HPCT and MOL

Bruce_Gregory6 · January 5, 2010, 1:42pm

[From Bruce Gregory 92010.01.05.1340 GMT)]

[From Bill Powers (2010.01.04.1305 MST)]

I'll be interested in your comments on the MOL paper I sent you.

Bill,

My own experience with stress focuses less on conflict than it does on a lack of control. I suspect I am not unique. The world has been very unpredictable of late, and things that you once could rely on, you no longer can. This is a good recipe for loss of control. This loss of control seems to be accompanied by a feeling I call anxiety or depression. Can this be modeled as internal conflict? I'm sure it can, but it is not obvious to me that it needs to be.

When it comes to conflicts, the brain seems to have a way to deal with them; activated neural circuits are able to inhibit potentially conflicting circuits. For example, when a cell in the visual cortex linked to detecting a horizontal edge fires, it inhibits the firing of nearby cells linked to detecting a vertical edge. Such a mechanism may exist at fairly high levels. Another way to deal with conflicts is to prevent them from occurring in the first place by ignoring or discounting evidence that conflicts with a primary goal. Republicans have become very adept at this. Do CBO projections show that health care costs will increase under a Democratic proposal? Praise the CBO for its independence and integrity. Do CBO projections show that health care costs will decrease under a Democratic proposal? Ignore the projections and label the CBO as a bunch of government bureaucrats. Again, I'm confident that "confirmation bias" can be modeled in terms of conflicting control systems in a hierarchy. Does it have to be? I'm not so sure.

I have a very limited experience with talk therapy. I got a few very helpful suggestions about how act, which I tried with considerable success. I soon got bored with talking about myself, however, and ended the sessions. My experience is that my brain is adept at making up stories to justify my actions. Changing my actions via changing my goals seems to be more effective for me than changing the stories I tell. When the actions change, the stories seem to follow along. I recently saw a meta-study of cognitive-behavioral therapy, one of the techniques that regularly shows good results. The meta-study showed that the behavioral modification component does most of the heavy lifting. Since I have had no experience with MOL, I can't give my reactions to it.

While emotions can certainly be a consequence of efforts to control, I think they may also play a role in establishing goals. One piece of evidence comes from the neuroscience. In at least one case, disrupting the link between the prefrontal lobes (normally thought of as the seat of planning) and the emotional center of the brain led to an individual who could describe alternatives but could not choose which alternative to act on. I know that this sounds very old-fashioned, but the possibility that some goals are established on the basis of their "emotional appeal" seems to fit my experience. This is especially true when the goal is otherwise "irrational." Nevertheless, one can always model any goal as a reference level established by a higher-level control system.

Thanks for the articles. I found them enjoyable and thought provoking. They do confirm one of my prejudices, however. I can imagine an HPCT system functioning perfectly well in world in which there are no other HPCT systems. In this sense, HPCT seems to me to be a little too-self contained to be completely convincing. The problem for me stems from the fact that reorganization seems to be the only mechanism for introducing truly novel goals into the system. I like your new flat-screen TV. Will I buy one for myself? Only if I have a pre-existing goal to buy whatever Bill Buys or if reorganization leads me to become determined to watch the Superbowl on a 50" Samsung. At least this is the way it seems to me.

Nothing I have said should be seen as constituting an objection to your reasoning. I merely want to suggest that other mechanisms may also be involved.

Bruce

Bill_Powers1 · January 6, 2010, 1:11am

[From Bill Powers (2010.01.05.1213 MST)]

Bruce Gregory 92010.01.05.1340 GMT) --

BG: My own experience with stress focuses less on conflict than it does on a lack of control. I suspect I am not unique. The world has been very unpredictable of late, and things that you once could rely on, you no longer can. This is a good recipe for loss of control. This loss of control seems to be accompanied by a feeling I call anxiety or depression. Can this be modeled as internal conflict? I'm sure it can, but it is not obvious to me that it needs to be.

BP: One can't generalize problems like this, but look at it from another angle. If an unresolved conflict between goals exists, the result is an inability to achieve either of the two goals. The unpredictability of the world is a problem only if one wants to predict it or rely on it, and that's not a conflict, it's just a problem. It could become a conflict if for some reason you wished not to have to predict the world or rely on it, so you find yourself simultaeously wanting something and not wanting it. A true conflict is an impossibility, like A = not-A. I want to experience A (for one set of reasons) and I want to not experience A (for different reasons). The problem is that you can't simultaneous experience and not experience the same thing. You can't make one perception match two different reference levels at the same time.

BG: When it comes to conflicts, the brain seems to have a way to deal with them; activated neural circuits are able to inhibit potentially conflicting circuits. For example, when a cell in the visual cortex linked to detecting a horizontal edge fires, it inhibits the firing of nearby cells linked to detecting a vertical edge. Such a mechanism may exist at fairly high levels. Another way to deal with conflicts is to prevent them from occurring in the first place by ignoring or discounting evidence that conflicts with a primary goal. Republicans have become very adept at this. Do CBO projections show that health care costs will increase under a Democratic proposal? Praise the CBO for its independence and integrity. Do CBO projections show that health care costs will decrease under a Democratic proposal? Ignore the projections and label the CBO as a bunch of government bureaucrats. Again, I'm confident that "confirmation bias" can be modeled in terms of conflicting control systems in a hierarchy. Does it have to be? I'm not so sure.

BP: Yes, we do have mechanisms for resolving conflicts (good or bad mechanisms), and normally they work for us. That's not what brings people to therapists. People seek help when they want to do something and can't bring themselves to do it, even though they know how. In MOL therapy what often happens is that reasons for not doing it show up, good reasons. One side of the conflict is visible, the other side is considered (at the moment anyway) "bad" and is out of awareness. A lot of the time the conflict is perfectly visible but the person is trying to reorganize at the wrong level, so the conflict just changes forms.

BG: I have a very limited experience with talk therapy. I got a few very helpful suggestions about how act, which I tried with considerable success. I soon got bored with talking about myself, however, and ended the sessions.

BP: In MOL therapy the therapist doesn't give advice or suggestions, or analyze and interpret, or prescribe, or have insights into the client's problems. It's not a diagnose-and-treat sort of process. The therapist simply helps the client explore the structure of the problem to find out where it's coming from and mostly to get reorganization working in some new places where it might do more good. As Tim Carey puts it, the client gets the client well. Get hold of Tim Carey's book: The Method of Levels, or how to do psychotherapy without getting in the way.

BG: My experience is that my brain is adept at making up stories to justify my actions. Changing my actions via changing my goals seems to be more effective for me than changing the stories I tell. When the actions change, the stories seem to follow along. I recently saw a meta-study of cognitive-behavioral therapy, one of the techniques that regularly shows good results. The meta-study showed that the behavioral modification component does most of the heavy lifting. Since I have had no experience with MOL, I can't give my reactions to it.

BP: I rather doubt the truth of the behavior-mod idea. Behavior does change when you reorganize, but changing the behavior isn't the point; it's changing the results that matters, now how you behave to do it. As we know in PCT, behavior changes with every disturbance; it's the controlled variable that matters to the control system.

BG: While emotions can certainly be a consequence of efforts to control, I think they may also play a role in establishing goals.

BP: Seeking or avoiding them certainly does. People don't like to be sad and do like to be happy, so they do things that result in nice feelings and not bad ones. I don't know if emotions themselves have any particular power; it seems to me that they are more consequences than causes.

BG: One piece of evidence comes from the neuroscience. In at least one case, disrupting the link between the prefrontal lobes (normally thought of as the seat of planning) and the emotional center of the brain led to an individual who could describe alternatives but could not choose which alternative to act on. I know that this sounds very old-fashioned, but the possibility that some goals are established on the basis of their "emotional appeal" seems to fit my experience. This is especially true when the goal is otherwise "irrational." Nevertheless, one can always model any goal as a reference level established by a higher-level control system.

BP: Yes, and that's my main point. Emotions arise when you want to do something. Bad ones get more intense when something prevents the action from succeeding. I just tried to put the simple facts into PCT: feelings are perceptions; an urge to do something comes from an error signal which is related to reference signals. I don't think any more than that is needed -- is it? When you stop thinking of emotions as being able to cause anything, as in SR theory, the PCT view makes more sense, I think.

BG: Thanks for the articles. I found them enjoyable and thought provoking. They do confirm one of my prejudices, however. I can imagine an HPCT system functioning perfectly well in world in which there are no other HPCT systems. In this sense, HPCT seems to me to be a little too-self contained to be completely convincing.

BP: Well, speaking literally, people are pretty self-contained, aren't they? I'm not sure what you mean here. Too independent of the environment? Too much inner-controlled and not enough subject to external control? Too mechanistic and not creative/free enough?

If you just recognize that higher levels of perception include other people and your relationships with them, and that when one person controls his own perceptions, the actions he uses to do that affect other people, there is plenty of basis in HPCT for social relationships.

BG: The problem for me stems from the fact that reorganization seems to be the only mechanism for introducing truly novel goals into the system. I like your new flat-screen TV. Will I buy one for myself? Only if I have a pre-existing goal to buy whatever Bill Buys or if reorganization leads me to become determined to watch the Superbowl on a 50" Samsung. At least this is the way it seems to me.

BP: Well, what other sources of novel organization are you looking for? Maybe we need to find a place for them in HPCT.

BG: Nothing I have said should be seen as constituting an objection to your reasoning. I merely want to suggest that other mechanisms may also be involved.

BP: Fine. Propose some!

By the way: the BP: and BG: stuff is there because I've notice that some people transmit plain text without styles, and the quoted materials simply aren't identified as such That makes for difficult reading. I'm trying to set an example. When I remember to,

Best,

Bill P.

Bruce_Gregory6 · January 6, 2010, 1:51pm

[From Bruce Gregory (2010.01.06.1350 GMT)]

Thanks , Bill, very helpful. Just a few comments.

[From Bill Powers (2010.01.05.1213 MST)]

Bruce Gregory 92010.01.05.1340 GMT) –

BP: Yes, we do have mechanisms for resolving conflicts (good or bad mechanisms), and normally they work for us. That’s not what brings people to therapists. People seek help when they want to do something and can’t bring themselves to do it, even though they know how. In MOL therapy what often happens is that reasons for not doing it show up, good reasons. One side of the conflict is visible, the other side is considered (at the moment anyway) “bad” and is out of awareness. A lot of the time the conflict is perfectly visible but the person is trying to reorganize at the wrong level, so the conflict just changes forms.

BG: That’s interesting. Since I do not have a problem doing something that I know how to do, MOL is clearly not appropriate for me.

BG: I have a very limited experience with talk therapy. I got a few very helpful suggestions about how act, which I tried with considerable success. I soon got bored with talking about myself, however, and ended the sessions.

BP: In MOL therapy the therapist doesn’t give advice or suggestions, or analyze and interpret, or prescribe, or have insights into the client’s problems. It’s not a diagnose-and-treat sort of process. The therapist simply helps the client explore the structure of the problem to find out where it’s coming from and mostly to get reorganization working in some new places where it might do more good. As Tim Carey puts it, the client gets the client well. Get hold of Tim Carey’s book: The Method of Levels, or how to do psychotherapy without getting in the way.

BG: In my case, the advice involved doing something I was perfectly capable of, and comfortable, doing. I tried it, and it worked. Again MOL seems aimed at problems that I fortunately do not have!

BG: My experience is that my brain is adept at making up stories to justify my actions. Changing my actions via changing my goals seems to be more effective for me than changing the stories I tell. When the actions change, the stories seem to follow along. I recently saw a meta-study of cognitive-behavioral therapy, one of the techniques that regularly shows good results. The meta-study showed that the behavioral modification component does most of the heavy lifting. Since I have had no experience with MOL, I can’t give my reactions to it.

BP: I rather doubt the truth of the behavior-mod idea. Behavior does change when you reorganize, but changing the behavior isn’t the point; it’s changing the results that matters, now how you behave to do it. As we know in PCT, behavior changes with every disturbance; it’s the controlled variable that matters to the control system.

BG: I’m sorry if I was unclear. The critical point is changing the goals. The behavior follows as you deal with disturbances.

BG: While emotions can certainly be a consequence of efforts to control, I think they may also play a role in establishing goals.

BP: Seeking or avoiding them certainly does. People don’t like to be sad and do like to be happy, so they do things that result in nice feelings and not bad ones. I don’t know if emotions themselves have any particular power; it seems to me that they are more consequences than causes.

BG: One piece of evidence comes from the neuroscience. In at least one case, disrupting the link between the prefrontal lobes (normally thought of as the seat of planning) and the emotional center of the brain led to an individual who could describe alternatives but could not choose which alternative to act on. I know that this sounds very old-fashioned, but the possibility that some goals are established on the basis of their “emotional appeal” seems to fit my experience. This is especially true when the goal is otherwise “irrational.” Nevertheless, one can always model any goal as a reference level established by a higher-level control system.

BP: Yes, and that’s my main point. Emotions arise when you want to do something. Bad ones get more intense when something prevents the action from succeeding. I just tried to put the simple facts into PCT: feelings are perceptions; an urge to do something comes from an error signal which is related to reference signals. I don’t think any more than that is needed – is it? When you stop thinking of emotions as being able to cause anything, as in SR theory, the PCT view makes more sense, I think.

BG: My problem is with how goals are established. Once they are established, your explanation makes perfect sense.

BG: Thanks for the articles. I found them enjoyable and thought provoking. They do confirm one of my prejudices, however. I can imagine an HPCT system functioning perfectly well in world in which there are no other HPCT systems. In this sense, HPCT seems to me to be a little too-self contained to be completely convincing.

