They said "could", not "should".

I believe the point is that it's much easier to create a plausible justification than an accurate justification. So simply requiring that the system produce some kind of explanation doesn't help, unless there are rigorous controls to make sure it's accurate.