Authors: Kirsty Watkinson*, University of Manchester, Jonathan Huck, University of Manchester, Angela Harris, University of Manchester
Topics: Geographic Information Science and Systems
Keywords: GeoAI, Volunteered geographic information, machine learning, humanitarian mapping, mapping inequalities
Session Type: Virtual Paper
Start / End Time: 11:10 AM / 12:25 PM
Room: Virtual 47
Presentation File: No File Uploaded
Global inequalities in map coverage persist (Huck et al., 2020). Geospatial Artificial Intelligence (GeoAI) has been vaunted as a solution to this: able to rapidly provide lower income, technologically disadvantaged countries with data vital for humanitarian response and global development initiatives. Despite promise, two key challenges remain for GeoAI: unmapped areas lack the large, high quality training datasets required for an AI model (Chen and Zipf, 2017); and the ability of the machine to accurately detect features is still inferior to its human counterpart (Antoniou and Potsiou, 2020). To balance the need to provide high accuracy data fast, researchers have proposed re-integrating the human into automated AI workflows. Such approaches have gained particular traction in the field of automated feature detection for humanitarian applications (Chen et al, 2019; Herfort et al., 2019), recognising the benefits of AI in collaboration with, rather than as a replacement for, human volunteers (Antoniou and Potsiou, 2020).
Here, the initial findings of an evaluation of a human-AI workflow known as Centaur VGI are presented. Centaur VGI utilises freely available Volunteered Geographic Information (VGI) to train an AI model, and verifies model output through human confirmation, editing or rejection; utilising these results to further refine the model. This research comprises the first user evaluation of such a system, utilising usability questionnaires and focus group discussions to evaluate speed, quality and usability compared with traditional VGI workflows. These findings will demonstrate the potential of such approaches to produce timely, high quality map data to support humanitarian activities.