What humans are very good at is visual perception, object recognition and cognitive processing. Also audial processing. You can listen to a song, isolate single instruments, or the voice, or change the entire tune, and replay it. Heck, you can take a piano tune and re-imagine it for an entire orchestra in your head. You can READ a book and VISUALIZE everything that is described in it. All the result of millions of years of evolution.
What humans are not very good at is maths, or rather cognitive maths. We can't solve complex equations when we are asked to, yet our brain solves complex equations implicitly under the hood when it does stuff like visual analysis, object recognition, etc...
Just imagine a Google search that is actually intelligent. A search that you could interact with like you would interact with a normal person. You show it ONE picture of an object (let's say the FRONT of the USS Enterprise from TOS), and it recognizes that object, and the results are pictures of the USS Enterprise from TOS. No image from TMP, no image from Abramstrek. And the ship from all sides, not just the front. And no duplicate results. And no images that have absolutely nothing to do with it.
A human can do all that with enough time. A computer, right now, can't even do that, not even with enough time. Because the algorithms for object recognition, image recognition, knowledge conjunction that are hardwired into our brains, are extremely complex and cannot be replicated properly right now.
But the point is: you do not need a self aware AI to achieve this. You need an AI, you need a complex AI. But it does not have to be sentient or self aware to do such a task.