It's fascinating how sample size differs based on who wants what policy to be enacted. What are the false arrest rates due to people just looking at a picture and thinking "That's the one"?
Furthermore, with a large black population like Baltimore, if they mostly ran the software on black people, then yeah the false IDs will also be mostly black people.
I don't want to let the cops off the hook here. The previous commenter suggested a mechanically plausible way: "decent software used in a racist way" would have the above problem
If you aren't aware of the specifics, research "qualified immunity" , a legal doctrine that was not passed into law but invented by the courts to shield police from the consequences of violating your civil rights.