The problem is that the people who will put this in place rate capability on a linear scale: in their view the ability to write software is sufficiently magic, so such an ability is obviously good enough to recognize criminals. From their perspective, there are hurdles to be crossed (like probable cause) and an AI flagging a suspect feels like a magical intelligence crossing those hurdles and allowing them to continue in the process.

They don't validate the results of their fellow officers, or the validity of warrants, or anything else that predicates an arrest. Why would they start with this?