Is task-irrelevant learning really task-irrelevant?

Is task-irrelevant learning really task-irrelevant?