I have been searching high and low on Discourse, and by extension, CGSNet. but I can’t find an answer to this question. I’m not looking for an answer involving evolution, brain processes or anatomy, or the source of intrinsic reference values, and we know that the answer will involve reorganisation and most likely be described as occurring in awareness, and we know that their purpose is to restore of improve control to reduce intrinsic error, often by helping to resolve conflict, and the end product of this process is often felt as a new perspective or even an epiphany. My question is more mechanistic: “How are new input functions created?”
I think this boils down to what are the conditions or processes that are necessary to create them, and specifically form the unique algorithm that is contained in each notation of ‘f’ where we have the input function at each level. It must involve bringing together signals from input functions at lower levels and applying transformations to those signals - at the very least summing them, but also differentiating them, integrating them possibly, and ordering this process?
Does anyone know if Bill answered this, as I would be very keen in quoting him if you have the quote and its source.
Hope you are al doing well and I look forward to hearing!
WM:I have been searching high and low on Discourse, and by extension, CGSNet. but I can’t find an answer to this question. I’m not looking for an answer involving evolution, brain processes or anatomy,
RM: Then I might not be able to help you because my answer will involve the evolution of brain processes. But I’ll give it to you anyway.
WM: My question is more mechanistic: “How are new input functions created?”
RM: This is an ambiguous question: are you asking how new types of input functions are created (input functions that can perceive what could not be perceived before) or are you asking how input functions of an existing type are added to one’s repertoire of control systems. I’ll actually try to answer both.
RM: According to PCT, organisms control a hierarchy of different types of perceptions. There is evidence that this is the case and that the types range from perceptions of sensory intensity to perceptions of principles and system concepts.
RM: The existence of different types of perceptions implies that there exist different mechanisms that produce them. For example, there is good evidence that perceptions of sensory intensity can be produced by input functions that are implemented as simple transduction mechanisms; perceptions of sensations, like color and pitch, are apparently produced by input functions that are implemented as a mechanism that integrates the outputs of several sensory intensity transducers; perceptions of more complex perceptions, such as principles, are produced by input functions that are implemented by neural mechanisms that we don’t understand at all yet.
RM: It’s highly unlikely that organisms develop, within the timeframe of a few weeks or even years, input functions that can produce new types of perceptions. I think what evidence we have of interspecies differences in what perceptual variables can be controlled suggest that new types of input functions – input functions that can perceive the world in new ways – evolve as an adaptive mechanism.
RM: Taking Powers’ proposed hierarchy as a working hypothesis, it seems to me that “lower” organisms – one’s with less developed brains than ours – cannot perceive (and control) the world in terms of the higher level types of perceptions that humans can. Lower organisms can certainly perceive and control the world in terms of lower level perceptual types – intensities, sensations, configurations, etc – although they often perceive different values of these variable, such as bats and dogs that can here pitch sensations well out of the range that we can hear. But I believe that even our closest relatives, the great apes, cannot perceive and control the world in terms of the higher level perceptions that we can.
RM: (I think this would be a great area for comparative research, by the way; I wish I had mentioned it in my book. The research would involve looking for differences across species in terms of the types of perceptual variables that can be controlled. The book did mention research by the Plooij’s where they looked at the age at which the ability to control different types of variables comes “on-line”. To the extent that ontogeny recapitulates phylogeny, this is a first step toward doing cross species studies of the types of variables the different species can control. )
RM: So how do we account for the fact that the different types of controlled variables come on line at different points in development? My guess is that it’s because we have to learn how our variations in lower level perceptions affect higher level perceptions; and we can’t find this out until we have learned how to vary still lower level perceptions in order to control those at the next higher level up.
RM: So we have to learn how to vary intensity perceptions (muscle tensions) in order to control sensation perceptions (limb position) and we have to learn how to vary sensation perceptions in order to control transition perceptions (limb movement). But I believe that the input functions that produce perceptions of intensity, sensation and transition are built in. We don’t learn to perceive in new ways, we learn how to vary outputs in order to have the desired effect on these perceptions.
