildings in OpenStreetMap. So they would have to come from I don’t know. Census data. AUDIENCE: Yeah. I mean, the best you can interpolate the house is across the street. To the best of my knowledge. One thing I didn’t mention was the Voting Rights Act. When you use an algorithmic approach to generate these districts, you totally fail to take into account plain law that requires that districts have certain percentages of minorities in them. So that’s another thorn in my side in this project. AUDIENCE: You have many thorns in your side. Article last week. The parties themselves have sophisticated software REDMAP. AUDIENCE: Right. And one of the critical pieces of data that to them is most interesting for creating the geographies is the voting history at the address level. Right. AUDIENCE: And so and I’m sure did you mention in your talk the Supreme Court? The Gill versus Whitford? AUDIENCE: Whatever the days is in front of the Supreme Court right now? I was there that day. Right outside of the Supreme Court. I was hijacking your question. AUDIENCE: No, it’s cooler than my question We’re there and chanting and have signs and stuff. And an entourage shows up and it’s Arnold Schwarzenegger. AUDIENCE: Was he a supporter of your work? Working on it. On the Schwarzenegger note, I think that’s all we have time for. Enjoy the speakers and enjoy the rest of the afternoon. [ Applause ]