BP: Well, speaking literally, people are pretty self-contained, aren’t they? I’m not sure what you mean here. Too independent of the environment? Too much inner-controlled and not enough subject to external control? Too mechanistic and not creative/free enough?

BG: There is a great deal of evidence that when people are uncertain they look to others like themselves to decide how to act. For example, if only one person arrives at the scene of an accident, that person is very likely to take responsibility by calling 911 and trying to help the victims. If, on the other hand, several people arrive at the same time, no one is likely to take charge or to act. One explanation is that each person looks to the others for clues as to how to act. (One lesson from this is that if you are the victim of an accident and are able to do so, take charge. Point to someone and say, “Call 911.” The are likely to promptly comply.) This behavior suggests that the perception being controlled is “act the way others like you are acting unless there are no others like you around.”

If you just recognize that higher levels of perception include other people and your relationships with them, and that when one person controls his own perceptions, the actions he uses to do that affect other people, there is plenty of basis in HPCT for social relationships.

BG: The problem for me stems from the fact that reorganization seems to be the only mechanism for introducing truly novel goals into the system. I like your new flat-screen TV. Will I buy one for myself? Only if I have a pre-existing goal to buy whatever Bill Buys or if reorganization leads me to become determined to watch the Superbowl on a 50" Samsung. At least this is the way it seems to me.

BP: Well, what other sources of novel organization are you looking for? Maybe we need to find a place for them in HPCT.

BG: Random reorganization is an elegant solution that avoids S-R thinking. It is does not seem to be a very good description of how higher level goals change, however. Perhaps I misunderstand the implications at the level of social relationships, for example.

BG: Nothing I have said should be seen as constituting an objection to your reasoning. I merely want to suggest that other mechanisms may also be involved.

BP: Fine. Propose some!

BG: I think you see my problem. It seems that I can adopt a goal without having to reorganize. Of course, I can always postulate a higher level controlled variable that sets this new reference level. But this solution seems somewhat ad hoc to me.

By the way: the BP: and BG: stuff is there because I’ve notice that some people transmit plain text without styles, and the quoted materials simply aren’t identified as such That makes for difficult reading. I’m trying to set an example. When I remember to,

BG: Thanks again. Your response was both clear and helpful.

Bruce

Bill_Powers1 · January 6, 2010, 5:20pm

[From Bill Powers (2010.01.06.0807 MST)]

Bruce Gregory (2010.01.06.1350 GMT)–

BG: Since I do not have a
problem doing something that I know how to do, MOL is clearly not
appropriate for me.

BP: (1) Do you know how to sock someone in the nose? (2) Have you ever
felt angry enough with someone to sock him or her in the nose? (3) Why
didn’t you?

That’s what I mean by a conflict. Most of us resolve that one by
suppressing goal (1) for reasons such as “I would never do that!
That’s not the kind of person I want to be! Physical violence is
abhorrent to me!” From remarks like those, the MOL therapist asking
about goal (1), just checking it out, would judge that there is reason to
think the person has a conflict and rather than resolving it so there is
no actual desire to sock anyone in the nose, the person is simply using
“will power” to prevent that goal from being achieved. The
desire is still there. So the therapist would pursue that subject a
little further, asking things like “Why wouldn’t you do that? What’s
wrong with physical violence?” and so on. The idea is not to get the
client to make a case for physical violence, but merely for the client to
look inside and report on any answers or other thoughts that are already
there. This idea is generally made clear through repeated little
exchanges like this. There’s no particular place the therapist is trying
to lead the client; the point is to get awareness focused on a possible
problem and see what reorganizes, if anything.

BG: In my case, the advice
involved doing something I was perfectly capable of, and comfortable,
doing. I tried it, and it worked. Again MOL seems aimed at problems that
I fortunately do not have!

BP: That’s great. Sometimes wise advice actually works. However, in
therapy a client’s normal response to advice is to say “Oh, I
couldn’t do that” or “that wouldn’t work for me because
…”. Resistance and noncompliance are big issues in conventional
therapies. I’ve listened to tapes of sessions in which this repeatedly
happens, tempting the therapist to get into an argument with the client:
“Well, how can you know unless you try it? This has helped a lot of
other people. Couldn’t you just try to do it a little bit at
first?”

I haven’t yet met anyone who has no conflicts. Perhaps my clarification
of “know how to do” enlarges the scope of that concept a
little.

BP earlier: I rather doubt the
truth of the behavior-mod idea. Behavior does change when you reorganize,
but changing the behavior isn’t the point; it’s changing the results that
matters, now how you behave to do it. As we know in PCT, behavior changes
with every disturbance; it’s the controlled variable that matters to the
control system.

BG: I’m sorry if I was unclear. The critical point is changing the goals.
The behavior follows as you deal with disturbances.

I rather thought you knew that, but I sort of doubt that the people doing
the metastudies do. PCT isn’t that well known yet.

BG: My problem is with how goals
are established. Once they are established, your explanation makes
perfect sense.

BP: There are two meanings to “how”: one is “why this goal
rather than that goal?” and the other is “what agency chooses
and sets goals?”
In HPCT the agency is a higher-level control system if there is one,
otherwise reorganization. There is a third possibility,
“volition,” but I think that is intimately related to
reorganization (as is evident in the way MOL works).
The answer I have proposed to the first why changes the question to
“why this perception rather than that one?” Goals are states of
perceptions that we wish to produce again. In order to have a goal, you
have to have a perception to seek, meaning you have probably had it
before and liked it (or “to avoid, meaning … didn’t like
it”). I have proposed that at least in the upper levels of the
hierarchy, higher systems don’t just “send reference signals”
to lower systems; I think memory is involved. The higher system somehow
selects from (local) memory a recording of a previous state of a
perception that a lower system has had, thus in effect telling the lower
system, “Have this perception again.”
This solves a “language” problem: for example, how does a
sequence-control system, which perceives and controls only sequences and
no variables found at any other levels, generate a reference signal that
a lower system will recognize as a goal-category, or a goal-relationship,
and so on? This problem isn’t completely solved, but saying that the
higher system sends the address of a stored perception to the
lower system means that the higher system doesn’t need to know anything
about what is stored at that address. It needs to know only that
selecting whatever is in that address results in a perception that the
higher system needs for its own input function – if the lower system
succeeds in matching its present-time perception to the stored one. This
is still too cognitive a view of what the higher system does (it doesn’t
even know that lower-order control systems exist), but it will serve for
now.
A more general answer to the question about goals comes from realizing
that in HPCT, goals are not just “set”: they are varied.
The higher system changes which stored values of a lower perception are
selected, according to its own error signals, which arise because of
disturbances or variations in the reference signals the higher system is
receiving from still-higher systems, or both. While one system generally
controls only one kind of perception, there can be more or less of
that perception without changing its nature: distance and loudness remain
distance and loudness, but higher systems can tell lower systems to
control for more or less of either one.

Some higher systems act by switching from one lower control system to
another: one if by day, two if by night. That’s a matter of selecting one
reference input rather than another at the lower level.

The big unanswered question, then, is where do new kinds of perceptions
come from? “Reorganization” is a sort of garbage can for all
sorts of hard questions (as is “evolution”), but we’ve gone
past that stage. The next big problem on the agenda is how input
functions get reorganized, or organized from scratch if they don’t
already exist. In LCS3 there are several demos showing how output
functions can be reorganized; they work very well. But I still don’t see
what criteria have to be used for input functions, or what kind of
precursor organization has to exist to permit various kinds of input
functions to come into existence. This is really an epistemological
problem: do the levels of perception as I’ve defined them exist because
that is how Real Reality is organized? Or are they just one of an
infinity of possible ways to slice the multidimensional pie? I don’t
know. I answer a lot of questions that way.

BG: There is a great deal of
evidence that when people are uncertain they look to others like
themselves to decide how to act. For example, if only one person arrives
at the scene of an accident, that person is very likely to take
responsibility by calling 911 and trying to help the victims. If, on the
other hand, several people arrive at the same time, no one is likely to
take charge or to act.

BP: One has to be careful with “facts” like these. I don’t
doubt that they are the results of well-done studies, but when you say
“people” are uncertain, the unavoidable message is that ALL
people are uncertain. Most studies of this kind actually show that
some people are sometimes uncertain under some
conditions – and often the number actually showing the characteristic
are a minority of the population (it’s just that no other pattern is more
visible). I am quite sure that there are people who look to others to
make decisions, but I’m just as sure, without denying the studies, that
there are also lots of people who just go ahead and decide.

This is a theme that has run through PCT circles as long as PCT has
existed. Talk about conflicts! Since all of us were educated in a world
in which there was no PCT, especially psychologists and social
scientists, we have been stuffed full of facts that are true of
individuals in only slightly more cases than they are false. These are
low-grade facts; you can’t combine many of them into a complex argument
without making the overall truth value “false.” But saying that
to people who enjoyed graduate school and loved their mentors and friends
is not a way to be popular with them. PCTers with this background often
have trouble with making the transition to real facts. They try, but it’s
hard. Conflict.

We don’t use that kind of facts in PCT or anyhow I try not to. Anything
we accept as a fact, rather than just as a possibility to be explored
further, is always true under all circumstances, and of practically all
people (except physically damaged ones). I don’t think the statement
about who makes decisions is of that nature. Sometimes it’s true,
sometimes (almost as often) it isn’t. And in any given case there’s no
way to predict which it will be.

BG: One explanation is that each
person looks to the others for clues as to how to act. (One lesson from
this is that if you are the victim of an accident and are able to do so,
take charge. Point to someone and say, “Call 911.” The are
likely to promptly comply.) This behavior suggests that the perception
being controlled is “act the way others like you are acting unless
there are no others like you around.”

BP: True, some people really do act this way. A lot don’t. Explanations
of why they act that way when they do also have to explain why they don’t
when they don’t. Otherwise the statements about why they act that way are
obviously incorrect in an uncomfortably large number of cases. To say
“people” are “likely” to comply with an order from
someone else is to say that there are enough people who don’t comply to
require at least mentioning them. And that spoils the generalization, as
far as I’m concerned. Any significant number of exceptions to a
generalization change it from a nice clean fact into a distribution of
mutually-exclusive possibilities.

…

BG: Random reorganization is an
elegant solution that avoids S-R thinking. It is does not seem to be a
very good description of how higher level goals change, however. Perhaps
I misunderstand the implications at the level of social relationships,
for example.

BP: Again, “how higher level goals change” is ambiguous. They
are changed by higher-level systems as a means of counteracting
disturbances of still-higher goals, except for the case where there are
no higher-level systems to change them. Then the answer is “A goal
becomes a different goal when an input function is reorganized so that
the old perception is replaced by a different kind of perception.”
It’s a question of whether the kind or degree of a constant kind of
perception is changing, or one perception is changed into a different
kind of perception. The latter is the subject-matter we call
reorganization.

BG: I think you see my problem.
It seems that I can adopt a goal without having to reorganize. Of course,
I can always postulate a higher level controlled variable that sets this
new reference level. But this solution seems somewhat ad hoc to
me.

BP: Yes, you can adopt a different value of an existing perception as a
goal without reorganizing. I contend that you can’t adopt a different
kind of perception (that doesn’t already exist) as a goal without
reorganizing.

The randomness of E. coli reorganization has bothered others beside you.
But all the randomness does is bring new possibilities into view which
never have been seen before. That’s what makes reorganization so
powerful: it doesn;t rely on past experience, reason, or any of those
things dear to the rational mind. We experience this, at higher levels,
as new ideas, perceptions, or goals, that just pop up out of nowhere,
like inspirations. Some theorists have proposed that they actually are
(divine) inspirations.

However, reorganization has another side, which is the evaluation of the
results of the random changes. If you want something non-random in
reorganization, that’s where to look. We reorganize at random when
something is wrong and neither past experience nor reason has been able
to fix it. But then we experience the consequences of the new
organization, and if it hasn’t made things better we reject it by
reorganizing again. So we are always filtering the results of
reorganization to see if they fit in with the rest of our makeup, and we
can throw out the results that don’t fit. That is definitely not done at
random.

Best,

Bill P.

Shannon_Williams · January 7, 2010, 5:32am

[From Shannon Williams (2010.01.06 23:30 CST)]

[From Bruce Gregory (2010.01.06.1350 GMT)]
BG: There is a great deal of evidence that when people are uncertain they
look to others like themselves to decide how to act. ... This
behavior suggests that the perception being controlled is "act the way
others like you are acting unless there are no others like you around."

No. The perception and goal do not change when more people are on the
scene. The environment changes. The environment changes which
changes which behavior is selected.

If several individuals arrive at the same time, and no one claims the
ability to control the situation (such as an EMT), each individual is
unsure of what behavior is required OF HIS SELF to control the
situation. Your studies probably show different outcomes in
scenarios where one of the first arrivals has been told (through
public awareness campaigns perhaps) that he needs to take charge or
one of the first arrivals is related to (or otherwise driven to meddle
on behalf of) the victim or one of the first arrivals is familiar with
controling in group situations (such as a competent project manager or
a competent person who is in a leadership position).

Just adding my $0.02.

Shannon

Bruce_Gregory6 · January 7, 2010, 1:21pm

[From Bruce Gregory (2010.01.07.1320 GMT)]

[From Bill Powers (2010.01.06.0807 MST)]

Bruce Gregory (2010.01.06.1350 GMT)–

BG: Since I do not have a
problem doing something that I know how to do, MOL is clearly not
appropriate for me.

BP: (1) Do you know how to sock someone in the nose? (2) Have you ever
felt angry enough with someone to sock him or her in the nose? (3) Why
didn’t you?

