The Greeks observed these limitations thousands of years ago. Below is an excerpt from Plato's "Theaetetus":
Socrates: That is certainly a frank and indeed a generous answer, my dear lad. I asked you for one thing [a definition of "knowledge"] and you have given me many; I wanted something simple, and I have got a variety.
Theaetetus: And what does that mean, Socrates?
Socrates: Nothing, I dare say. But I'll tell you what I think. When you talk about cobbling, you mean just knowledge of the making of shoes?
Theaetetus: Yes, that's all I mean by it.
Socrates: And when you talk about carpentering, you mean simply the knowledge of the making of wooden furniture?
Theaetetus: Yes, that's all I mean, again.
Socrates: And in both cases you are putting into your definition what the knowledge is of?
Theaetetus: Yes.
Socrates: But that is not what you were asked, Theaetetus. You were not asked to say what one may have knowledge of, or how many branches of knowledge there are. It was not with any idea of counting these up that the question was asked; we wanted to know what knowledge itself is.--Or am I talking nonsense?
The other one (also perhaps relevant to the general topic of this thread): WYSIATI (What You See Is All There Is).
Also, seeing computers perform tasks they haven't done before can convince people that the model behind the scenes is closer to AGI than it really is. The fact that deep neural networks are very hard to decifer only furthers the mystical nature of the "intelligence" of the model.
Also, tasks like playing starcraft are very impressive, but are not very close to true AGI in my opinion. Perhaps theres a more formal definition that I'm not aware of, but in my mind, AGI is not being good at playing starcraft, AGI is deciding to learn to play starcraft in the first place.
That's my 2 cents, anyways.