AI can do something that is impossible for human doctors

AI-powered deep learning models can identify the ethnicity of a human by simple X-rays, something that even doctors can’t do. This new finding raises concerns from experts, due to possible racial bias in diagnosis and evaluation and medical treatment.

It all started when a team of researchers did trained his AI using hundreds of thousands of X-ray images existing labels labeled with the patient’s ethnicity details. Indeed, the researchers “fed” information about humans to the AI, but the goal was not to recognize them during exams.

It turned out, however, that artificial intelligence is incredibly good at this. AI could predict a patient’s ethnic identity with surprising accuracy, even when scans were done by people of the same age and sex. Levels of the 90% accuracy with some groups of images.

Scientists aren’t sure why the AI ​​system was able to identify ethnicity with such great accuracy from images that don’t contain such information, at least not on the surface. Even when limited information is provided, such as removing clues to bone density or focusing on a small part of the body, the models still performed surprisingly well in guessing it.

Our finding indicates that AI can accurately predict self-reported human ethynia, even from damaged, cropped, and noisy medical images, often when clinical experts can’t. [Questo] creates a huge risk for all model implementations in medical imaging“, write the researchers.

This would normally not be a problem, but since AIs are known to currently have racial biases due to the set of images they are trained with (predominantly of Caucasian ethnicity) it could become so.

Read also  where to watch live today online for free on Facebook