Re: So what could go wrong.
The entire idea of the D-PO is dangerously misguided, right from the name. Being a decent police officer – and there are many, even in the US – is enormously more complex than image recognition and querying databases. It's one of a handful of professions at the very top of the list for requiring nuanced human capabilities we don't understand well, such as evaluating the emotional states of multiple people in a charged situation.
"Scientific policing" tools have a very mixed record. Fingerprint and DNA evidence mostly seems to be reliable if the lab is rigorous1 and ethical2, and if applied in a scientific manner and not overrated when described to judges and juries. Some other forensic techniques are highly suspect (e.g. facial reconstruction from skulls) or outright fraudulent.
Elevating one of these tools – particularly one that's overly general and poorly understood – to the status of "police assistant" or, worse, "police officer", is an astonishingly bad idea.
1Various studies have shown wildly differing results from multiple labs when evaluating partial fingerprint matches, for example.
2There have been numerous documented cases of labs faking results to give police and prosecutors false evidence.