RE: [pcp-discuss:] Goal-directedness, control and freedom

From: Paulo Garrido (Paulo.Garrido@DEI.UMINHO.PT)
Date: Fri Mar 02 2001 - 15:01:03 GMT

  • Next message: Alexei Sharov: "Re: [pcp-discuss:] Goal-directedness, control and freedom"

    Perhaps a Control Engineering perspective may be of help...

    For an engineer to control an X, means that a variable or variables V in X
    are to be controlled. This means that the evolution in time of V is to
    satisfy some criterion C. This is referred as the "control problem". If the
    "control system" manages to solve the control problem, its ok and one can
    invoice the customer. Otherwise, back to work...;-)

    For engineers it is useful to classify (and split) control problems in
    - regulation problems, where one wants to maintain the controlled variable
    value constant, according to C, against perturbations effects;
    - servo problems, where one wants the controlled variable to follow,
    according to C, some (possibly unknown in advance) evolution of a reference
    variable (which is not constant).

    The rational behind the classification is that in regulation one doesnt care
    about the reference variable (which is assumed to be constant) and in
    "servoing" one doesnt care about perturbations.

    No real control problem is only "regulation" or only "servo", but the
    classification is useful, if the problem is separable (one may define a
    control for regulation and a control for servoing) or mainly falls in one of
    the classes.

    For what concerns strategies an engineer may apply, mainly two are
    recognized: feedback and feedforward.

    Feedforward or antecipation is based on knowing how a system reacts to
    commands or perturbations to generate a command which (presumably) will
    yield the behavior required by the criterion. It can be used in the servo
    problem and in regulation, if one knows or is able to measure the evolution
    of reference or perturbations.
    As a rule, feedforward is very sensitive to our ignorance of the system to
    be controlled. It breaks down with (low levels of) model-to-reality
    mismatch. Then one adds

    Negative feedback. Which under reasonable constraints has the capability of
    enforcing a system behavior in agreement with the criterion, compensating
    perturbations by measuring their effects - in which one may include our
    ignorance about the system.

    In situations involving growth (of biomass, of an organization) one may also
    use positive feedback as a strategy. (Positive feedback earned a bad
    reputation because it leads easily to unstable systems. But note that in
    growth the criterion calls for instability. Note also that negative feedback
    is not a sufficent condition of stability.)

    Is goal-directedness (as expressed by the criterion) necessary to the
    control concept? In engineering, yes. In a mathematical sense, no. If one
    aplies a "control system" to a "system to be controlled", then (if one is
    able to send the invoice...;-) a "controlled system" with less *potential
    behavior variety* was created. Less than the system to be controlled! So, I
    think that one may talk of control whenever "potential behavior variety" is
    diminished even if a goal is not distinguishable.

    (One should not forget that "less variety at the system level" may mean
    "more variety at the meta-system level").

    Paulo Garrido

    ========================================
    Posting to pcp-discuss@lanl.gov from Paulo Garrido <Paulo.Garrido@dei.uminho.pt>



    This archive was generated by hypermail 2b29 : Fri Mar 02 2001 - 14:59:05 GMT