Mapping with Artificial Intelligence

By using machine learning in the Tasking Manager, we can identify the more complex tasks (in dark red) and break them up into smaller, more manageable tasks for users.

In the last few years, we've seen huge advances in the way artificial intelligence (AI) and machine learning have assisted mappers. Humanitarian OpenStreetMap Team is proud to be leading in the use of this innovative technology in an effort to improve people’s lives in a meaningful way by applying the Principles of Digital Development for using machine learning responsibly in OpenStreetMap.

We’re excited to share a few of the ways we’ve used machine learning in 2019 to enable our mappers to collect better quality life-saving information faster and more efficiently:

  1. Using AI to detect buildings in East Africa. In partnership with Microsoft, we used machine learning to map buildings for disaster relief in two countries: Uganda and Tanzania. The mapping covered an area of 1.2 million km2 and over 18 million buildings were identified - data that our teams in the two countries started verifying remotely and on the ground.

  2. Simplifying tasks in the Tasking Manager with machine learning. We are testing a new machine learning-assisted Tasking Manager that can detect the complexity of a mapping task and suggests more manageable task sizes, ultimately distributing the workload and balancing the amount of time and effort it takes to complete a task.

  3. Experimenting with AI-assisted mapping in OpenStreetMap. We also contributed to the development of a ‘Tasking Manager-machine-learning-playground’, where users could test open-source solutions and try AI-derived datasets to assist their mapping.

  4. Creating the ML-Enabler tool to bring together all parallel machine learning-integration efforts happening across OpenStreetMap. In partnership with Development Seed, we developed an open-source programming framework that enables several machine learning-models used across OpenStreetMap to be accessible through one consistent API.

Our approach to this inclusive and collaborative community effort is to design together with the user, test iteratively to carefully understand the ecosystem we are moving in, rely heavily on open data, open source, and build on existing efforts. This is based on our commitment to applying the Principles for Digital Development for innovation in OpenStreetMap.