Any test that humans can pass and AIs cannot is a stepping stone on the way to AGI.

When you run out of such tests then it's evidence that you have reached AGI. The point of these tests is to define AGI objectively as the inability to devise tests that humans have superiority on.