How will we know when human level AI has been achieved?

When human level intelligence creates human level AI, said AI will be able to improve and create new AI's... so the theory goes, the answer is we will know, because it will easily be able to take control from humans.

I'm not certain I agree to all points though; this would assume it's not only human level but "human nature". All our current AI's are designed for just one purpose alone in mind, and while it's likely a human level AI would be more generic in nature, less human-programmed and more self-learning, there's still little reason to assume it wouldn't come with human built-in limits.