That’s what I mean by a conflict. Most of us resolve that one by
suppressing goal (1) for reasons such as “I would never do that!
That’s not the kind of person I want to be! Physical violence is
abhorrent to me!” From remarks like those, the MOL therapist asking
about goal (1), just checking it out, would judge that there is reason to
think the person has a conflict and rather than resolving it so there is
no actual desire to sock anyone in the nose, the person is simply using
“will power” to prevent that goal from being achieved. The
desire is still there. So the therapist would pursue that subject a
little further, asking things like “Why wouldn’t you do that? What’s
wrong with physical violence?” and so on. The idea is not to get the
client to make a case for physical violence, but merely for the client to
look inside and report on any answers or other thoughts that are already
there. This idea is generally made clear through repeated little
exchanges like this. There’s no particular place the therapist is trying
to lead the client; the point is to get awareness focused on a possible
problem and see what reorganizes, if anything.

BG: In my case, the advice
involved doing something I was perfectly capable of, and comfortable,
doing. I tried it, and it worked. Again MOL seems aimed at problems that
I fortunately do not have!

BP: That’s great. Sometimes wise advice actually works. However, in
therapy a client’s normal response to advice is to say “Oh, I
couldn’t do that” or “that wouldn’t work for me because
…”. Resistance and noncompliance are big issues in conventional
therapies. I’ve listened to tapes of sessions in which this repeatedly
happens, tempting the therapist to get into an argument with the client:
“Well, how can you know unless you try it? This has helped a lot of
other people. Couldn’t you just try to do it a little bit at
first?”

I haven’t yet met anyone who has no conflicts. Perhaps my clarification
of “know how to do” enlarges the scope of that concept a
little.

BG: I have experienced conflicts and know how paralyzing they can be. However, I was dealing with moods linked to perceptions beyond my ability to control. The solution was to “see” that I could establish goals that I could realize. Recognizing the pattern that anxiety was invariably liked to a control issue seemed to banish the anxiety even without further actions on my part. Anxiety became an old friend that I did not have to read more into than necessary. Dwelling on anxiety made it stronger, acknowledging it and its likely cause let it rapidly diminish.

I can’t resist pointing out that you are talking about “normal” responses, something you call me on.

BP earlier: I rather doubt the
truth of the behavior-mod idea. Behavior does change when you reorganize,
but changing the behavior isn’t the point; it’s changing the results that
matters, now how you behave to do it. As we know in PCT, behavior changes
with every disturbance; it’s the controlled variable that matters to the
control system.

BG: I’m sorry if I was unclear. The critical point is changing the goals.
The behavior follows as you deal with disturbances.

I rather thought you knew that, but I sort of doubt that the people doing
the metastudies do. PCT isn’t that well known yet.

BG: Yes, I know. But if I tell you to change your behavior (adopt a different grip on your tennis racket) I am actually telling you change a perceptual goal. You can be a perfectly good tennis coach without knowing a thing about PCT, although it might help!

BG: My problem is with how goals
are established. Once they are established, your explanation makes
perfect sense.

BP: There are two meanings to “how”: one is “why this goal
rather than that goal?” and the other is “what agency chooses
and sets goals?”

In HPCT the agency is a higher-level control system if there is one,
otherwise reorganization. There is a third possibility,
“volition,” but I think that is intimately related to
reorganization (as is evident in the way MOL works).

BG: I’m not fond of volition as an explanation of anything. I suspect volition is a story we tell about our behavior and not a cause of that behavior. When the brain activates two patterns that have not been active at the same time in the past, it seems to have an “aha!” moment. The action that it initiates following this realization draws on its memories of actions it initiated in the past in similar circumstances.

The answer I have proposed to the first why changes the question to
“why this perception rather than that one?” Goals are states of
perceptions that we wish to produce again. In order to have a goal, you
have to have a perception to seek, meaning you have probably had it
before and liked it (or “to avoid, meaning … didn’t like
it”). I have proposed that at least in the upper levels of the
hierarchy, higher systems don’t just “send reference signals”
to lower systems; I think memory is involved. The higher system somehow
selects from (local) memory a recording of a previous state of a
perception that a lower system has had, thus in effect telling the lower
system, “Have this perception again.”

BG: The way I think about it is in terms of memories of situations encountered (perceptions) and actions initiated to control those perceptions. But your understanding is much the same as mine.

This solves a “language” problem: for example, how does a
sequence-control system, which perceives and controls only sequences and
no variables found at any other levels, generate a reference signal that
a lower system will recognize as a goal-category, or a goal-relationship,
and so on? This problem isn’t completely solved, but saying that the
higher system sends the address of a stored perception to the
lower system means that the higher system doesn’t need to know anything
about what is stored at that address. It needs to know only that
selecting whatever is in that address results in a perception that the
higher system needs for its own input function – if the lower system
succeeds in matching its present-time perception to the stored one. This
is still too cognitive a view of what the higher system does (it doesn’t
even know that lower-order control systems exist), but it will serve for
now.

BG: I think the notion of an address for a stored perception is based on a metaphor (the brain as computer) that is likely to be misleading. Let me elaborate a bit. A perception can be thought of as a network of neural activity. Each time the network is activated, it becomes easier to activate it in the future. This is essentially what learning is. If the pathways are sufficiently long, as is the case with motor neurons, myelination facilitates the process. The brain (and nervous system) is a giant record of past perceptions and actions, not a computer. Present perceptions activate pathways that have been activated in the past. The brain is an auto-associative memory; activating part of a pattern provides access to the entire pattern

A more general answer to the question about goals comes from realizing
that in HPCT, goals are not just “set”: they are varied.
The higher system changes which stored values of a lower perception are
selected, according to its own error signals, which arise because of
disturbances or variations in the reference signals the higher system is
receiving from still-higher systems, or both. While one system generally
controls only one kind of perception, there can be more or less of
that perception without changing its nature: distance and loudness remain
distance and loudness, but higher systems can tell lower systems to
control for more or less of either one.

Some higher systems act by switching from one lower control system to
another: one if by day, two if by night. That’s a matter of selecting one
reference input rather than another at the lower level.

The big unanswered question, then, is where do new kinds of perceptions
come from? “Reorganization” is a sort of garbage can for all
sorts of hard questions (as is “evolution”), but we’ve gone
past that stage. The next big problem on the agenda is how input
functions get reorganized, or organized from scratch if they don’t
already exist. In LCS3 there are several demos showing how output
functions can be reorganized; they work very well. But I still don’t see
what criteria have to be used for input functions, or what kind of
precursor organization has to exist to permit various kinds of input
functions to come into existence. This is really an epistemological
problem: do the levels of perception as I’ve defined them exist because
that is how Real Reality is organized? Or are they just one of an
infinity of possible ways to slice the multidimensional pie? I don’t
know. I answer a lot of questions that way.

BG: All we know is that our perceptions “work” (or not). We have many stories about reality, but unless they lead to predictions, I am not sure one is more preferable than another.

BG: There is a great deal of
evidence that when people are uncertain they look to others like
themselves to decide how to act. For example, if only one person arrives
at the scene of an accident, that person is very likely to take
responsibility by calling 911 and trying to help the victims. If, on the
other hand, several people arrive at the same time, no one is likely to
take charge or to act.

BP: One has to be careful with “facts” like these. I don’t
doubt that they are the results of well-done studies, but when you say
“people” are uncertain, the unavoidable message is that ALL
people are uncertain. Most studies of this kind actually show that
some people are sometimes uncertain under some
conditions – and often the number actually showing the characteristic
are a minority of the population (it’s just that no other pattern is more
visible). I am quite sure that there are people who look to others to
make decisions, but I’m just as sure, without denying the studies, that
there are also lots of people who just go ahead and decide.

BG: I agree. I take the results of such studies as evidence that certain things can happen, not how frequently they happen in practice. What I have useful about the “others like me” phenomenon is that it explains a great deal of human behavior without the need to call on “character.” If others like you are originating “liar loans” and making money hand over fist, you are likely to do so as well. Which does not mean that you will. You may have goals that rule out establishing a new goal of making as much money as you can, any way you can.

This is a theme that has run through PCT circles as long as PCT has
existed. Talk about conflicts! Since all of us were educated in a world
in which there was no PCT, especially psychologists and social
scientists, we have been stuffed full of facts that are true of
individuals in only slightly more cases than they are false. These are
low-grade facts; you can’t combine many of them into a complex argument
without making the overall truth value “false.” But saying that
to people who enjoyed graduate school and loved their mentors and friends
is not a way to be popular with them. PCTers with this background often
have trouble with making the transition to real facts. They try, but it’s
hard. Conflict.

We don’t use that kind of facts in PCT or anyhow I try not to. Anything
we accept as a fact, rather than just as a possibility to be explored
further, is always true under all circumstances, and of practically all
people (except physically damaged ones). I don’t think the statement
about who makes decisions is of that nature. Sometimes it’s true,
sometimes (almost as often) it isn’t. And in any given case there’s no
way to predict which it will be.

BG: One explanation is that each
person looks to the others for clues as to how to act. (One lesson from
this is that if you are the victim of an accident and are able to do so,
take charge. Point to someone and say, “Call 911.” The are
likely to promptly comply.) This behavior suggests that the perception
being controlled is “act the way others like you are acting unless
there are no others like you around.”

BP: True, some people really do act this way. A lot don’t. Explanations
of why they act that way when they do also have to explain why they don’t
when they don’t. Otherwise the statements about why they act that way are
obviously incorrect in an uncomfortably large number of cases. To say
“people” are “likely” to comply with an order from
someone else is to say that there are enough people who don’t comply to
require at least mentioning them. And that spoils the generalization, as
far as I’m concerned. Any significant number of exceptions to a
generalization change it from a nice clean fact into a distribution of
mutually-exclusive possibilities.

…

BG: Random reorganization is an
elegant solution that avoids S-R thinking. It is does not seem to be a
very good description of how higher level goals change, however. Perhaps
I misunderstand the implications at the level of social relationships,
for example.

BP: Again, “how higher level goals change” is ambiguous. They
are changed by higher-level systems as a means of counteracting
disturbances of still-higher goals, except for the case where there are
no higher-level systems to change them. Then the answer is “A goal
becomes a different goal when an input function is reorganized so that
the old perception is replaced by a different kind of perception.”
It’s a question of whether the kind or degree of a constant kind of
perception is changing, or one perception is changed into a different
kind of perception. The latter is the subject-matter we call
reorganization.

BG: In much the same way that you try to stick to facts, I try to stick to networks of neural activity. A perception is a pattern of neural activity. There are different patterns of neural activity, but as far as we know, there are not different kinds of networks. Feeding signals from a TV camera to electrical outputs to the tongue leads to very coarse “vision.” This suggests that ultimately it is the pattern of activity that matters, not the region of the cortex where the pattern takes place. The brain is a pattern matching engine. I am not sure how to describe one kind of perception changing into another. This may be a case where the PCT model involves a property that does not show up at the level of neural networks. I realized this morning that you and I have somewhat different goals. Your goal is to model human behavior as a hierarchical system of control systems incorporating a reorganization mechanism. My goal is to understand human behavior by telling stories that draw on what we know about the behavior of networks of neurons. Your models are quantitative, mine are not. I have no reason to believe that our goals are inconsistent, but that remains to be seen.

BG: I think you see my problem.
It seems that I can adopt a goal without having to reorganize. Of course,
I can always postulate a higher level controlled variable that sets this
new reference level. But this solution seems somewhat ad hoc to
me.

BP: Yes, you can adopt a different value of an existing perception as a
goal without reorganizing. I contend that you can’t adopt a different
kind of perception (that doesn’t already exist) as a goal without
reorganizing.

BG: I understand. That makes perfect sense in the context of HPCT.

The randomness of E. coli reorganization has bothered others beside you.
But all the randomness does is bring new possibilities into view which
never have been seen before. That’s what makes reorganization so
powerful: it doesn;t rely on past experience, reason, or any of those
things dear to the rational mind. We experience this, at higher levels,
as new ideas, perceptions, or goals, that just pop up out of nowhere,
like inspirations. Some theorists have proposed that they actually are
(divine) inspirations.

However, reorganization has another side, which is the evaluation of the
results of the random changes. If you want something non-random in
reorganization, that’s where to look. We reorganize at random when
something is wrong and neither past experience nor reason has been able
to fix it. But then we experience the consequences of the new
organization, and if it hasn’t made things better we reject it by
reorganizing again. So we are always filtering the results of
reorganization to see if they fit in with the rest of our makeup, and we
can throw out the results that don’t fit. That is definitely not done at
random.

BG: By “makeup” I take it that you are referring to our other goals. In this case, the system is testing to see the extent to which the new goal leads to actions that conflict with the actions taken to achieve existing goals. This can happen in “imagination” which in turn is based on memory.

Sorry this post has gotten so lengthy, but I hope it makes some sense to someone not engaged in the exchange.

Bruce

Bruce_Gregory6 · January 7, 2010, 1:50pm

[From Bruce Gregory (2010.01.07.1350)]

[From Shannon Williams (2010.01.06 23:30 CST)]

[From Bruce Gregory (2010.01.06.1350 GMT)]
BG: There is a great deal of evidence that when people are uncertain they
look to others like themselves to decide how to act. … This
behavior suggests that the perception being controlled is “act the way
others like you are acting unless there are no others like you around.”

No. The perception and goal do not change when more people are on the
scene. The environment changes. The environment changes which
changes which behavior is selected.

If several individuals arrive at the same time, and no one claims the
ability to control the situation (such as an EMT), each individual is
unsure of what behavior is required OF HIS SELF to control the
situation. Your studies probably show different outcomes in
scenarios where one of the first arrivals has been told (through
public awareness campaigns perhaps) that he needs to take charge or
one of the first arrivals is related to (or otherwise driven to meddle
on behalf of) the victim or one of the first arrivals is familiar with
controling in group situations (such as a competent project manager or
a competent person who is in a leadership position).

