[From: Bruce Nevin (Thu 93048 08:57:33)]
[Gary Cziko 930407.2214 GMT] --
"IN A FUNCTIONING CONTROL SYSTEM, IS THERE USEFUL INFORMATION
ABOUT THE DISTURBANCE IN THE PERCEPTUAL SIGNAL?"
Gary, given the pernicious ambiguity of "information," this seems
like a badly formulated question. Does the answerer mean
log(R/r) or "knowledge"?
Also, "useful" is ambiguous: useful to the control system or
useful to an observer, such as a designer of control systems, or
possibly a reorganization function.
How about:
IN A FUNCTIONING CONTROL SYSTEM, IS THERE INFORMATION ABOUT THE
DISTURBANCE IN THE PERCEPTUAL SIGNAL THAT IS USED BY THE CONTROL
SYSTEM TO CONTROL THE PERCEPTUAL SIGNAL?
[Allan Randall (930407.1600 EDT)] --
no one is claiming there is 100% of the disturbance information
in the percept. [...] the goal of the ECS is for there to be
*no* information about disturbances in the percept. However, the
control system's only source of information is via this percept,
so perfect control is not possible. [...] The percept is the
only information source the control system has. [...] Where do
you think the information is coming from to allow the system to
control against the disturbance? Do you think the system needs
*any* information at all about the world in order to control? If
not, I just have to shake my head. Magic.
Allan, these statements and others like them show that you
conceive of the ECS as controlling information.
The goal of an ECS is the instantaneous value r(t) with which
the instantaneous value p(t) is compared to produce an
instantaneous value o(t). The ECS does not get information from
this instantaneous value. That would require a function within
the ECS to register values of p(t) over time.
Since the ECS controls only the instantaneous value of p(t) at
instant t (continuously through time), it is incumbent on you now
to show how there is information in a single instantaneous value
of p(t).
···
************************************************************
* Yes, there is information in p . . . *
************************************************************
*************************************************************
* . . . but not for the ECS, only for an outside observer. *
*************************************************************
This outside observer of ECS A could be a higher-level ECS B. In
that case, the information in the signal p input to A would be
reduced for B to a single instantaneous value of p(t) input to B.
Obviously, this account is recursive e.g. for an outside observer
of B with respect to information about the information in A,
i.e. pattern in the changing values of p(t) input to B over
time.
In any case, the information in p is inaccessible as such to the
ECS into which p is an input. The ECS only "knows" the
instantaneous value of p(t), which it compares with the
instantaneous value of r(t), computing the instantaneous value of
o(t). The information in these signals is observable only over
time. The ECS has no capacity to observe any of them over time,
it "knows" only the present instant.
The appearance of information in o and information in p and
information in d and of a correspondence between information in d
and information in o is a byproduct of the ECS maintaining the
entrainment of p to r by means of varying o, instant by instant,
over time--that is, information preservation in o with respect to
d is a byproduct of control. When you understand control from
the point of view of an ECS (not an outside observer), using only
the capacities that are available to the ECS, you will see how
there can be a correspondence, and even in a sense a
`transmission' of information, without the information itself
being at any point relevant to PCT. If the information did not
correspond, were not `transmitted' (that is a metaphor, you
know), then it can only be because the control system is not
controlling p. But you would know that by observing the
instantaneous value(s) o(t) and r(t). To measure information
quantity or information rate in order to diagnose the performance
of a control system would be worse than the cowboy in the joke.
Asked how he could count the cattle so much faster than his
peers, he said "Easy! I count the legs and divide by four!"
So here's the proposition again: if you want the information
that is in p to be relevant to PCT, in the sense of being useful
to the ECS into which p is an input signal, then . . .
*********************************************
* SHOW HOW THERE IS INFORMATION *
* IN A SINGLE INSTANTANEOUS VALUE P(T). *
*********************************************
Bruce
bn@bbn.com