I get it and I have no disagreement with qualifying the numbers. It's a very complicated issue, but, if it is in fact true, when controlled for all other factors, if a particularity defined demographic group engages in disproportionate amount of criminal activity (less or more), then wouldn't you expect that that would filter down to the individual interactions?
If an individual police officer is choosing who to frisk (or interact with), and the only things they have to go on are visible superficial characteristics (i.e. age, gender, ethnicity, clothing, etc.), and the global fact that those characteristics are statically correlated with more or less crime, that would be impossible for an individual to control their biases.
We know this is a painful exercise, because people are not just those superficial characteristics, and it's unfair for an individual to be singled out for those characteristics - but that's the ONLY information available to the officer.
The officer cannot control their bias, because if they try, they will either over or under compensate and because they are human. It's why we have double-blind trials and the scientific method - we know even well-meaning humans cannot control their biases. And this isn't an example of racism, because it could clothing or gender that trigger the frisk (I guarantee you that men are stopped more than women - WHY?!?!?), but aspect of human cognition. The only way to control for that is by introducing a non-biased random decision maker. For example, in airport security, there will be a device that will randomly flag passengers for extra screening. Perhaps something like that should be the case in these 'stop-and-frisk' policies? But even that has limited value. If a particular neighborhood with issues of crime, is dominated by an ethnic group, even random sample will involve disproportionate 'harassment'.
A while back Sam Harris had a debate with an airport security expert on profiling in airport security[1]. Sam Harris advocated for profiling and security expert was not. Sam Harris was 100% wrong and didn't admit it. The security expert talked about how proper airport security should work (and the problems with profiling and how to control for that). That debate has analogues to this conversation, because police should adopt some of those strategies because profiling is socially painful and breeds resentment and has limited success ... and individuals will not make the right decision in context.
[1]https://samharris.org/to-profile-or-not-to-profile/