BG: I don’t differ with your description of what is likely to happen, but you seem to be saying that the environment selects behavior. That does not seem to be consistent with HPCT.

Bruce

Bjorn · January 7, 2010, 11:12pm

[From Bjorn Simonsen (2010.01.07.0010 EU ST)]

From Bruce Gregory 92010.01.05.1340 GMT)

My own experience with stress focuses less on conflict than it

does on a lack of control. I suspect I am not unique. The world

has been very unpredictable of late, and things that you once

could rely on, you no longer can. This is a good recipe for loss

of control. This loss of control seems to be accompanied by

a feeling I call anxiety or depression. Can this be modeled as
internal conflict? I’m sure it can, but it is not obvious to me

that it needs to be.

I wish to discuss some of your statements above viewed in the light of your book “Inventing Reality”.

… and things you could rely on, you no longer can
last section, chapter one. “Whether or not God is a mathematician, we know that some human beings are mathematicians, and we will see that the language they speak will be interpreted more and more as being the language in which nature is written.”

You know a lot about mathemathics. I think you and we know more mathematics today than in the years before the time of late. Is it correct to say that “things that you once could rely on, you no longer can.”?

You also say in yourr second section in your preface that " This book tells the story of how physicists invented a language in order to talk about the world." And I like this sentence.

In your last sentence, chapter two you say " But the future was not written in the language of the starsthe future was written in the language of Newton." I understand that the laws scientists express are written in their language. So are also all the expressions psychologists, sosiologist and other …logists expresed and expresses. These languages are based on the scientists and the …logists imaginations. Scientiststs have verified that their laws (theories ) are confirmed by their perceptions of experiments using dead objects. Many …logists have confirmed (often using funny views on statistics) their imaginations (theories) using experiments where the objects are living organisms.

Dead objects do not control their perceptions, they do not perceive anything. Living organisms control their perceptions and the …logists do not know what they perceive.

(Your last sentence in your first section of your prologue) “The visual world is not a faithful reflection of the images on the retinas of our eyes but a world somehow constructed out of such images.” Ib think this applies both scientists and …logists.

bjorn

Bill_Powers1 · January 7, 2010, 11:59pm

[From Bill Powers (2010.01.07.1606 MST)]

Bruce Gregory (2010.01.07.1320 GMT)--

BG: I have experienced conflicts and know how paralyzing they can be. However, I was dealing with moods linked to perceptions beyond my ability to control. The solution was to "see" that I could establish goals that I could realize....

Sounds like a very successful reorganization.

BG: Yes, I know. But if I tell you to change your behavior (adopt a different grip on your tennis racket) I am actually telling you change a perceptual goal. You can be a perfectly good tennis coach without knowing a thing about PCT, although it might help!

Interesting example, in that the coach is telling the tennis player trying to improve his game to go DOWN a level and attend to the grip, not the whole game. MOL therapists take note.

BG: I'm not fond of volition as an explanation of anything. I suspect volition is a story we tell about our behavior and not a cause of that behavior.

I don't know what you mean by volition, then. I think I can do things just to see what will happen, and that is pretty much my canonical example of volition.

When the brain activates two patterns that have not been active at the same time in the past, it seems to have an "aha!" moment. The action that it initiates following this realization draws on its memories of actions it initiated in the past in similar circumstances.

Isn't that Hebb's theory of learning? Or maybe something to do with Guthrie's (contiguity)?

BG: I think the notion of an address for a stored perception is based on a metaphor (the brain as computer) that is likely to be misleading.

BP: I agree, if the "digital" aspect comes into it. I was just trying to think of a way that one level of control, which works in terms of one kind of perception, can select goals for a lower level that works in different terms. As far as the higher-level system is concerned, it would just be sending outputs into its environment -- it doesn't know they're reference signals for a lower system, or addresses. It also doesn't know that the inputs it gets back from this "environment" are actually copies of perceptual signals from lower levels. So the problem was, how can it select relevant lower-order perceptions to serve its own purposes when it doesn't even know they are there? Just a technicality of the model, of course, but it's a loose end I wanted to tie up or snip off. I don't think I got very far with this.

BG: Let me elaborate a bit. A perception can be thought of as a network of neural activity. Each time the network is activated, it becomes easier to activate it in the future. This is essentially what learning is.

BP: This is very close to the Hebbian idea. The problem I see is that the form of an input function (a neural network) requires more than just increasing its activation -- both excitation and inhibition have to be adjusted quantitatively, to generate a particular form of function. Also, adjusting synaptic weightings is fine for creating sensation-perceptions out of intensity signals, but not of obvious use in creating configuration perceptions out of sensations or intensities, or introducing transitions as in the perception of a "spinning" cube, which spins but also remains a cube.

BG: If the pathways are sufficiently long, as is the case with motor neurons, myelination facilitates the process. The brain (and nervous system) is a giant record of past perceptions and actions, not a computer. Present perceptions activate pathways that have been activated in the past. The brain is an auto-associative memory; activating part of a pattern provides access to the entire pattern.

BP: This sounds as if you're more in the S-R camp than the PCT camp, though I don't really believe that. I, too proposed (in B:CP) that memory is associative, but I haven't really done much of anything in that area. As to the whole brain being nothing but a record of the past, I can't go along with that -- there's too much else it does that doesn't need to involve memory.

BG: All we know is that our perceptions "work" (or not). We have many stories about reality, but unless they lead to predictions, I am not sure one is more preferable than another.

BP: I can't agree with you -- more. Models are just ways of making clear predictions from the properties of a theory. Stories are not predictions or explanations. They're used mostly to create an illusion of explaining things that have already happened ("predicting the past"), not predicting what will happen next. Models say, IF reality is organized this way, then THIS is what has to happen.

BG: I realized this morning that you and I have somewhat different goals. Your goal is to model human behavior as a hierarchical system of control systems incorporating a reorganization mechanism. My goal is to understand human behavior by telling stories that draw on what we know about the behavior of networks of neurons. Your models are quantitative, mine are not. I have no reason to believe that our goals are inconsistent, but that remains to be seen.

Actually, my goal is to find a model that accuractely represents and predicts human behavior. It just happens that the control-system model is the only one I've seen so far that does this adequately even in simple situations.

I think our interests are quite consistent, because the neural-network approach, especially the multilayer perceptron approach, is highly compatible with the concept of functions becoming reorganized to satisfy certain criteria. The exploration of neural networks is the exploration of what I call neural functions -- input, comparator, and output functions, perhaps. It may be that somewhere in neural network theory is a solution to the problem of how input functions get organized. Unfortunately all the research I have seen relies on an external teacher, is mostly concerned with categorizing (only one level of the PCT model), and is firmly S-R in its way of connecting perception to behavior (if this is an "A", respond by typing "A"). My model of reorganization shows how to bypass the external teacher. So far it hasn't helped with understanding how we come to perceive higher-level things. Maybe in neural network theories there lurk some solutions to that problem.

BG: By "makeup" I take it that you are referring to our other goals. In this case, the system is testing to see the extent to which the new goal leads to actions that conflict with the actions taken to achieve existing goals. This can happen in "imagination" which in turn is based on memory.

BP: Well said. Yes, that's what I meant. Sometimes this can be done in memory/imagination; other times the conflict isn't discovered until you actually try to carry out the action. Think of a "fork" in chess -- players sometimes fail to see them in time and let go of the piece before seeing the other threat.

Best,

Bill P.

Bruce_Gregory6 · January 8, 2010, 4:06pm

[From Bruce Gregory (2010.01.08.1605 GMT)]

[From Bill Powers (2010.01.07.1606 MST)]

BG: I'm not fond of volition as an explanation of anything. I suspect volition is a story we tell about our behavior and not a cause of that behavior.

BP: I don't know what you mean by volition, then. I think I can do things just to see what will happen, and that is pretty much my canonical example of volition.

BG: Let me tell you a little story about volition. The following is taken from something I have been working on:

For most of us, the left hemisphere of the cerebral cortex initiates spoken language. Normally the left hemisphere and the right hemisphere communicate via the corpus callosom, but after an operation to eliminate the debilitating effects of severe epileptic seizures the two hemispheres are no longer in contact. Most surprisingly, this radical surgery seems to have a minimal impact on those who undergo it. Nevertheless, clever experiments have been designed that allow the experimenter to communicate with each hemisphere individually. The results are sometimes dramatic. By taking advantage of the way the optic nerves deliver information from the visual field to each hemisphere, experimenters can send different information to each hemisphere. The researcher Michael Gazzaniga describes one such experiment:

ï¿½The basic observations have been made hundreds of times in three patients. In the two patients who can respond to printed commands present exclusively to the right hemisphere it is simple to create the same kind of experimental conditions. A simple command such as ï¿½walkï¿½ is given to mute right half-brain and the patient will typically respond by pushing his or chair back and starting to leave the testing area. When asked where he or she is going, a typical response is ï¿½Going into my house to get a Coke.ï¿½ As before, the left hemisphere is faced with the task of explaining an overt behavior that was initiated not by that hemisphere but by the disconnected right half-brain. (Gazzaniga, 1985).ï¿½ Gazzaniga goes on to say, ï¿½It is hard to describe the spell-binding power of seeing such things. Manipulating mind variables is awesome.ï¿½

What exactly is being manipulated here? In split-brain patients, speech generation in the left hemisphere has no direct access to information sent to the right hemisphere. The speech-generating hemisphere has no access to what the right hemisphere saw (the word ï¿½walkï¿½). But more importantly, the left hemisphere apparently has no way to know that it lacks this information. The left hemisphere does not know why the body got up and started walking. But the left hemisphere does not say, "I don't know why I did that." This suggests that the left hemisphere needs information to arrive at the conclusion that it does not know something. Absent this "something is missing" information, the left hemisphere works with the information at hand to construct a plausible reason for the action.

I suspect that the subject of Gazzaniga's experiment was quite certain he had acted of his own volition.

Classic studies by Benjamin Libet showed that the brain intiates and action some 300 msec before the person is aware of having an urge to initiate the action. As the neuroscientist Cris Frith says, "So Libet's results still stands. at the moment at which we are choosing an action, our brain has already made the choice. But this does not mean that the action was not chosen freely. It simply means that we were not _aware_ of making the choice at an earlier time." (By the way, Frith's book, _Making Up the Mind: How the Brain Creates Our Mental World_ is a real gem; highly recommended.)

When the brain activates two patterns that have not been active at the same time in the past, it seems to have an "aha!" moment. The action that it initiates following this realization draws on its memories of actions it initiated in the past in similar circumstances.

Isn't that Hebb's theory of learning? Or maybe something to do with Guthrie's (contiguity)?

BG: Yes. Hebbian learning seems to be the best model anyone has yet come up with.

BG: I think the notion of an address for a stored perception is based on a metaphor (the brain as computer) that is likely to be misleading.

BP: I agree, if the "digital" aspect comes into it. I was just trying to think of a way that one level of control, which works in terms of one kind of perception, can select goals for a lower level that works in different terms. As far as the higher-level system is concerned, it would just be sending outputs into its environment -- it doesn't know they're reference signals for a lower system, or addresses. It also doesn't know that the inputs it gets back from this "environment" are actually copies of perceptual signals from lower levels. So the problem was, how can it select relevant lower-order perceptions to serve its own purposes when it doesn't even know they are there? Just a technicality of the model, of course, but it's a loose end I wanted to tie up or snip off. I don't think I got very far with this.

BG: Yes, I appreciate the problem! I find it a real discipline to keep in mind that the brain, unlike the mind, doesn't know anything. I'm getting better at, but I still fall into "mentalism" from time to time. Whatever understanding is, it must eventually reduce to patterns of neural activity.

BG: Let me elaborate a bit. A perception can be thought of as a network of neural activity. Each time the network is activated, it becomes easier to activate it in the future. This is essentially what learning is.

BP: This is very close to the Hebbian idea. The problem I see is that the form of an input function (a neural network) requires more than just increasing its activation -- both excitation and inhibition have to be adjusted quantitatively, to generate a particular form of function. Also, adjusting synaptic weightings is fine for creating sensation-perceptions out of intensity signals, but not of obvious use in creating configuration perceptions out of sensations or intensities, or introducing transitions as in the perception of a "spinning" cube, which spins but also remains a cube.

BG: If you look at the structure of the neocortex, it seems that higher level perceptions are constructed by the same sort of neural networks that lower level perceptions are. I agree it is not easy to understand how this happens, but the evidence suggests that it does happen.

BG: If the pathways are sufficiently long, as is the case with motor neurons, myelination facilitates the process. The brain (and nervous system) is a giant record of past perceptions and actions, not a computer. Present perceptions activate pathways that have been activated in the past. The brain is an auto-associative memory; activating part of a pattern provides access to the entire pattern.

BP: This sounds as if you're more in the S-R camp than the PCT camp, though I don't really believe that. I, too proposed (in B:CP) that memory is associative, but I haven't really done much of anything in that area. As to the whole brain being nothing but a record of the past, I can't go along with that -- there's too much else it does that doesn't need to involve memory.

BG: First, it might be helpful to distinguish between initiating an action and controlling the action after it is initiated. The premotor cortex is the source of action but I have no reason to think it is involved in controlling actions. If I understand the HPCT model correctly, actions are generated as the result of errors. If this is correct, the analog to initiating an action is setting a reference level. So you could say that the brain remembers reference levels. I can appreciate your reluctance to accept the notion that the brain (the neocortex) is largely a record of the past; it took me quite a while to come to this conclusion. In any case, it is not critical to the present discussion.

BG: All we know is that our perceptions "work" (or not). We have many stories about reality, but unless they lead to predictions, I am not sure one is more preferable than another.

