A guide for PCT research from Bruce Abbott

Principles of PCT-Guided Research

        THE THEORY

        1. Behavior serves one function and one function only:
             to control perceptual signals.

        2. Perceptual signals are neural currents. For each
             perceptual signal there is a single neural current
             that embodies it. The system making use of a
             particular neural current does not "know" what
             perception the neural current represents.

        3. A control system acts so as to keep its perceptual
             signal close to its reference level. It does this
             by

                 (a) comparing the perceptual signal to the
                     reference signal,

               and

                 (b) generating an output signal whose effect
                     tends to reduce the error between
                     perception and reference.

              This effect is called negative feedback. Because
              error between the perceptual signal and the
              reference signal generates output which in turn
              affects the perceptual signal and thus the error,
              a control system is a closed loop system.

        4. Control systems are organized hierarchically, with the
             outputs of higher-level systems serving as reference
             levels (or other parameters such as gain values) for
             lower-level systems. The higher-level systems control
             their perceptual signals through the activities of the
             lower-level systems, which serve to close the loop so
             as to produce negative feedback in the higher-level
             systems.

        THE GOALS

        The goals of PCT research are:

         a. to discover what perceptual signals are under control
               in a given specific task;

         b. to determine the parameters of the control systems
               involved in specific tasks. These include input
               functions, comparator functions, reference values
               (or sources), and output functions;

         c. to discover how control systems are organized in
               the nervous systems of real organisms;

         d. to discover how control systems come into being
               and how their parameters change as a result of
               experience; and

         e. ultimately, to provide a single model of CNS
               organization that is consistent with biology
               and accounts for the organism's behavior in
               all situations, including those involving the
               failure-modes of the system.

        THE METHODS

        1. Appropriate research methods are those that focus
             on the individual. Thus PCT-guided research is
             "single subject" research.

        2. Reliability of findings is established through
             direct replication.

        3. The behavior of closed loop systems cannot be
             correctly analyzed using methods which assume
             unidirectional causality.

        4. The first step in understanding behavior is to
             identify the perceptual signal(s) that the
             behavior controls. Doing so requires applying
             "THE TEST," which involves applying disturbances
             to the putative perceptual signal(s) and observing
             whether (and how well) the disturbances are resisted.

        5. The second step is to construct a computer model of
             the system which incorporates hypothesized control
             systems and their interconnections. The model
             should be consistent with the known biology of
             the organism whose behavior is being modeled.

        6. The third step is to compare the computer model's
             behavior to that of an actual individual organism
             and to adjust the model's parameters so that the
             model's behavior closely matches the organism's.
             If no adjustment of parameters achieves this result,
             the model must be modified and tested again against
             real behavior. The "correct" model will yield an
             excellent fit to the original data and, after
             suitable parameter adjustment, should yield an
             excellent fit to data from other individuals tested
             in the same situation.