Using the same inputs and outputs as a human operator, the model views the screen and decides on a series of mouse and keyboard actions to reach an objective.