BP: I can't agree with you -- more. Models are just ways of making clear predictions from the properties of a theory. Stories are not predictions or explanations. They're used mostly to create an illusion of explaining things that have already happened ("predicting the past"), not predicting what will happen next. Models say, IF reality is organized this way, then THIS is what has to happen.

I agree with you about the nature of models. Quantitative models, when they can be constructed, are definitely preferable. I think you'll agree, however, that it is not always possible to build the kind of models we would like to have. Climate is one area where the models are too simplified to be useful. If you want to predict the effects of global warming, you are better off relying on paleoclimate data and recognizing that things are changing much more rapidly now than in the past (short of rare catastrophic events). For anyone interested in the science of climate change, I recommend Jim Hansen's new book, _ Storms of My Grandchildren: The Truth About the Coming Climate Catastrophe and Our Last Chance to Save Humanity_. (The situation is as dire as the title makes it seem.)

BG: I realized this morning that you and I have somewhat different goals. Your goal is to model human behavior as a hierarchical system of control systems incorporating a reorganization mechanism. My goal is to understand human behavior by telling stories that draw on what we know about the behavior of networks of neurons. Your models are quantitative, mine are not. I have no reason to believe that our goals are inconsistent, but that remains to be seen.

Actually, my goal is to find a model that accuractely represents and predicts human behavior. It just happens that the control-system model is the only one I've seen so far that does this adequately even in simple situations.

I think our interests are quite consistent, because the neural-network approach, especially the multilayer perceptron approach, is highly compatible with the concept of functions becoming reorganized to satisfy certain criteria. The exploration of neural networks is the exploration of what I call neural functions -- input, comparator, and output functions, perhaps. It may be that somewhere in neural network theory is a solution to the problem of how input functions get organized. Unfortunately all the research I have seen relies on an external teacher, is mostly concerned with categorizing (only one level of the PCT model), and is firmly S-R in its way of connecting perception to behavior (if this is an "A", respond by typing "A"). My model of reorganization shows how to bypass the external teacher. So far it hasn't helped with understanding how we come to perceive higher-level things. Maybe in neural network theories there lurk some solutions to that problem.

BG: I share your observations. Jeff Hawkins' company Numenta is building some models that seem promising, but are too complex for me to understand. Many of my views of the functioning of the neocortex can be traced to Jeff's excellent book, _On Intelligence_. Anyone interested in human behavior owes it to themselves to take a look at this book.

Bruce

Bill_Powers1 · January 8, 2010, 7:20pm

[From Bill Powers(2010.01.08.1055 MST)]

Bruce Gregory (2010.01.08.1605 GMT) --

BG: Let me tell you a little story about volition. The following is taken from something I have been working on:
...
Classic studies by Benjamin Libet showed that the brain intiates and action some 300 msec before the person is aware of having an urge to initiate the action. As the neuroscientist Cris Frith says, "So Libet's results still stands. at the moment at which we are choosing an action, our brain has already made the choice. But this does not mean that the action was not chosen freely. It simply means that we were not _aware_ of making the choice at an earlier time." (By the way, Frith's book, _Making Up the Mind: How the Brain Creates Our Mental World_ is a real gem; highly recommended.)

These observations, or rather the conclusions drawn from them, are very model-dependent. Since my model differs so radically from the ones that neuroscience still seems to accept, it's hard to make comparisons. I doubt that Libet did the kinds of experiments I would do.

Also, there seems to be an epidemic of ambiguous writing going on. What does it mean to say "the brain" has already made the choice? The brain is not homogeneous like a potato. And to say "we" were not aware of making the choice implies that "we" made the choice and were also unaware of making the choice. The 300-millisecond lag is a lag between what and what? How do you determine the time at which the choice was actually made and the time the action started?

I'm pretty sure I don't understand the meaning of the above, because it seem to me that I can announce, "In about five seconds, I'm going to lower this finger which is pointing at you," and about 5 seconds later, do it. I'm certainly aware that I am choosing to move my finger at some time, and that I have initiated that choice, and that the action will follow within a very short time. I don't think the action would have occurred unless I willed it to occur (barring coincidences like a meteorite hitting my arm and breaking it).

My treatment of awareness (discussed mainly under the heading of MOL) doesn't pin it down to any physiological function, because I can't think of one that would account for the properties of awareness. So I just treat awareness as the operation of some system of unknown nature that can receive information selectively from different parts of the brain, with one side-effect of dragging the locus of reorganization around with it. A ghost in the machine, as far as any present knowledge can tell us. When awareness is located in, or identified with, or focused upon, some part of the brain's operations, it experiences the signals that the perceptual input functions are generating, and in more general terms sees a world through the perceptual signals of some set of control systems. That combination is what we call consciousness: awareness plus neural perceptual signals. See the Glossary in B:CP.

Another implication of this model is that large parts of the control hierarchy normally operate without the presence of awareness. The odd thing is the way awareness can apparently move around through the brain's subsystems, bringing some types of perceptual signals into awareness while others disappear from it. The kinds of perceptions involved in this phenomenon are obviously still there when not in awareness, because the control systems that are associated with them go right on controlling -- breathing, for example, or simply sitting upright in a chair.

You may or may not think this picture fits observations, but it certainly offers a different explanation for the phenomena of volition to which you refer, because the underlying model is different. A delay between the moment a reference signal changes and the moment one becomes aware of the effect on behavior becomes understandable, because very often awareness is not focused on the highest levels of organation. Those higher levels can react to disturbances by changing reference signals for lower systems, producing activities at the lowest levels which are the first thing awareness knows about the action. Reference signals always change before changes of action, so it's not surprising that a lag would be observed there (though 300 milliseconds seems too long except at the top few levels of organization).

It's not surprising, under my model, that awareness/volition might (a) not be the origin of an action (which can be produced by disturbances or control systems not currently in the conscious mode), and (b) might see the consequences of the changed reference signals only after they have resulted in some overt action. I can understand (and have experienced) the confusion that results from doing things I was not aware of initiating -- when I say "I" did that, I mean only that it's pretty sure that my brain and muscles rather than someone else's caused it, but at the same time any "I" that I can point to is clearly not me, because I'm here and I'm observing it over there.

When we get into areas like this we've clearly gone beyond any neuromuscular model and are just fishing around for some kind of order. We're defining phenomena that demand explanations that are not available. But that's how all modeling starts, and I'm sure this one will improve with time.

When the brain activates two patterns that have not been active at the same time in the past, it seems to have an "aha!" moment. The action that it initiates following this realization draws on its memories of actions it initiated in the past in similar circumstances.
>
> Isn't that Hebb's theory of learning? Or maybe something to do with Guthrie's (contiguity)?

BG: Yes. Hebbian learning seems to be the best model anyone has yet come up with.

No, it isn't. Mine is. As far as I have been able to make models of it work.

BG: Yes, I appreciate the problem! I find it a real discipline to keep in mind that the brain, unlike the mind, doesn't know anything. I'm getting better at, but I still fall into "mentalism" from time to time. Whatever understanding is, it must eventually reduce to patterns of neural activity.

But what is "mentalism" but the pattern of neural activity? The pattern has no physical existence; it exists in the eye of some beholder. Neural activity, at the physiological level, is just electrochemistry, but the pattern -- the algorithm for computing the square root of a number by Newton's Method -- would never be discovered by a chemist. We're talking about the difference between a computer program and the computer in which the program is running. The program can be transferred from one machine to another just by transmitting information; it's clearly not made of silicon.
...

BG: If you look at the structure of the neocortex, it seems that higher level perceptions are constructed by the same sort of neural networks that lower level perceptions are.

Well, not exactly the same, are they? Yes, all information is ferried around as neural impulses, but the operations involved can't be the same. You can't do logic using only weighted summations. You can't detect differences in sequence by the same operations used to detect relationships or categories. And I doubt very much that pyramid cells would be of much use in the cerebellum, not would Purkinje cells from the cerebellum be of aid in visual pattern recognition.

I get the idea that you're thinking of the brain as just one big randomly connected sponge, in which any part of the brain could become organized to perform any function. I don't think that's true. I'm aware of the studies of brain plasticity in which some forms of vision can be reproduced by patterns of stimulation on the tongue or the back, but those patterns are not recognized by the tongue or the back; they are recognized by areas in the brain. In HPCT there are 11 levels of perception, and each one extends across all (or at least many) sensory modalities. It's not surprising that visual patterns can be detected even by the auditory cortex (in ferrets, I think), if there are similar levels in the auditory and visual parts of the hierarchy. The computations through which configurations are created don't depend on the origins of the neural signals that are input to the computations.

I agree it is not easy to understand how this happens, but the evidence suggests that it does happen.

Reorganization theory helps us to understand it, I think. Recent findings about brain plasticity support the idea that brain functions can be altered even in adults, and I've seen mention of the randomness of those alterations that occur in the beginning of a learning process. I think reorganization theory works a lot better than the Hebbian theory of simultaneous firings. Naturally, I speak objectively.

...

BG: First, it might be helpful to distinguish between initiating an action and controlling the action after it is initiated. The premotor cortex is the source of action but I have no reason to think it is involved in controlling actions. If I understand the HPCT model correctly, actions are generated as the result of errors. If this is correct, the analog to initiating an action is setting a reference level. So you could say that the brain remembers reference levels. I can appreciate your reluctance to accept the notion that the brain (the neocortex) is largely a record of the past; it took me quite a while to come to this conclusion. In any case, it is not critical to the present discussion.

I just don't see output functions, comparators, or motor neurons as records of the past (except metaphorically). I already assume (in B:CP) that every input function at every level is connected to memory, both to record and to play back what is stored. But memory doesn't perform the computations that create one perception as a function of subsets of other perceptions. Memory doesn't detect the difference between a reference signal and a present-time perceptual signal. Memory doesn't cause an error at one level to end up adjusting reference signals in systems that deal with lower-level variables.

This is all very interesting, but there's no point in 'tis-so-'taint-so arguments about models. We can make proposals, but we can't decide which is the better until they are turned into working models and used to make predictions about experimental results. That's what I spend most of my time doing, or did until CSGnet got so active lately.

I'm going to have to focus more on the modelling because I've finally got in touch with the department of psychology at CU in Boulder (specifically, Lew Harvey) and have been invited to join in the Monday brown-bag lunches. It's a hotbed of neuroscientists there so I'm going to have to keep my mouth shut and let my demos do the talking. Maybe I'll learn more about how to get along with naive realists; I can put on a pretty good imitation of one when the occasion demands. But at least Lew Harvey is halfway to PCT: he told me that he's been using B:CP in his graduate classes for years. And he's a proficient programmer, so he'll appreciate LCS3, too.

Best,

Bill P.

Bruce_Gregory6 · January 8, 2010, 9:21pm

[From Bruce Gregory (2010.01.08.2120 GMT)]

[From Bill Powers(2010.01.08.1055 MST)]

Bruce Gregory (2010.01.08.1605 GMT) --

BG: Let me tell you a little story about volition. The following is taken from something I have been working on:
...
Classic studies by Benjamin Libet showed that the brain intiates and action some 300 msec before the person is aware of having an urge to initiate the action. As the neuroscientist Cris Frith says, "So Libet's results still stands. at the moment at which we are choosing an action, our brain has already made the choice. But this does not mean that the action was not chosen freely. It simply means that we were not _aware_ of making the choice at an earlier time." (By the way, Frith's book, _Making Up the Mind: How the Brain Creates Our Mental World_ is a real gem; highly recommended.)

These observations, or rather the conclusions drawn from them, are very model-dependent. Since my model differs so radically from the ones that neuroscience still seems to accept, it's hard to make comparisons. I doubt that Libet did the kinds of experiments I would do.

Also, there seems to be an epidemic of ambiguous writing going on. What does it mean to say "the brain" has already made the choice? The brain is not homogeneous like a potato. And to say "we" were not aware of making the choice implies that "we" made the choice and were also unaware of making the choice. The 300-millisecond lag is a lag between what and what? How do you determine the time at which the choice was actually made and the time the action started?

BG: The subjects were told to move a finger at the time of their choosing. Subjects recorded the time at which they had "the urge" to lift their finger by watching a special clock. Typically, the time the reported preceded the actual motion by 200 msec. Using EEG records, experimenters found that the brain activity that typically precedes finger motion began 500 msec before the finger started to move. Thus this activity preceded the subjects awareness that he had initiated movement by 300 msec. The data (and references) are in Frith, 2007.

I'm pretty sure I don't understand the meaning of the above, because it seem to me that I can announce, "In about five seconds, I'm going to lower this finger which is pointing at you," and about 5 seconds later, do it. I'm certainly aware that I am choosing to move my finger at some time, and that I have initiated that choice, and that the action will follow within a very short time. I don't think the action would have occurred unless I willed it to occur (barring coincidences like a meteorite hitting my arm and breaking it).

BG: Without doing the experiment the findings are of course conjecture. I suggest that we might find that before you began to speak the actions required to say, "In about five seconds..." Even more conjecture, the decision to say anything at all occurred before you were aware that you had decided anything.

My treatment of awareness (discussed mainly under the heading of MOL) doesn't pin it down to any physiological function, because I can't think of one that would account for the properties of awareness. So I just treat awareness as the operation of some system of unknown nature that can receive information selectively from different parts of the brain, with one side-effect of dragging the locus of reorganization around with it. A ghost in the machine, as far as any present knowledge can tell us. When awareness is located in, or identified with, or focused upon, some part of the brain's operations, it experiences the signals that the perceptual input functions are generating, and in more general terms sees a world through the perceptual signals of some set of control systems. That combination is what we call consciousness: awareness plus neural perceptual signals. See the Glossary in B:CP.

