That would be a fair point if the dog were given the opportunity to make both type-1 and type-2 errors. With the process behind a roadside traffic stop, we actually have no idea how often the dog would generate a false negative. We don't know its specificity in the real-world contexts where it actually matters. We just know that it is very, very sensitive... to something other than the presence of drugs.