In yet another demonstration that artificial intelligence can detect things that humans fundamentally cannot, researchers have shown that AI may be able to determine race from X-ray scans, even though there is no discernible difference between human specialists in this regard.
Artificial intelligence detected race with an accuracy of about 90 percent using just X-ray and CT pictures; nevertheless, the experts are at a loss to explain how it can accomplish this feat.
The findings imply that significant caution must be exercised when employing AI in medical technology since it may still be able to describe supplied photos despite being considered anonymous, indistinct, or otherwise distorted.
The study’s findings were presented in a paper published in The Lancet Digital Health.
A deep-learning model was initially provided with CT and X-ray scans of many different body locations on diverse groups of people, with each image tagged with the person’s race. It was done so that the AI could be tested on its capacity to recognize images.
After that, the tags were removed, along with any potentially distinguishing characteristics, such as the color of the subject’s skin and hair. The artificial intelligence was given each photograph and tasked with determining the individual’s race.
Even without any distinguishing characteristics, artificial intelligence could determine a person’s race with an accuracy of almost 90 percent throughout the body.
The researchers wanted to be sure they covered all their bases, so they questioned whether the AI was playing a trick on them and utilized statistics to make the assumption based on covariables like body-mass index (BMI) or breast density may imply one race over another.
Even when the researchers eliminated all the potential covariables and only gave the AI datasets of people with comparable BMIs and body types, the AI could still correctly identify the participants’ races.
How is it even possible? The researchers, to tell the truth, are not certain.
Previous research has demonstrated that AI can detect race even when the picture has been significantly damaged or changed. This is not the first time AI has startled scientists by identifying race in virtually impossible settings.
The elimination of covariables, on the other hand, produces perplexing findings. The researchers have a theory that AI can recognize differences in the levels of melanin found in the skin of white and black persons.
These differences may be visible in CT and X-ray scans, but humans have never seen them. That is but one possibility, though, and the real job starts now with trying to comprehend the outcomes.
Regardless, the facts should raise major concerns regarding using artificial intelligence in hospitals and how they may behave differently to various races. This deployment of AI has been observed multiple times.
Sign up for our newsletter to get the best of The Sized delivered to your inbox daily.