What's the false negative rate?
So 24 in 25 flagged matches are wrong. So what? Without knowing the false negative rate, we still don't know enough to tell whether it's useful.
Hypothetically, if there are *zero* false negatives, this would still be a very useful system. If you have one suspect to identify in a crowd of 1000 faces, it's entirely worthwhile to have a computer just show you 25 faces to take a closer look at, rather than the full 1000.
Of course I know it won't be that accurate, but without knowing *how* accurate it is this "96%" figure *still* isn't enough to pronounce it useless.
And yes, I realise it will also victimise people based on skin colour - but let's be clear, that's an entirely separate issue, the cops don't need any automated help doing that anyway.