BG: I suspect that awareness is the activation of perceptual circuits in the brain that reach some threshold. Awareness does not receive information so much as it is information. There is no ghost in the machine; there is just the machine.

Another implication of this model is that large parts of the control hierarchy normally operate without the presence of awareness. The odd thing is the way awareness can apparently move around through the brain's subsystems, bringing some types of perceptual signals into awareness while others disappear from it. The kinds of perceptions involved in this phenomenon are obviously still there when not in awareness, because the control systems that are associated with them go right on controlling -- breathing, for example, or simply sitting upright in a chair.

BG: That is certainly true. Most neural activity occurs without triggering awareness. Again, I think awareness does not move around; some circuits are activated more strongly than others. The level of activation is determined by electrochemical activity.

You may or may not think this picture fits observations, but it certainly offers a different explanation for the phenomena of volition to which you refer, because the underlying model is different. A delay between the moment a reference signal changes and the moment one becomes aware of the effect on behavior becomes understandable, because very often awareness is not focused on the highest levels of organation. Those higher levels can react to disturbances by changing reference signals for lower systems, producing activities at the lowest levels which are the first thing awareness knows about the action. Reference signals always change before changes of action, so it's not surprising that a lag would be observed there (though 300 milliseconds seems too long except at the top few levels of organization).

BG: If you are willing to give up the ghost we do not disagree.

It's not surprising, under my model, that awareness/volition might (a) not be the origin of an action (which can be produced by disturbances or control systems not currently in the conscious mode), and (b) might see the consequences of the changed reference signals only after they have resulted in some overt action. I can understand (and have experienced) the confusion that results from doing things I was not aware of initiating -- when I say "I" did that, I mean only that it's pretty sure that my brain and muscles rather than someone else's caused it, but at the same time any "I" that I can point to is clearly not me, because I'm here and I'm observing it over there.

BG: I have no problem with that.

When we get into areas like this we've clearly gone beyond any neuromuscular model and are just fishing around for some kind of order. We're defining phenomena that demand explanations that are not available. But that's how all modeling starts, and I'm sure this one will improve with time.

When the brain activates two patterns that have not been active at the same time in the past, it seems to have an "aha!" moment. The action that it initiates following this realization draws on its memories of actions it initiated in the past in similar circumstances.
>
> Isn't that Hebb's theory of learning? Or maybe something to do with Guthrie's (contiguity)?

BG: Yes. Hebbian learning seems to be the best model anyone has yet come up with.

No, it isn't. Mine is. As far as I have been able to make models of it work.

BG: I don't know anything about your model of learning (as contrasted with control) so I am not in a position to judge.

BG: Yes, I appreciate the problem! I find it a real discipline to keep in mind that the brain, unlike the mind, doesn't know anything. I'm getting better at, but I still fall into "mentalism" from time to time. Whatever understanding is, it must eventually reduce to patterns of neural activity.

But what is "mentalism" but the pattern of neural activity? The pattern has no physical existence; it exists in the eye of some beholder. Neural activity, at the physiological level, is just electrochemistry, but the pattern -- the algorithm for computing the square root of a number by Newton's Method -- would never be discovered by a chemist. We're talking about the difference between a computer program and the computer in which the program is running. The program can be transferred from one machine to another just by transmitting information; it's clearly not made of silicon.

BG: There are no programs running in the brain, except the one's we put there. If you think about it I suspect you may come to the conclusion that natural selection is incapable of developing a computer. Of course if one appears as a result of natural selection, I will have to revise my thinking (or accept Intentional Design). Mentalism is the idea that the brain ponders, decides, understands, make sense of. Mentalism is the ghost in the machine.

...

BG: If you look at the structure of the neocortex, it seems that higher level perceptions are constructed by the same sort of neural networks that lower level perceptions are.

Well, not exactly the same, are they? Yes, all information is ferried around as neural impulses, but the operations involved can't be the same. You can't do logic using only weighted summations. You can't detect differences in sequence by the same operations used to detect relationships or categories. And I doubt very much that pyramid cells would be of much use in the cerebellum, not would Purkinje cells from the cerebellum be of aid in visual pattern recognition.

I get the idea that you're thinking of the brain as just one big randomly connected sponge, in which any part of the brain could become organized to perform any function. I don't think that's true. I'm aware of the studies of brain plasticity in which some forms of vision can be reproduced by patterns of stimulation on the tongue or the back, but those patterns are not recognized by the tongue or the back; they are recognized by areas in the brain.

BG: That is true,but I fail to take your point.

In HPCT there are 11 levels of perception, and each one extends across all (or at least many) sensory modalities. It's not surprising that visual patterns can be detected even by the auditory cortex (in ferrets, I think), if there are similar levels in the auditory and visual parts of the hierarchy. The computations through which configurations are created don't depend on the origins of the neural signals that are input to the computations.

BG: Fine, but I doubt any computations are going on. I assume you are using the word metaphorically.

I agree it is not easy to understand how this happens, but the evidence suggests that it does happen.

Reorganization theory helps us to understand it, I think. Recent findings about brain plasticity support the idea that brain functions can be altered even in adults, and I've seen mention of the randomness of those alterations that occur in the beginning of a learning process. I think reorganization theory works a lot better than the Hebbian theory of simultaneous firings. Naturally, I speak objectively.

BG: You'll forgive me if I recall the cartoon in which a scientist writing on a blackboard reaches his conclusion by postulating "and then a miracle occurs." "And then reorganization occurs..." There is a great deal of neuroscience behind Hebbian learning. As yet there seems to be little neuroscience behind reorganization. I hope that comes next.

...

BG: First, it might be helpful to distinguish between initiating an action and controlling the action after it is initiated. The premotor cortex is the source of action but I have no reason to think it is involved in controlling actions. If I understand the HPCT model correctly, actions are generated as the result of errors. If this is correct, the analog to initiating an action is setting a reference level. So you could say that the brain remembers reference levels. I can appreciate your reluctance to accept the notion that the brain (the neocortex) is largely a record of the past; it took me quite a while to come to this conclusion. In any case, it is not critical to the present discussion.

I just don't see output functions, comparators, or motor neurons as records of the past (except metaphorically). I already assume (in B:CP) that every input function at every level is connected to memory, both to record and to play back what is stored. But memory doesn't perform the computations that create one perception as a function of subsets of other perceptions. Memory doesn't detect the difference between a reference signal and a present-time perceptual signal. Memory doesn't cause an error at one level to end up adjusting reference signals in systems that deal with lower-level variables.

BG: In what I am saying there is no separate memory. It is all memory. Or, if you prefer, it is all neural networks (since the memory resides in the strength of the networks). The error signal is the difference between a present perception (which will soon be a memory) and a resident memory. The brain has a error detection (or error amplification) center in the anterior cingulate cortex, but this probably is sensitive to conflicting actions generated by attempts to achieve inconsistent goals.

This is all very interesting, but there's no point in 'tis-so-'taint-so arguments about models. We can make proposals, but we can't decide which is the better until they are turned into working models and used to make predictions about experimental results. That's what I spend most of my time doing, or did until CSGnet got so active lately.

BG: I'll do my best to help by keeping my opinions to myself.

I'm going to have to focus more on the modelling because I've finally got in touch with the department of psychology at CU in Boulder (specifically, Lew Harvey) and have been invited to join in the Monday brown-bag lunches. It's a hotbed of neuroscientists there so I'm going to have to keep my mouth shut and let my demos do the talking. Maybe I'll learn more about how to get along with naive realists; I can put on a pretty good imitation of one when the occasion demands. But at least Lew Harvey is halfway to PCT: he told me that he's been using B:CP in his graduate classes for years. And he's a proficient programmer, so he'll appreciate LCS3, too.

BG: Good luck. By your description I guess that I'm a naive realist. It seems that you think that the brain really is a hierarchical control system filled with the components of the model. Modelers do have a tendency to take their models more seriously than others are likely to do. I liken it to thinking that the ultimate physical reality must be probability amplitudes because quantum mechanics makes use of probability amplitudes. No harm done unless you begin to stand on street corners shouting, "There is no God, there are only probability amplitudes!"

No hard feelings,

Bruce

Bruce_Gregory6 · January 8, 2010, 9:27pm

[From Bruce Gregory (2010.01.08.2127 GMT)]

[From Bjorn Simonsen (2010.01.07.0010 EU ST)]

You know a lot about mathemathics. I think you and we know more mathematics today than in the years before the time of late. Is it correct to say that “things that you once could rely on, you no longer can.”?

BG: I was too cryptic. I was referring to the kind of financial planning and job security that seem to have become a thing of the past for many citizens.

You also say in yourr second section in your preface that " This book tells the story of how physicists invented a language in order to talk about the world." And I like this sentence.

In your last sentence, chapter two you say " But the future was not written in the language of the starsthe future was written in the language of Newton." I understand that the laws scientists express are written in their language. So are also all the expressions psychologists, sosiologist and other …logists expresed and expresses. These languages are based on the scientists and the …logists imaginations. Scientiststs have verified that their laws (theories ) are confirmed by their perceptions of experiments using dead objects. Many …logists have confirmed (often using funny views on statistics) their imaginations (theories) using experiments where the objects are living organisms.

Dead objects do not control their perceptions, they do not perceive anything. Living organisms control their perceptions and the …logists do not know what they perceive.

BG: Of course, all we really have access to is our perceptions. But you are quite right, we are all given to seeing what we expect to see.

Bruce

Bill_Powers1 · January 9, 2010, 12:47am

[From Bill Powers (2010.01.08.1555 MST)]

Bruce Gregory (2010.01.08.2120 GMT) –

BG: The subjects were told to
move a finger at the time of their choosing. Subjects recorded the time
at which they had “the urge” to lift their finger by watching a
special clock. Typically, the time the reported preceded the actual
motion by 200 msec. Using EEG records, experimenters found that the brain
activity that typically precedes finger motion began 500 msec before the
finger started to move. Thus this activity preceded the subjects
awareness that he had initiated movement by 300 msec. The data (and
references) are in Frith, 2007.

BP: How long did it take from the time they felt the urge to move the
finger to the time when they reported it and the clock reading was
recorded? I don’t have easy access to “Frith 2007” so maybe you
could look that up for me.

BP earlier: I’m pretty sure I
don’t understand the meaning of the above, because it seem to me that I
can announce, “In about five seconds, I’m going to lower this finger
which is pointing at you,” and about 5 seconds later, do it. I’m
certainly aware that I am choosing to move my finger at some time, and
that I have initiated that choice, and that the action will follow within
a very short time. I don’t think the action would have occurred unless I
willed it to occur (barring coincidences like a meteorite hitting my arm
and breaking it).

BG: Without doing the experiment the findings are of course conjecture. I
suggest that we might find that before you began to speak the actions
required to say, “In about five seconds…” Even more
conjecture, the decision to say anything at all occurred before you were
aware that you had decided anything.

BP: Yes, possible, but hard to check. That’s more like faith in an
underlying model than knowledge. Even in PCT, it’s likely that there is a
higher-order system setting the reference condition for the sentences to
be spoken, as a way of achieving a still higher goal such as
demonstrating some theoretical proposal. When you can find such a higher
control system, volition doesn’t come into it. Volition in the sense I’m
thinking of is evident only when there isn’t some higher learned system
acting, or when the voluntary act disrupts what a higher system is
doing.

BP earlier: My treatment of
awareness (discussed mainly under the heading of MOL) doesn’t pin it down
to any physiological function, because I can’t think of one that would
account for the properties of awareness.

BG: I suspect that awareness is
the activation of perceptual circuits in the brain that reach some
threshold. Awareness does not receive information so much as it is
information. There is no ghost in the machine; there is just the
machine.

BP: Well, that’s the issue, isn’t it? Is there or isn’t there some actual
receiver of information? I have trouble imagining how blips moving along
a nerve fiber could experience anything. What I’m trying to account for
is this sense of observing something else, of being an observer as
opposed to observing any particular thing. Maybe I’m the only organism
that has this sort of experience, so I’m just weird. I think, however,
that other people have this same sort of experience, but it’s very hard
to check what lies behind their words. Maybe I’m just projecting my
delusions onto other people who are talking about something quite
different. Do you think that could be the case?

Again, I think awareness does
not move around; some circuits are activated more strongly than others.
The level of activation is determined by electrochemical
activity.

My problem with that is that when I’m being aware of things, I don’t feel
like a different person just because different things come to attention.
If awareness is a receiver of some sort, and it doesn’t move around, then
the only remaining explanation is that there are information channels
coming into it from everywhere in the brain, and switches which determine
which of its input channels is active. I haven’t come across any brain
structure, except perhaps the reticular formation, that would seem to
provide such a large number of information channels – and expecially
channels from low-order systems.

BP earlier: Reference signals
always change before changes of action, so it’s not surprising that a lag
would be observed there (though 300 milliseconds seems too long except at
the top few levels of organization).

BG: If you are willing to give up the ghost we do not
disagree.

BP: I’m not ready to give it up yet. It may not be a good explanation for
all these phenomena of awareness, but without it I don’t have any
explanation at all.

BP earlier: … when I say
“I” did that, I mean only that it’s pretty sure that my brain
and muscles rather than someone else’s caused it, but at the same time
any “I” that I can point to is clearly not me, because I’m here
and I’m observing it over there.

BG: I have no problem with that.

BP: You have me stumped now. What is that observer that is here,
observing the self over there, if not awareness? In fact this observer is
pretty much what I mean by “awareness.” If you have no
problem with the existence of this observer, then we have no argument at
all.
*…*BP: When we get into areas like this we’ve clearly gone beyond any
neuromuscular model and are just fishing around for some kind of order.
We’re defining phenomena that demand explanations that are not available.
But that’s how all modeling starts, and I’m sure this one will improve
with time.

