The potential of humanitarian and AI-assisted mapping in OpenStreetMap and their effect on representation bias and data quality

Benjamin Herfort

Over the past decade there has been an unprecedented momentum around collaborative mapping in OpenStreetMap (OSM). This was driven by technological developments which enabled a rapid increase in data production. Without these innovations, contributing data to OSM, e.g. through digitizing buildings from satellite imagery or adding the opening hours for your local corner shop on your smartphone, wouldn't be as simple as it is nowadays. At the same time the political context has shaped the very purpose why these datasets and maps are needed in the first place. Major political frameworks have defined data needs on a global scale, e.g. for the purpose of monitoring SDG progress. In this context the OSM project has evolved as a social product and forms a large community of people loosely connected through the joint work on a global geographic database.

The objective of this lecture is to provide an overview of the past dynamics of mapping in OSM and the outcomes it produced. It demonstrates how researchers can better understand the spatio-temporal evolution of OSM in respect to data quality and issues of representation.

First, I will raise the question of who is currently represented through mapping in OSM and how this has changed over the past decade in the light of humanitarian mapping efforts. Second, I will investigate data quality in OSM and treat this as a dedicated form of representation bias. Here I go beyond analysing mapping activity but investigate the completeness of OSM data. Finally, the enabling and inhibiting aspects of machine learning, and artificial intelligence assisted mapping for crowdsourcing community projects are presented. The focus for this analysis lies on the potential to combine automated and crowdsourced means of creating human settlement maps.

In essence, this lecture invites the audience to discuss these three questions related to OSM and open mapping communities in general:
- Where are the data gaps?
- What is the role of AI-assisted mapping?
- Who is represented?

Over the past 15 years OSM has been embraced by researchers in the domain of GIScience. Building upon the achievements of this community we can now paint a bigger picture. We have discovered that OSM data shows a much more spatially diverse pattern than previously considered, which is shaped by regional, socio-economic and demographic factors across several scales. It would be great if this lecture helps us to look ahead and allows us to discuss a shared vision of future OSM use cases and research.