Building AI-Assisted Mapping Tools: Progress And Lessons Learned

Saturday, 1:00 pm – 20 min
Drew Bollinger
Development Seed

The use of AI to analyze Earth observation imagery is growing, but there remain some difficult questions about how this trend will affect OpenStreetMap and its users. At Development Seed, our aim is to develop “AI-Assist” tools that will empower existing mappers rather than replace them. In the last year, we’ve started building the AI-Assist toolbox that we proposed at SotMUS 2017. Our presentation will focus on a new plugin for the Humanitarian OpenStreetMap Team’s Tasking Manager, which uses AI to help mappers visualize where infrastructure is most under-mapped. The tool works by autonomously detecting building and road infrastructure from satellite imagery, subtracting the known infrastructure stored in OpenStreetMap, and presenting the difference to mappers to help prioritize their efforts. Our presentation will also cover the lessons we learned when putting these AI-powered tools in the hands of real mappers. We believe that AI can significantly improve OpenStreetMap, but we must be mindful of this technology’s limitations as it is integrated into the mapping community.