RM: But what about learning how to perceive and control new instances of the same type of perception? For example, once we know how to control the position of our wrist, how do we learn to control for a new wrist position perception – the one that gives us the best forehand stroke in tennis? I think what is being learned here is not a new perception but a new state of a perceptual variable that we have come into the world “knowing” how to perceive. Learning a new wrist position is a memory phenomenon – what you learn is a stored reference specification for the wrist position (lower level controlled perception) to use to make a great forearm stroke (higher level controlled perception).
RM: This evolutionary explanation of how we develop new input functions seems to explain something I have always had trouble understanding: How could humans, with a brain that evolved to be just smart enough to give us a slight adaptive advantage over our ape and homo ancestors, have used that brain to invent the incredibly complex science and technologies that we have today? I think the answer has something to do with the fact that the new types of perceptual input functions that evolved beyond those processed by the great apes and early homos – possibly the program, principle and system concept types – made it possible to do a lot more than the things we could do when we first developed those perceptual abilities. For example, the program type input functions, that might have evolved to allow us to perceive and control more effective hunting strategies, could also be used to perceive and control other kinds of programs, such as the program of actions required plan and build a Boeing 787.
RM: Anyway, that’s my story and I’m stickin’ to it – at least until someone does some relevant research to test it.
Thanks Rick. Nearly everything you say makes sense and it helps me direct the attention to where the exceptions might be - the input functions not prepared by evolution- as that’s my aim.
So is it implicit in what you are saying that principle and system concepts input functions must be created rather than evolutionarily prepared. Obviously the machinery to construct those levels is evolutionarily prepared but what about the functions that determine the ‘identity’ of the specific principle or system concept (the term identity as a property of a variable I got from Powers).
Second, surely it’s necessary that the function that establishes the identity of a specific category of object is constructed through one’s early life rather than prepared. We aren’t born to recognise a chair, a phone, or even a duck billed platypus? So many of the ‘mid’ level functions are created through reorganisation too?
I’d rather work with you (or someone else) in the details of this than argue about it if you think it’s possible.
Talk to you soon
Hi have another thought about ‘information’ that has been supported by this article, but it takes it to PCT:
So ‘control information’ is information that which can be reliably re-experienced and identified by the perceiver regardless of disturbances. To do so requires an input function that identifies this source of information, and a means to act against disturbances (e.g. movements, distortions, loss of data) that might render it unidentified. Otherwise the purpose or function of the information to the perceiver is redundant – it is just a meaningless pattern impinging on the senses.
WM: So is it implicit in what you are saying that principle and system concepts input functions must be created rather than evolutionarily prepared. Obviously the machinery to construct those levels is evolutionarily prepared but what about the functions that determine the ‘identity’ of the specific principle or system concept (the term identity as a property of a variable I got from Powers).
RM: I think this is close to getting close to my view. I would say that what you call “the machinery to construct those levels” is what I call the input functions themselves. And what is “constructed” are the lower level perceptual inputs to these functions.
WM: Second, surely it’s necessary that the function that establishes the identity of a specific category of object is constructed through one’s early life rather than prepared.
RM: I think the functions that produce categorical perceptions are built in; the categories themselves are learned in terms of what inputs go into these category perception functions. But maybe this can be tested. It just seems highly unlikely to me that if people were always constructing (via reorganization) perceptual input functions it would be unlikely that everyone would end up constructing functions that always ended up perceiving the world in terms of the 9 or ten types that seem to exist.
WM: We aren’t born to recognise a chair, a phone, or even a duck billed platypus?
RM: No, but I think we are born to recognize categories (and sequences and principles, etc); what we learn is how to categorize the lower order perceptions that are the inputs to these functions.
WM: So many of the ‘mid’ level functions are created through reorganisation too?
RM: Not the functions themselves; I think it’s the inputs to the functions that are reorganized.
WM: I’d rather work with you (or someone else) in the details of this than argue about it if you think it’s possible.
RM: In one of my earlier posts I suggested one kind of research that I think could be very interesting and could possibly be done by just going to archival sources. The proposal was to look through the literature and do a comparative analysis of the types of variables different species can control. Maybe just limit it to primates.
RM: If you are interested in studying “perceptual learning” I would be interested in hearing what phenomenon you are trying to explain and what kind of research you are thinking of doing to test your explanation.
WM: "How are new input functions created?”
I found one short paragraph in LCS III, page 133, the footnote 32.