Isn’t that Hebb’s
theory of learning? Or maybe something to do with Guthrie’s
(contiguity)?

BG: Yes. Hebbian learning seems to be the best model anyone has
yet come up with.

No, it isn’t. Mine is. As far as I have been able to make models of
it work.

BG: I don’t know anything about your model of learning (as contrasted
with control) so I am not in a position to judge.

BP: It’s called reorganization. Are you sure you haven’t come across it?
There are several pretty good demonstrations of it in LCS III.

BG: There are no programs
running in the brain, except the one’s we put there.

BP: Yes, those are the ones I’m talking about. They represent the
activities in the brain at the level where logic is the prime kind of
computation. What did you think I meant?

BG: If you think about it I
suspect you may come to the conclusion that natural selection is
incapable of developing a computer.

BP: This is probably just a matter of what we mean by a computer. I don’t
mean writing numbers and symbols down and then manipulating them
according to the rules of arithmetic – or, since we actually do such
things, I don’t mean doing them inside the brain either actually or in
imagination.

My first acquaintance with electronic computers had to do with analog
computers, which do not work with numbers, logic, or rules of
mathematics. That’s my meaning whenever I say the nervous system computes
something but don’t mean the pencil-and-paper thing. I don’t know how
familiar you are with analog computing – I don’t want to insult you by
explaining things you’ve known for years.

Of course if one appears
as a result of natural selection, I will have to revise my thinking (or
accept Intentional Design). Mentalism is the idea that the brain ponders,
decides, understands, make sense of. Mentalism is the ghost in the
machine.

So what’s wrong with that? Don’t we ponder, decide, understand, and make
sense of things? I don’t know about you, but I do. I wish I had a more
advanced model to bolster my position. When you say you
“suspect” I may decide something, isn’t suspecting a mental
activity?

I get the idea that you’re
thinking of the brain as just one big randomly connected sponge, in which
any part of the brain could become organized to perform any function. I
don’t think that’s true. I’m aware of the studies of brain plasticity in
which some forms of vision can be reproduced by patterns of stimulation
on the tongue or the back, but those patterns are not recognized by the
tongue or the back; they are recognized by areas in the brain.

BG: That is true,but I fail to take your point.

BP: My point is that different parts of the brain handle different kinds
of perceptual functions, so to get similar perceptual results from
different kinds of inputs, the neural signals have to go to the parts of
the brain that handle the appropriate types of functions.

BP earlier: The computations
through which configurations are created don’t depend on the origins of
the neural signals that are input to the computations.

BG: Fine, but I doubt any computations are going on. I assume you are
using the word metaphorically.

Not at all. I mean literal analog computations, just like in an analog
computer except that instead of using capacitors, resistors, inductors,
and diodes as computing functions, the brain uses neurons and
synapses.

BG: I agree it is not easy to
understand how this happens, but the evidence suggests that it does
happen.

Reorganization theory helps us to understand it, I think. Recent
findings about brain plasticity support the idea that brain functions can
be altered even in adults, and I’ve seen mention of the randomness of
those alterations that occur in the beginning of a learning process. I
think reorganization theory works a lot better than the Hebbian theory of
simultaneous firings. Naturally, I speak objectively.

BG: You’ll forgive me if I recall the cartoon in which a scientist
writing on a blackboard reaches his conclusion by postulating “and
then a miracle occurs.” “And then reorganization
occurs…” There is a great deal of neuroscience behind Hebbian
learning. As yet there seems to be little neuroscience behind
reorganization. I hope that comes next.

BP: I hope so too. I’ve actually run across several references (in
Science) to random processes involved in “adaptation,” which is
a form of reorganization, but unfortunately have lost them because I
wasn’t in the right place and didn’t have time to take notes. I don’t
think there is a lot of neuroscience behind Hebbian learning; just a lot
of overinterpreted details of synaptic interactions. Neuroscientists are
not known for being too timid to leap to conclusions across vast chasms
of missing information.

BG: In what I am saying there is
no separate memory. It is all memory. Or, if you prefer, it is all neural
networks (since the memory resides in the strength of the networks).

BP: The “strength” of networks? Do you mean just how many
impulses per second come out per impulse that went in?

BG: The error signal is the
difference between a present perception (which will soon be a memory) and
a resident memory. The brain has a error detection (or error
amplification) center in the anterior cingulate cortex, but this probably
is sensitive to conflicting actions generated by attempts to achieve
inconsistent goals.

BP: The brain has error detection and amplification centers practically
everywhere. The spinal cord and muscles are full of them. They exist in
every reflex. The cerebellum is overflowing with them. Every control
system has to have one. What’s so special about the anterior cingulate
cortex? I’m sure it has them, too, but where else did these scientists
look for them?

BP earlier: This is all very
interesting, but there’s no point in 'tis-so-'taint-so arguments about
models. We can make proposals, but we can’t decide which is the better
until they are turned into working models and used to make predictions
about experimental results. That’s what I spend most of my time doing, or
did until CSGnet got so active lately.

BG: I’ll do my best to help by
keeping my opinions to myself.

BP: Now I’m beginning to be reminded of what happened the last time we
tried to have a discussion. What’s with the paranoia? I didn’t say I
wanted you to keep your opinions to yourself. I was saying that a better
context for discussion of these disagreements is one in which we look at
models and test them, not one where we just make flat statements back and
forth to each other. I was more or less apologizing for not having stuck
to that way of exploring the subject, because I have spent so much time
on CSGnet talking (not just with you) but not doing.

BG: Good luck. By your
description I guess that I’m a naive realist. It seems that you think
that the brain really is a hierarchical control system filled with the
components of the model.

BP: Yes, I rather think there is a lot of evidence that this is true. We
can take it down to the neural circuit level in a few places, but of
course a great deal more has to be done before we can carry that much
higher than the first two or three levels in a spotty way. Our best
evidence right now is the way models based on PCT match real experimental
behavior with excellent precision.

BG: Modelers do have a tendency to take their models more seriously than
others are likely to do.

BP: And naive realists do have a tendency to think they can understand
something without a model.

Well, look at that. Another conversation that started out friendly down
the drain. I think you and I are in some kind of positive feedback thing,
so what starts out stable and normal begins to jiggle, then oscillates,
and then runs away. It would be interesting if it weren’t so
frustrating.

Pooey.

Bill P.

Bruce_Gregory6 · January 9, 2010, 1:21am

[From Bruce Gregory (2010.0109.0120 GMT)]

[From Bill Powers (2010.01.08.1555 MST)]

Well, look at that. Another conversation that started out friendly down the drain. I think you and I are in some kind of positive feedback thing, so what starts out stable and normal begins to jiggle, then oscillates, and then runs away. It would be interesting if it weren't so frustrating.

BG: If I responded to this post I fear I would only prevent you from getting on with what you really want to do, constructing and testing models. So I will not respond. I do not believe that our exchange went down any drain. I'm sorry you feel that way. I'm simply trying to avoid taking up your time with qualitative rather than quantitative discussions. I am trying to respect your goals.

Best,

Bruce

Bruce_Gregory6 · January 9, 2010, 1:30pm

[From Bruce Gregory (2010.01.09.1330 GMT)]

A (probably futile) attempt at clarification. If the question is do I think that human behavior can be successfully modeled using HPCT, the answer is yes. This essentially means that I think human beings have goals and act to achieve those goals. HPCT has been shown to be a powerful tool in modeling goal-oriented behavior. (I suspect, but cannot prove, that it is a minimal-energy solution to the problem; that is, that other solutions require more computational resources. This is pure conjecture on my part.)

I do not think there is any evidence that brains are computers executing programs. I think this is a misleading analogy. I think a better analogy is that the brain is a memory that takes the form of a pattern matching engine. To the extent a computer program can simulate pattern matching, the program is useful; it is unlikely, however, that the steps in the program correspond to activities in the brain.

I had thought I might bring something useful to the discussion of HPCT. I now see that I was hopelessly naive. I apologize for wasting people's time, and I wish the enterprise well.

Bruce

Bill_Powers1 · January 9, 2010, 2:39pm

[From Bill Powers (2010.01.09.0639 MST)]

Bruce Gregory (2010.01.09.1330 GMT) --

I do not think there is any evidence that brains are computers executing programs. I think this is a misleading analogy. I think a better analogy is that the brain is a memory that takes the form of a pattern matching engine.

Bruce, I apologize for snapping at you, and I don't want you to leave again. Our problem is what to do when we disagree. In the above paragraph, you express a conclusion you have drawn, and suggest a better analogy. I clearly have a different view of things. So what do we do about that?

One approach, which is unfortunately very popular, is just to reiterate our points of view more emphatically, which ends up in a fight or leaving the playing field. Another is to explain the basis for each point of view, and if that fails, to explore that basis, and so on, iterating until the bedrock is exposed. I assume that long before the bedrock is reached, the causes of the disagreement will be found, and some means of resolving it can be proposed if it doesn't resolve itself as soon as it's understood.

I don't think the basis of our disagreements is visible yet. Clearly we have different views, but what makes them different? I assume it's not just some trivial random preference of the moment. Mine certainly isn't, and I don't think yours is, either.

Can we put aside, for the time being, the fact that we disagree, and simply start dissecting the issues to see their structure? Assuming you will want to do this, too, let me start.

We could start with your first statement above: you do not think there is any evidence that brains are computers executing programs. This suggests a concept of what a computer executing a program is. I can ask, is the sticking point for you the concept that a brain is a computer, or that the brain executes programs?

My concept is based on the latter. Let's start with the apparent evidence that the brain can, among all the other things it can do, execute procedures -- that is, it can produce sequences of results by repeating a sequence of operations. For example, the brain can carry out Newton's method for obtaining square roots. Starting with a guess, you divide the number whose square root you want by the initial guess, add the quotient to the first guess, and divide by two. That is your second guess. As you continue executing this procedure, you find that the guesses very quickly approach some value, getting closer and closer to it so you have to carry more and more significant figures to see a difference from the last guess.

We see a procedure being carried out again and again. But why does it ever stop?

It stops when the difference between one guess and the next is smaller than some amount, which is determined by how accurate a square root we want. If I want to know the square root of 12 to 3 significant figures, I continue repeating the procedure until the difference between successive guesses is less than 5 in the fourth figure. There is no easy way to predict how many repetitions that will require; it depends on the initial guess.
We don't normally try to predict the number and then do the procedure that many times, which may or may not be exactly the right number of times. Instead, we compare the difference between guesses with the accuracy we want, a reference value, testing after each iteration of the procedure. We stop when the difference is less than or equal to the reference value.

That makes the sequence of operations into a program. Miller, Galanter, and Pribram called it a "TOTE unit" in their "Plans and the Organization of Behavior" published in 1960, the same year the first paper in the series that led to B:CP was published. It's a very simple program, and more than a TOTE unit might be required to accomplish more complex ones, but this example will do for now.

This is not to say that brains are computers; I've said only that brains execute programs. I don't think there is any doubt that people execute learned programs, but if you have doubts about that, this would be the time to express and explain them. I hope you will.

Best,

Bill P.

Bruce_Gregory6 · January 9, 2010, 6:16pm

[From Bruce Gregory (2010.01.09.1815 GMT)]

[From Bill Powers (2010.01.09.0639 MST)]

Bruce Gregory (2010.01.09.1330 GMT) --

I do not think there is any evidence that brains are computers executing programs. I think this is a misleading analogy. I think a better analogy is that the brain is a memory that takes the form of a pattern matching engine.

Bruce, I apologize for snapping at you, and I don't want you to leave again. Our problem is what to do when we disagree. In the above paragraph, you express a conclusion you have drawn, and suggest a better analogy. I clearly have a different view of things. So what do we do about that?

BG: Apology accepted. I am sorry that I overreacted.

One approach, which is unfortunately very popular, is just to reiterate our points of view more emphatically, which ends up in a fight or leaving the playing field. Another is to explain the basis for each point of view, and if that fails, to explore that basis, and so on, iterating until the bedrock is exposed. I assume that long before the bedrock is reached, the causes of the disagreement will be found, and some means of resolving it can be proposed if it doesn't resolve itself as soon as it's understood.

I don't think the basis of our disagreements is visible yet. Clearly we have different views, but what makes them different? I assume it's not just some trivial random preference of the moment. Mine certainly isn't, and I don't think yours is, either.

Can we put aside, for the time being, the fact that we disagree, and simply start dissecting the issues to see their structure? Assuming you will want to do this, too, let me start.

We could start with your first statement above: you do not think there is any evidence that brains are computers executing programs. This suggests a concept of what a computer executing a program is. I can ask, is the sticking point for you the concept that a brain is a computer, or that the brain executes programs?

BG: I find there is a tendency to imagine the brain is executing all sorts of arcane programs. I think we share this view when it comes a picture in which the brain computes the forces need to throw a baseball and then executes a program that carries out the activity in the absence of feedback.

My concept is based on the latter. Let's start with the apparent evidence that the brain can, among all the other things it can do, execute procedures -- that is, it can produce sequences of results by repeating a sequence of operations. For example, the brain can carry out Newton's method for obtaining square roots. Starting with a guess, you divide the number whose square root you want by the initial guess, add the quotient to the first guess, and divide by two. That is your second guess. As you continue executing this procedure, you find that the guesses very quickly approach some value, getting closer and closer to it so you have to carry more and more significant figures to see a difference from the last guess.

BG: I agree that the brain can carry out procedures. Procedures involve sequences of actions. The brain learns to carry out such sequences. I think we agree that the actions are accompanied by changing perceptions. The way I think of the process is that the brain recalls what worked in the past. It then initiates a sequence of actions and controls these actions by negative feedback. You might say that the brain has a series of pictures and acts to match what it perceives with each of these pictures in turn.

