If the AI is designed to be an a self-motivated decision-making agent, some form of utility function will be already built in by its architects. One both hopes and fears that it will also be provided with a manner of updating and actualising that utility function in altering circumstances.