I would think that ideally, it would be a system where an AI would somehow have intimate knowledge of every human connected to its network. It would then make regulatory decisions for humanity based on that data on things like infrastructure, energy, criminal justice, trade, and monetary policy. The problem is that AI would lack emotion. Double edged sword, because it would lack ambition, jealousy, and greed, but would also lack empathy. In stories like the Matrix and Dues Ex, they tried to fix the problem by having a person at the core of the AI so that the AI could understand human empathy and individuality in order to get some counterbalance to its cold logic.
You'd most likely wind up with a dystopian surveillance state at the end of the day, though. That Atlantic article above has some interesting information.
