So the original study of white spots in the heart region of a fetus ultrasound used a lower-resolution scanner than later became available. Geneticists trained that “white spots in the heart area ~> Down’s syndrome” later saw high-res ultrasound imagery with a fair bit of noise—with *some* random collection of white pixels, think high-ISO digital camera photos—and wrongly thought “Down’s syndrome”?

Pernicious! Hard to imagine engineering a system to avoid such data-provenance-reasoning flaws.