We see a procedure being carried out again and again. But why does it ever stop?

BG: Just a note. Sometimes the procedure does not stop. Compulsive behavior suggests the breakdown of a mechanism that normally stops a procedure after a goal is reached.

It stops when the difference between one guess and the next is smaller than some amount, which is determined by how accurate a square root we want. If I want to know the square root of 12 to 3 significant figures, I continue repeating the procedure until the difference between successive guesses is less than 5 in the fourth figure. There is no easy way to predict how many repetitions that will require; it depends on the initial guess.
We don't normally try to predict the number and then do the procedure that many times, which may or may not be exactly the right number of times. Instead, we compare the difference between guesses with the accuracy we want, a reference value, testing after each iteration of the procedure. We stop when the difference is less than or equal to the reference value.

That makes the sequence of operations into a program. Miller, Galanter, and Pribram called it a "TOTE unit" in their "Plans and the Organization of Behavior" published in 1960, the same year the first paper in the series that led to B:CP was published. It's a very simple program, and more than a TOTE unit might be required to accomplish more complex ones, but this example will do for now.

This is not to say that brains are computers; I've said only that brains execute programs. I don't think there is any doubt that people execute learned programs, but if you have doubts about that, this would be the time to express and explain them. I hope you will.

No, I have no doubts about this. The brain can and does learn to carry out sequences of actions. Where we seem to differ is the extent to which in its operations the brain functions like a computer. As far as we know the brain has no analog to a CPU. The memory of the brain does not appear to be divided into bins capable of containing arbitrary information. Memories do not seem to be transmitted from one location to another location in the brain in order to be processed. I try to stick to the picture that the brain is a complex system of overlapping neural networks. What the brain "knows" is the result of the strengthening and weakening of parts of this network. When the brain encounters a new situation, it reactivates portions of the existing network. This seems to be consistent with the neuroscience and seems to me to be the result one would expect as the outcome of natural selection. I do not doubt that this system can be simulated by a computer.

Bruce

Bill_Powers1 · January 9, 2010, 8:27pm

[From Bill Powers (2010.01.09.1226 MST)]

Bruce Gregory (2010.01.09.1815 GMT) –

BG: Apology accepted. I am sorry that I overreacted.

BP: My son and I get into these escalating arguments once in a while, but
have learned to see the signs when they start and cut them off before
damage is done. It only takes one to break the positive feedback loop, so
whoever notices first just stops reacting, which leaves nothing to start
the next round. Let’s make a deal to do this with each other, OK? I wish
I had thought of this while my father was still alive.

BP earlier: We could start with
your first statement above: you do not think there is any evidence that
brains are computers executing programs. This suggests a concept of what
a computer executing a program is. I can ask, is the sticking point for
you the concept that a brain is a computer, or that the brain executes
programs?

BG: I find there is a tendency to imagine the brain is executing all
sorts of arcane programs. I think we share this view when it comes a
picture in which the brain computes the forces need to throw a baseball
and then executes a program that carries out the activity in the absence
of feedback.

BP: We’re definitely on the same side there.

BG: I agree that the brain can
carry out procedures. Procedures involve sequences of actions. The brain
learns to carry out such sequences. I think we agree that the actions are
accompanied by changing perceptions. The way I think of the process is
that the brain recalls what worked in the past. It then initiates a
sequence of actions and controls these actions by negative feedback. You
might say that the brain has a series of pictures and acts to match what
it perceives with each of these pictures in turn.

BP: That’s a good possibility. In B:CP I tried to come up with a
“recognizer” for a sequence of phonemes, but mostly as a way of
illustrating how simple functions implemented with neurons could be used
to do it. No serious proposal about a model. Note that I distinguish a
program from a procedure. A procedure is a fixed list of actions. A
program introduces tests and comparisons at a higher level: criteria that
decide whether to terminate one procedure and switch to a different one,
or restart the same one.

Later on I’ll tweak you about saying that the brain recalls what worked
in the past. If you think about it, though, I’m sure you will see that
this is exactly PCT says does NOT happen. Of course you may not be
referring to behaviors, actions, that “worked in the
past.”

BP earlier: We see a procedure
being carried out again and again. But why does it ever stop?

BG: Just a note. Sometimes the procedure does not stop. Compulsive
behavior suggests the breakdown of a mechanism that normally stops a
procedure after a goal is reached.

That’s a good idea. It sort of turns the problem upside down. Instead of
saying that compulsive behavior is caused by some ongoing problem, you’re
saying that whatever is supposed to turn it off when it succeeds has
failed. In fact, isn’t that the perfect PCT explanation? An error gives
rise to the behavior, but the behavior fails to correct the effect of a
disturbance, or the system fails to perceive that the effect has been
corrected, so the behavior just goes on and on. Reminds me of anorexia;
those poor skinny starving girls saying “I’m too fat!”

…

BP: This is not to say that brains are computers; I’ve said only that
brains execute programs. I don’t think there is any doubt that people
execute learned programs, but if you have doubts about that, this would
be the time to express and explain them. I hope you will.

BG: No, I have no doubts about
this. The brain can and does learn to carry out sequences of actions.
Where we seem to differ is the extent to which in its operations the
brain functions like a computer. As far as we know the brain has no
analog to a CPU. The memory of the brain does not appear to be divided
into bins capable of containing arbitrary information. Memories do not
seem to be transmitted from one location to another location in the brain
in order to be processed.

BP: I certainly agree with that, especially the last point. One of the
improbabilities involved in the “modern control theory” idea is
that a mental model can be designed and its parameters can be determined,
and then by inverting the matrix, those parameters can found and be
transfered to another place in the brain for use in equations that
compute “commands for action.” I’m sure I could do that on my
computer (if someone gave me the software), but I sincerely doubt that it
could happen in a brain. I don’t know of any way that synaptic weightings
could be detected and transferred to some other set of synapses in
another neural network. The brain just can’t know about things at that
level of organization. There are no sensors for synaptic weights, that I
have ever heard of.

BG: I try to stick to the
picture that the brain is a complex system of overlapping neural
networks. What the brain “knows” is the result of the
strengthening and weakening of parts of this network. When the brain
encounters a new situation, it reactivates portions of the existing
network. This seems to be consistent with the neuroscience and
seems to me to be the result one would expect as the outcome of natural
selection. I do not doubt that this system can be simulated by a
computer.

The only problem for me there is knowing what to simulate – the details
of what is in the networks. I don’t try to do much of that; the
neurological facts just aren’t there yet. My concerns are primarily with
WHAT has to be simulated, and figuring out a way to simulate it on a
computer. I figure that if I can see how to do it on a computer, the
brain can probably do it with neurons, although perhaps not the same way
I do it. Knowing that there is one way of doing it tells us that it’s not
impossible; it’s an existence theorem. That’s enough to tell us we can
get somewhere with it. I don’t know how the brain is organized to control
perceptions, except perhaps at the one or two lowest levels. But I can
show that one way works, and reproduces real behavior very well, so that
is a sign that we should look deeper to find out how the brain really
does it – when the technology catches up.

One thing about my computer simulations should be understood by all.
While they are done on a digital computer, the computations are not
digital; they are analog. My methods involve making a digital computer
act like an analog computer. I didn’t invent this (at one time I thought
I had); it’s used in a number of simulation programs like Stella and
Vensim. As I like to put it, an analog computer doesn’t solve systems of
equations; it acts them out. While you can write equations that can be
solved for variables in a control loop, my simulations don’t do that.
They just keep repeating the computations that describe each part of the
neural network, transforming inputs to a component into outputs, which
become inputs to one or more other components. The only way to know the
solution of the equations is to turn the simulation on and record how
variables at all points in the network behave through time. So my
simulations are a great deal more like observing the behavior of a neural
network than they are like solving equations, inverting matrices, and all
the rest of that. And all parts of the neural network are active at the
same time – there is no CPU. It’s a bit tricky to get a digital computer
with a single CPU to act that way, but not too tricky to do.

I’m glad that you understood and accepted my peace offering. Let’s not
forget how to do that.

Best,

Bill P.

Bruce_Gregory6 · January 10, 2010, 12:14pm

[From Bruce Gregory (2010.01.10.1215 GMT)]

[From Bill Powers (2010.01.09.1226 MST)]

BP: My son and I get into these escalating arguments once in a while, but
have learned to see the signs when they start and cut them off before
damage is done. It only takes one to break the positive feedback loop, so
whoever notices first just stops reacting, which leaves nothing to start
the next round. Let’s make a deal to do this with each other, OK? I wish
I had thought of this while my father was still alive.

BG: Deal!

BP earlier: We could start with
your first statement above: you do not think there is any evidence that
brains are computers executing programs. This suggests a concept of what
a computer executing a program is. I can ask, is the sticking point for
you the concept that a brain is a computer, or that the brain executes
programs?

BG: I find there is a tendency to imagine the brain is executing all
sorts of arcane programs. I think we share this view when it comes a
picture in which the brain computes the forces need to throw a baseball
and then executes a program that carries out the activity in the absence
of feedback.

BP: We’re definitely on the same side there.

BG: I agree that the brain can
carry out procedures. Procedures involve sequences of actions. The brain
learns to carry out such sequences. I think we agree that the actions are
accompanied by changing perceptions. The way I think of the process is
that the brain recalls what worked in the past. It then initiates a
sequence of actions and controls these actions by negative feedback. You
might say that the brain has a series of pictures and acts to match what
it perceives with each of these pictures in turn.

BP: That’s a good possibility. In B:CP I tried to come up with a
“recognizer” for a sequence of phonemes, but mostly as a way of
illustrating how simple functions implemented with neurons could be used
to do it. No serious proposal about a model. Note that I distinguish a
program from a procedure. A procedure is a fixed list of actions. A
program introduces tests and comparisons at a higher level: criteria that
decide whether to terminate one procedure and switch to a different one,
or restart the same one.

BG: Fair enough. I use to term procedure because the term program is likely to suggest a computer is at work. Neural networks can “make decisions” on the basis of changing activation and inhibition. We can call these programs.

Later on I’ll tweak you about saying that the brain recalls what worked
in the past. If you think about it, though, I’m sure you will see that
this is exactly PCT says does NOT happen. Of course you may not be
referring to behaviors, actions, that “worked in the
past.”

BG: The brain seems to recall the “gist” of a situation rather than the details. Studies of “change blindness” reveal that great deal can change (including the stranger we are talking to!) without the brain registering that something has happened. I suspect that the brain remembers the gist of the action. Enough to initiate an action with is then controlled by negative feedback. (Drive to Adams Supermarket to pick up a container of milk.)

BG: No, I have no doubts about
this. The brain can and does learn to carry out sequences of actions.
Where we seem to differ is the extent to which in its operations the
brain functions like a computer. As far as we know the brain has no
analog to a CPU. The memory of the brain does not appear to be divided
into bins capable of containing arbitrary information. Memories do not
seem to be transmitted from one location to another location in the brain
in order to be processed.

BP: I certainly agree with that, especially the last point. One of the
improbabilities involved in the “modern control theory” idea is
that a mental model can be designed and its parameters can be determined,
and then by inverting the matrix, those parameters can found and be
transfered to another place in the brain for use in equations that
compute “commands for action.” I’m sure I could do that on my
computer (if someone gave me the software), but I sincerely doubt that it
could happen in a brain. I don’t know of any way that synaptic weightings
could be detected and transferred to some other set of synapses in
another neural network. The brain just can’t know about things at that
level of organization. There are no sensors for synaptic weights, that I
have ever heard of.

BG: I definitely agree with that.

BG: I try to stick to the
picture that the brain is a complex system of overlapping neural
networks. What the brain “knows” is the result of the
strengthening and weakening of parts of this network. When the brain
encounters a new situation, it reactivates portions of the existing
network. This seems to be consistent with the neuroscience and
seems to me to be the result one would expect as the outcome of natural
selection. I do not doubt that this system can be simulated by a
computer.

The only problem for me there is knowing what to simulate – the details
of what is in the networks. I don’t try to do much of that; the
neurological facts just aren’t there yet. My concerns are primarily with
WHAT has to be simulated, and figuring out a way to simulate it on a
computer. I figure that if I can see how to do it on a computer, the
brain can probably do it with neurons, although perhaps not the same way
I do it. Knowing that there is one way of doing it tells us that it’s not
impossible; it’s an existence theorem. That’s enough to tell us we can
get somewhere with it. I don’t know how the brain is organized to control
perceptions, except perhaps at the one or two lowest levels. But I can
show that one way works, and reproduces real behavior very well, so that
is a sign that we should look deeper to find out how the brain really
does it – when the technology catches up.

BG: Again , I definitely agree.

One thing about my computer simulations should be understood by all.
While they are done on a digital computer, the computations are not
digital; they are analog. My methods involve making a digital computer
act like an analog computer. I didn’t invent this (at one time I thought
I had); it’s used in a number of simulation programs like Stella and
Vensim. As I like to put it, an analog computer doesn’t solve systems of
equations; it acts them out. While you can write equations that can be
solved for variables in a control loop, my simulations don’t do that.
They just keep repeating the computations that describe each part of the
neural network, transforming inputs to a component into outputs, which
become inputs to one or more other components. The only way to know the
solution of the equations is to turn the simulation on and record how
variables at all points in the network behave through time. So my
simulations are a great deal more like observing the behavior of a neural
network than they are like solving equations, inverting matrices, and all
the rest of that. And all parts of the neural network are active at the
same time – there is no CPU. It’s a bit tricky to get a digital computer
with a single CPU to act that way, but not too tricky to do.

BG: I did not know that before. I am glad that I do now.

Bruce