<[Bill Leach 960121.19:41 U.S. Eastern Time Zone]
[Bob Clark (960121.1442 EST)]
That there is a difference between living and engineered control systems
is, I believe, without dispute.
It seems to me that the position that seems to be taken by most of us
concerning "caring" and the question about caring being the difference
between human and engineered systems is reasonably sound however, it does
rest upon only a particular aspect of the term as it is often applied
specifically to ourselves.
From the standpoint of an operating control system "caring" must mean
that we have a reference for a particular value for some perception.
Attempting to pursue this a bit further in the direction that I think you
are "coming from"...
We don't generally use the term with low priority control issues. We
also possibly DO use the term with very high priority control issues even
if we have never experienced a control error with respect to that issue.
That is, we are aware of what might occasionally be a sharp change in
our emotional perceptions for even imagined disturbance... even when we
"rationally" know that the disturbance is _only_ imagined.
We also "care" about issues for which we have little or no control
capability (even though we know we do not have a control capability with
respect to the issue).
The "intrinsics" have not been specified and even the suggestions have
been _very_ tentative (at least within the PCT community). However, in
a very real sense, every engineered control system has one or more
clearly identifiable "intrinsics". In simple control systems this is the
single reference set by the designer or user of the system. Thus talking
of not being able to add "intrinsics" to an engineered control system
does not make a lot of sense.
It is also possible to add reorganization to a control system (and this
has been done a time or two with various results). I suspect that Martin
may have the "highest level of experience" with such efforts.
There appears not to be a completely uniform belief even here on CSGNET
concerning the operation of the reorganization system though there does
seem to be a general agreement about certain characteristics.
For example, everyone seems to agree that the systems' operation must be
essentially random and that the system functions on the basis of some
sort of perception of the existence of error in other control loops.
Beyond that, opinions diverge. Generally, I think that most agree that
the reorganization system do not immediately affect the entire neural
structure but rather begins its operations quite localized to the
physical area that the "offending" control loop resides in. Many of us
then believe that the intensity, scope and maybe frequency of the
reorganizing increases with such things as intensity, persistence and
importance of the error to be corrected.
Also, some of us do not believe that the activities of the reorganizing
system is limited to reduction of error in the intrinsic control systems
only. Personally, I believe that any control loop with persistent error
will be reorganized.
I suspect that most of us believe that the reorganization system could be
simulated and would result in the sort of behaviour that we actually
observe from "frustrated" humans but that such a simulation is probably
"rather far down the road" from here.
As has been discussed many times, we do our simulations based upon the
principle that a single control loop can accurately emulate many
thousands of parallel control loops for our current purposes.
Reorganization happens to be one of those phenomenon where the number and
relative relationships between these multiple loops _IS_ an important
consideration.
Given the almost completely unimaginable complexity of the environment
that humans must deal with it would seem that the reorganization system
is highly effective. It appears to make few "mistakes" of significant
consequence (though in a single human it may well make many millions of
mistakes in a sort period of time).
In principle there is no reason why the ability to try random variations
in output could not be designed into a machine. I think that there are
several reasons such ability does not currently exist.
In the first place, a massive increase in complexity of the machine is
necessary (though with today's computing power that in itself is not as
much of an issue). Secondly, the control hierarchy design becomes much
more complex. The criteria for the operation of the reoganization system
is anything but obvious. Failures in any of these design issues can (and
likely will) have catastrophic results.
Additionally, most control systems are designed for operation is rather
well defined environments. Most disturbances can be and are predicted,
the systems equipped with appropriate perceptions (sensors, etc) and
fixed, preordained control functions. These systems work amazingly well
and have "served" us well. Intentionally designing in a capability to
make catastrophic errors is not exactly the top priority of most control
system engineers.
When a system does fail, a failure analysis is normally performed which
includes determining what environmental condition(s) resulted in the
control failure. Some analysis and maybe even testing is conducted to
determine if the capability of detecting and controlling these
environment disturbance is performed. Then a decision is made as to the
possibility and cost effectiveness of changes.
In computer based control systems there is indeed some measure of
self-modification being introduced but for the most part the use of
"random" processes is severely limited at this time but where used the
nature of such random processes _is_, I believe, very much consistent
with reorganization as it is generally understood in PCT.
... But they have no SENSE of IDENTITY since they are completely
lacking means with which to perceive themselves.
And in this, I at least, agree with you. Though we talk a great deal
about "self-image", "self-worth" and the like, the perception set that
we refer to as "self" is not at all well defined or understood. That a
similar perception set is lacking in any machine is doubtlessly true.
That such a set _can not_ exist in a machine is not so certain.
Our knowledge, experience and understanding of hierarchical control
systems is just too limited to make "flat" statements about what is and
is not possible.
... the Intrinsic Systems. Thus the learned systems will come to
respond to any event that may seem likely to affect the Intrinsic
System. Thus the primary element of CARING is the effect on the
Intrinsic System, in combination with the operation of the Learned
Hierarchical Systems.
While I agree that learning as you referred to it will have the effect
that the infant will likely create a functional hierarchy that operates
to reduce intrinsic error I missed the leap from that to the assertion
that the primary element of "caring" is the effect on the intrinsic
system ...
It is reasonable to consider error in any control loop (that is error as
in failure to control adequately) to always be an aspect of intrinsic
error at some point since the hypothesis postulates a "few" intrinsic
references at the very top as well as the idea that the system as a whole
routes references from the top to the bottom. Thus, a reference at a
comparator at the very lowest level must, by the hypothesis, be traceable
up the hierarchy to one or more intrinsic references. This path, by the
way will also take you up past the "self" (whatever that actually is).
I also find your following paragraph to effectively be saying nothing.
I happen to agree that HPCT does provide a "vehicle" for explaining
"emotions" and does so in what I think is the ONLY way that has ever made
any sense to me but I hardly agree that such concepts are "readily"
included unless by that you mean that one first really understands PCT as
an explanation for their own behaviour. Even at that, the PCT
explanation of "emotion" is rather superficial at best. The biological
chemistry of emotional response is itself at quite an infant state.
-bill