You should probably view this in context. Any “true” AI – like a sentient, intelligent being – that is much better and more durable than us, will likely have a stream of consciousness. Part of that includes continuously pulling diagnostic information and evaluating it; sort of like you will know if you suddenly got a leg cramp in the next milliseconds of it happening.
Truth is, having an artificially sentient entity will mean a lot of not-so-energy-efficient activities happening all the time; it cannot just react to things, it has to take the initiative and continuously process sensory information… sort of like us.
To create a software that fixes itself you’ll need a ton of specifications, goal-oriented programming, and you also must have a feedback mechanism (which answers the question “is the result of this function correct”?). I am not saying it’s impossible; it definitely is possible. And you’ll also probably need some logic programming as well (Prolog).
