AI tool uncovers hidden biases in medicine
Researchers have developed two tools to detect hidden bias in datasets used to train medical artificial intelligence. These tools examine the data for unintended patterns that could lead to incorrect conclusions, threatening the accuracy of patient care and the reliability of models. A study conducted by Johns Hopkins University and the FDA demonstrated that the tool, called "Generalized Attribute Benefit and Detectable Bias Test" (G-AUDIT), identifies features that may cause AI to draw inaccurate inferences and aids in improving the reliability of clinical models. Data testing revealed that biases can stem from differences in imaging conditions, image quality, or incorrect data, which can result in inaccurate assessments of patient health. The goal of the tool is to make AI in healthcare safer and more effective, with the potential to expand its use to other fields.
Notice: This Is an AI-Generated Summary
Comments (0)