A doctor can’t tell if somebody is Black, Asian, or white, just by looking at their X-rays. But a computer can, according to a surprising new paper by an international team of scientists, including researchers at the Massachusetts Institute of Technology and Harvard Medical School.
The study found that an artificial intelligence program trained to read X-rays and CT scans could predict a person’s race with 90 percent accuracy. But the scientists who conducted the study say they have no idea how the computer figures it out.
I bet the AI is using the machine learning process of “symbolic regression”.
I don’t know why this is so mysterious. Every artist knows that the shape of the skull, cheekbones, eyebrow ridges and jaw is characteristic of races. Any reasonably good artist can show the race of a person with a simple charcoal sketch, not as a caricature but in a subtle way that is clear. Of course, there is plenty of overlap in shapes, especially in racially-blended groups like African-Americans (who average almost 25% European ancestry). https://www.science.org/content/article/genetic-study-reveal…
An X-ray can detect the shape of teeth, which is also different. (For example, the back of the front teeth is flat in Europeans but concave in Asians.)
AI facial recognition programs use bony landmarks and eye shapes every day of the week. The facial recognition programs construct realistic portraits of all races. Take a look at the Land’s End catalog which uses AI models, not live models, of all races.
If the scientists don’t understand how this is done, it’s because they aren’t artists and haven’t worked with facial recognition.
Using the word “racist” is pure click-bait. AI isn’t racist. It’s doing what it’s programmed to do by artists. No mystery at all.
You are not surprised to learn that poverty, nutrition, diet preferences, etc cause our bodies to develop differently. Or that AI is able to identify those differences.
When it comes to medical, those differences allow treatments to be customized one would hope for better outcomes.
The first thought that comes to mind is “Who cares?”
Generally, X-rays are not required for anything other than medical healthcare issues. Healthcare
issues are typically requiring rather intimate knowledge of patients by those providing the
services.
Do you really want healthcare providers to not be aware of the patient’s race and general
medical history?
The study found that an artificial intelligence program trained to read X-rays and CT scans could predict a person’s race with 90 percent accuracy. But the scientists who conducted the study say they have no idea how the computer figures it out.
Law of Large Numbers. Get enough data, you can see/figure out correlations.
You can do the same thing with EKGs, lab, and other data points, and have a certain degree of success. Doesn’t mean AI or medicine is racist.
could predict a person’s race with 90 percent accuracy.
And, of course, a statistic like that is assuming that there is some other method of assessing race which is 100% accurate to serve as the reference … which doesn’t exist.