Regulation
Given that evolution implies goal-directedness, we now need to study how goals can
be achieved in practice. This is the subject matter of cybernetics, the science of
governance, communication and control [Ashby, 1964; Heylighen & Joslyn, 2001].
Cybernetics has resolved the old conundrum of how to model purposive action
without contradicting causality by introducing circular processes [Rosenblueth,
Wiener & Bigelow, 1943; Heylighen & Joslyn, 2001]. Goal-directed behavior is
understood as a process of control or regulation that reduces any deviation from the
goal by means of a negative feedback loop [Powers, 1973]. It consists of the
following components:
1) perception, i.e. the process whereby an agent gathers information about its
situation;
2) evaluation, whereby the agent compares the perceived situation with the ideal
or desired situation, i.e. its goal;
3) decision, whereby the agent chooses the action or strategy most likely to
minimize the difference between the perceived situation and the goal;
4) action, whereby the agent intervenes in the situation;
5) feedback, whereby the agent re-assesses the new situation as changed by the
action and by possible outside perturbations.
If the action was successful, the difference or deviation between perceived situation
and goal will have diminished. Therefore, the feedback is negative: it reduces
deviations from the goal state. Phase (5) (feedback) is equal to phase (1) (perception)
of a new feedback loop, intended to reduce the remaining differences between
situation and goal.
In this way, the agent will constantly be monitoring the result of its actions,
intervening whenever necessary to come closer to, or remain at, its goal state. Further
intervention is necessary when the previous action was insufficient or unsuccessful.
Failure to achieve the intended result may be due either to an error made by the agent
(e.g. miscalculating the effect of the action, or executing it clumsily), or by a
hindrance originating in the environment (e.g. appearance of a predator or obstacle,
unforeseen change in the conditions). All these phenomena that obstruct or endanger
the achievement of the goal can be subsumed under the generic term disturbance. The
beauty of the cybernetic regulation loop is that the nature of the disturbance does not
really matter: whatever the origin of the deviation from the desired course, it will
merely trigger another corrective action. As long as actions are more likely to
decrease the deviation than to increase it, a sufficiently long sequence of actions is
bound to dependably advance the agent towards its goal.
This negative feedback loop is a robust method to deal with uncertainty. The
environment of an agent is normally complex, dynamic and to an important degree
unpredictable. This means that the agent will sooner or later be confronted with a
problem, i.e. an unintended deviation from its goal. But whether this disturbance was
foreseen or not does not matter for the mechanism of regulation: the only thing that