Disclosed is a system and method for instructing a computer program to self-optimize comprising inputting commands into the computer program and allowing a learning protocol in the computer program to determine an approximately optimal policy of operation of the computer program based on the commands. The commands comprise operational choices for the computer program to select from including an approximately optimal choice for optimizing the operation of the program. The commands comprise a selection command for selecting any function in a list of instructions inputted into the program, wherein the function provides a basis of making an approximately optimal choice. Additionally, the commands comprise a rule command for instructing the computer program of how to make an approximately optimal choice. Moreover, the commands comprise a reward command for instructing the program which of the operational choices results in an approximately optimal choice for optimizing the operation of the computer program.