Does it matter if an algorithm can’t explain how it knows what it knows?