Maryland today | UMD co-leads $750,000 NSF, Amazon Project to tackle AI…

A satellite image of California farms (top) contrasts with an AI-generated crop map (bottom) that humans must examine and label for AI algorithms to be free from bias. Two boxes on the left show the areas where the labels appear in line with the satellite image; boxes to the right indicate areas where there are discrepancies. (Visual by Yiqun Xie. Images courtesy of ArcGIS World Imagery basemap, top, and USDA Cropland data layer, bottom.)

AI algorithms used to create crop maps, flood maps, road network maps and many more need ground truth labels to learn how to make predictions. Such labels are typically expensive and time-consuming to collect, often requiring an expert to go into the field to determine whether a pixel in satellite imagery should be labeled “corn” or “soybean.”

AI bias particularly affects people in regions with high poverty rates, such as sub-Saharan Africa, where a lack of knowledge of local cultures – especially crops grown by smallholders – leads to farmland being classified as sterile by algorithms more adapted to the agricultural dynamics that exist in high-income countries.

“Existing AI algorithms have no control over the quality of predictions at different places. They can and do compromise accuracy in some places while seeking a better result in other places, which injects bias in the results,” explained Xie, an assistant professor of geospatial information science.

Other consequences of AI bias, such as incorrectly assessing the extent of flooding in certain areas after a hurricane or natural disaster, or blocking medical supplies and other deliveries during crises, crop failures in these underdeveloped regions could go unnoticed in satellite data.

“Machine learning methods have the potential to advance and improve geoinformation products for agricultural monitoring; however, biases in maps can lead to biases in agricultural production estimates,” said Skakun, an assistant professor in the Department of Geographic Sciences with a cross appointment in the College of Information Studies. Skakun also works with NASA’s UMD-led Harvest Team, whose mission is to improve food security by accelerating the adoption of satellite-based agricultural monitoring.

After confirming the existence of location-based AI bias and measuring its negative societal impact, researchers are now developing a system through which experts, including those at NASA Harvest, can select an AI algorithm best suited to overcome the bias often associated with location. they wish to study and the type of map they are trying to create.

They plan to complete their “Advancing Deep Learning Towards Spatial Fairness” project by summer 2025, tackling the problem of AI bias in mapping, and possibly other issues.

“Currently, the issue of fairness is a barrier in many important areas where people want to deploy AI, so in a certain sense, it can also help accelerate the use of AI in different companies,” concluded Xie.

About Barbara Johnson

Check Also

IUB-AAUP deplores Rokita’s attacks on IU Professor Caitlin Bernard; Calls on administration to defend Dr Bernard – The Bloomingtonian

The following press release was sent to the Bloomingtonian this week: « July 26, 2022 …