The definition of a narrow AI isn't very complicated. It's a computer program which can solve a task, for example, a text to speech software. According to the capabilities in doing so, there are different narrow AI systems available. Human-level narrow AI is a powerful one that comes close to human capabilities in solving problems.
What remains an open question is, if a narrow AI can solve problems, what exactly is the task of so-called Strong AI systems? A hypothesis is, that a strong AI can't solve a problem but its task is to determine if the opponent is able to do so. A captcha test, which checks if the opponent is a human or a bot, is an example of a strong AI. Am I right?