Under sustained questioning, the algorithm admitted its recommendations are produced by confidence, not knowledge, a distinction it called "a human problem".

It then recommended this article to you, correctly.