Public Policy and Superintelligent AI: A Vector Field Approach