The official launch of WGIC’s latest report, “Geospatial AI/ML (GeoAI) Applications and Policies – A Global Perspective”, took place on April 13th, 2021, during AI for Good, an event organized by the International Telecommunication Union (ITU).
In her introduction, WGIC’s Executive Director Barbara Ryan quoted the American Congress, when she remarked that “artifical intelligence has the potential to disrupt every sector of society in both anticipated and unanticipated ways.” Under the title ‘Geospatial AI, Future Policies’, she and Arnout Desmet, Chair of WGIC’s Policy Development & Advocacy Committee, presented a host of prominent geospatial experts. Together, they highlighted crucial parts of the new report.
Lokendra Chauhan of Qen Labs is the main author of the GeoAI/ML report. In his overview, he explained how the methodology was executed in three parts. Firstly, Desk Research focused on the geospatial industry and its applications of AI and the overall AI/ML landscape. Next to this, the regulatory landscape of AI was assessed. Secondly, there were consultations in the form of interviews with WGIC members and external experts. These produced GeoAI use cases and their successes, lessons learned and ongoing concerns. Thirdly and lastly, Chauhan showed how the synthesis of the first two elements resulted in the report that was presented today.
Representatative Coverage and Definitions
With the aim of a globally representative coverage, the GeoAI/ML report takes note of the policy landscape in Australia, Brazil, China, the European Union, India, Israel, Qatar, Saudi Arabia, Singapore, South Korea, the United Arab Emirates, the United Kingdom and the United States of America. To be able to make valid comparisons, Lokendra Chauhan stated, one first needs to set the record straight about the definitions of both Artifical Intelligence itself and what would be described as GeoAI. Data input is typically restricted to geospatial data such as location, time (oftentimes in conjunction with global positioning or GNSS systems), spectral bandwidth and so forth. Expected answers or outcomes include proximity information, labels describing the object and other features of geographic information. Just like ‘regular’ Artificial Intelligence, GeoAI is rule-based. Chauhan sketched out what a GeoAI phrase would sound like in every-day language, using an example of object recognition: “Observed attributes imply this object is a ‘road’, a ‘farm’, a ‘vehicle’, etcetera.” Thus, the report highlights the unique needs for analyzing geospatial domain, including vector data analysis and also the dearth of specialized AI talent.
The report classifies Geospatial AI (GeoAI) current use cases as descriptive technologies, followed by predictive technologies for the medium term of up to 5 years and then prescriptive technologies to emerge beyond that; GeoAI systems by then are expected to prescribe specific solutions and even act automatically in many case. Chauhan: “You might want to summarize it in terms of Object Detection, which is what we are currently doing, going towards Predictions, like forecasting, resulting in Presciptions in the form of specific recommendations”. Of course, these trends imply an increase in quality, variety, coverage and frequency of input data and a decrease in cost.
On the Organization Level, the report recommends to seek alignment with the emerging consensus around the ethics and governance of AI. Geospatial industry members involved in this document feel that ‘self-regulation in good faith’ would increase trust. As for Government Policies on AI/ML, they should not restrict innovation. In order to work properly, they should be easily enforceable and adaptive to new developments in AI. Interestingly enough, the Geospatial industry also posed some recomendations on itself. “Classify use cases by the risk of potential harm to enable calibrated policy responses; Create tests and checklists for self-audit of GeoAI applications; Develop training datasets, benchmarks and tests for measuring GeoAI Performance.”
As part of the Virtual Launch of “Geospatial AI/ML (GeoAI) Applications and Policies – A Global Perspective”, an Expert Panel was assembled. Featured WGIC Members were Kumar Navulur, Senior Director of Strategic Business Development of Maxar Technologies, Jim Van Rens, Senior Vice President of RIEGL, Stephanie Leonard, Head of Traffic Innovation and Policy of TomTom and Siva Ravada, Vice President Product Development of Oracle.
All of these companies contributed to the report, the panelists took it on themselves to highlight different aspects of it. Maxar’s Kumar Navulur pointed to the shortage of data scientists: “While the demand for geospatial AI/ML is high, the risks due to the scarcity of sufficiently trained data scientists and engineers are also high relative to the consumer tech industry”. RIEGL’s Jim Van Rens placed geospatial data quality, and especially its accuracy and completeness in the heart of GeoAI/ML to be effective. “Data veracity, reliability and trust is critical. The need for global standards is paramount. As the technology and the data change – so too must the standards, which must be authoritative and relevant”. TomTom’s Stephanie Leonard dived into GeoAI/ ML applications, policies and future use of AI. She concluded her presentation with some ethical considerations: “Organizations should lead by example, and in good faith, build norms for ethical use of AI, so that policymakers, and citizens at large, have strong reasons to trust the geospatial industry.” Oracle’s Siva Ravada advocated a broader access to geospatial data for GeoAI/ML usage, model sharing and the development of no-code/ low-code tools built on Spatial AI technology. Ravada quoted from the report: “GeoAI can provide public agencies and businesses the ability to make decisions that will result in sustainable development and growth and the preservation of natural resources”.
External Expert Conclusions
Closing off the Expert Panel was one of the ‘external experts’ who contributed to the report: Shashi Shekar, McKnight Distinguished University Professor at the University of Minnesota, Minneapolis, MN, USA. Before answering the question ‘What’s special about GeoAI’, he emphasized the fact that we are not quite there yet: “The ‘machine’ is still learning”. All in all, according to Professor Shekar and the GeoAI report: “GeoAI is a highly interdisciplinary field bridging disciplines like computer science, engineering, statistics, and spatial science. As this field focuses on real-world problems, the impact on society and the economy is very high and critical.”
Download the report here
Watch the Virtual Launch of the GeoAI/ML Report at ITU’s AI For Good Event on YouTube