AI Just Signed Its Own Death Warrant

 I saw this mentioned at Instapundit.  It is somewhat old news. 5/18/22 Boston.com:

Scientists at Harvard and MIT are part of an international team of researchers who found that artificial intelligence programs can determine someone’s race with over 90% accuracy from just their X-rays.

The problem is that no one knows how the AI programs do it.

“When my graduate students showed me some of the results that were in this paper, I actually thought it must be a mistake,” Marzyeh Ghassemi, an MIT assistant professor and coauthor of the paper analyzing the subject, told The Boston Globe. “I honestly thought my students were crazy when they told me.”

The researchers wrote in the study that many studies have shown that AI diagnostic systems seem to be using race in their considerations for diagnosis and treatment, to the detriment of patient health.

In the paper, they gave an example in which an AI program that examined chest X-rays was more likely to miss signs of illness in Black and female patients.

Thus, the aim of the study, which was published Wednesday in the medical journal The Lancet Digital Health, was to determine the degree to which AI systems can detect race from medical imaging, and to find out more about how these AI systems are detecting race. 

To do this, the research team trained AI systems for the study using standard data sets of X-rays and CT scans of different parts of the body.

Each image was labeled with the person’s self-reported race, but contained no obvious racial markers, such as hair texture or skin color, or medical racial trends, such as BMI or bone density. The team then fed the AI systems images without race labelling.

The researchers found that the AI systems were somehow able to determine the race of the person who the images were taken from with over 90% accuracy. The AI systems were even able to detect race from medical images regardless of what part of the body the image was of.

 Now if this is actually making decisions to the patient's detriment, this is a problem.  But that the AI was 90% of the time correectly guessing patient race is unsurprising.  From American Journal of Human Genetics (Dec. 29. 2004):

We have analyzed genetic data for 326 microsatellite markers that were typed uniformly in a large multiethnic population-based sample of individuals as part of a study of the genetics of hypertension (Family Blood Pressure Program). Subjects identified themselves as belonging to one of four major racial/ethnic groups (white, African American, East Asian, and Hispanic) and were recruited from 15 different geographic locales within the United States and Taiwan. Genetic cluster analysis of the microsatellite markers produced four major clusters, which showed near-perfect correspondence with the four self-reported race/ethnicity categories. Of 3,636 subjects of varying race/ethnicity, only 5 (0.14%) showed genetic cluster membership different from their self-identified race/ethnicity.

This should be no surprise.  What we identify as race is not terribly subtle.  It would be startling indeed if skin color, lip shape, hair and eye color. etc. that is plainly visible by sight had no genetic origin.

Part of the problem driving the current insanity is that after the Holocaust what had been a legitimate line of scientific inquiry became hopelessly intertwined with German Rassenkunde (racial science). That decent people would choose to distance themselves from all that makes perfect sense. 

 But forensic anthropology is a science. You can look at a skeleton and determine with some certainty what sex this person was; it is not something uncertain or dependent on how you feel you should be regarded.

Metric and morphological techniques employed by forensic anthropologists for determination of race are reviewed. Included are several studies which examine cranial morphological techniques such as presence of the oval window of the inner ear, which occurs more frequently in Whites than in Native Americans; or the shape of the alveolar region which distinguishes between Asian, African, and North American Indian groups. A table of common cranial morphologic traits is presented. Metric techniques have also been used to determine race from the skull. Regression equations derived from measurements of the cranial base indicate a 70-90% accuracy for classifying Blacks and Whites, while multivariate discriminant functions for discriminating Blacks, Whites, and Native Americans correctly classify 82.6% of the males and 88.1% of the females. FORDISC, a computer program developed at the University of Tennessee, is another metric technique reviewed that not only distinguishes Whites, Blacks, and Native Americans but also male Hispanics, Chinese, and Vietnamese. Platycnemia, femoral curvature and other morphological attributes of the post-cranial skeleton may be used in support of a racial determination; however, several investigators have turned to post-cranial elements not only to use in support of cranial findings but for use when cranial information is not available. As a result, several discriminant functions from measurements of the pelvis, femur, tibia or combinations of these elements have been developed. Accuracy for these techniques varies from 57% to 95%, depending on the sample and technique used.