Action is an evolutionary process in which a cybernetic system adapts the environment to itself.
The flow of information is reversed. The criterion became autonomous and is now directed at the original source, the environment. At first, the system will go about its objectives in a playful way or on a trial-and-error basis. In the course of this, it will learn what actions are successful and what aren't. The selection of accidentally successful actions will take place at a later point in the model. The potential success of actions can be predicted. The model turns into the universal source, and actions become more targeted. At first, the selection via trial and error shifts to the model. Later, the perfected higher-order model, as initial value, offers the action leading to a desired environmental condition.