They're just behind. And the ACLU "bias" study was thin and unscientific. Data sets and weights could have bias, but that bias can also be controlled. "Facial recognition" does not have bias.