“The subject of energy conservation, although irrelevant to this demonstration, may be important in understanding how the input function might be reorganized. […] A condition of minimal internal conflict is reached when the three sets of input weights approach orthogonality. Therefore, total energy usage could possibly work as a criterion for reorganizing the input functions to achieve independent dimensions of sensing”
It would be great to test something like this, though I’m not sure what is the order or timescale of reorganization in the output and input. For example - are the input functions created simultaneously as the output functions, both going from 'undifferentiated to ‘precisely defined’, or one of them is more fully formed, and then the other one; etc.
There is also From Bill Powers (920711.0800) with some vector input functions reorganizing.
We might also consider Bill’s idea about bacteria chemotaxing, for example BP (920702.0800).
Let’s say some microorganism has several receptors, and starts with a randomized number and gain for - it does not “know” if it should swim up gradient or down gradient. Reorganization started by intrinsic variable error might change the input function to make some of those receptors have a different gain. Then the organism might swim in a different direction.
Hi Rick, I came across this from Bill’s comments on Maturana. Don’t you think that we need to have a place for ‘primitive’ discrimination of inputs without going to the category level?
Powers (1987) “The first six of these levels create an analogue world of perception, in which all perceptions can vary smoothly over their whole possible ranges. Many of these perceptions are under skillful control, in a hierarchy that operates on continuous variables without benefit of symbol manipulations. This is the world that provides a continuum to cleave.”
ME: But what about different configurations? And different senses? The different senses make discriminations purely based on the location and nature of the sense organs. The different configurations can be discriminated – e.g. snake shape or spider shape, by prey, and they need to be. A whole area of ethology involves predator and prey discrimination in quite primitive animals and mimicry. Surely that doesn’t require a category level perception?
WM: Hi Rick, I came across this from Bill’s comments on Maturana. Don’t you think that we need to have a place for ‘primitive’ discrimination of inputs without going to the category level?
RM: I’m not sure if I have an opinion on that.
WM: Powers (1987) “The first six of these levels create an analogue world of perception, in which all perceptions can vary smoothly over their whole possible ranges. Many of these perceptions are under skillful control, in a hierarchy that operates on continuous variables without benefit of symbol manipulations. This is the world that provides a continuum to cleave.”
RM: I found the reference; it’s in LCS II, the chapter on Control Theory, Constructivism and Autopoesis. Can that man write or what!
WM: But what about different configurations? And different senses? The different senses make discriminations purely based on the location and nature of the sense organs. The different configurations can be discriminated – e.g. snake shape or spider shape, by prey, and they need to be. A whole area of ethology involves predator and prey discrimination in quite primitive animals and mimicry. Surely that doesn’t require a category level perception?
RM: Maybe, maybe not. I think it’s an empirical question. For example, what perceptions are being controlled when an organism “discriminates” predator from prey, male from female, food from non-food? It’s probably different for different organisms. Lower organisms may do it based lower level perceptions; higher organisms may do it based on higher level sensations.
RM: By the way, I put “discriminates” in scare quotes because I think the idea that organisms discriminate, say, predator from prey, is a description of what’s going on from the observer’s perspective. The organism itself may simply be controlling to keep certain perceptual variables in reference states; perceptions corresponding to what you see as a predator are controlled by avoiding them and perceptions corresponding to what you see as prey are controlled by approaching them.
RM: By the way, I don’t see what this has do do with how new input functions are created. What am I missing?
It’s not clear what distinction you’re making, Rick.
They’re not inputs until they enter an input function. ‘Reorganizing the inputs’ is precisely how the input function is reorganized.
The other possible interpretation is that you mean the control systems which control lower-level perceptual signals are reorganized at a point before those signals become inputs at the higher level. That says nothing about how input functions are created or reorganized. If the lower-level reorganization affects lower-level perceptual input functions, the implied successive regression is not infinite, and at some level input functions at that level have to be created or reorganized and the punting down-hierarchy has to stop.
Warren, I’m glad you’re asking this question. We have almost no detailed understanding of the synapsing of signals to create one signal of a different order. This essential topic has been almost entirely neglected, though it is the grit and substance of HPCT. It includes the converse decomposition of an error signal into branched and weighted reference inputs and it includes reference input functions receiving a plurality of such signals, ideally one at a time but if more than one arrives synapsing them to create one ‘conflicted’ reference signal.
Hi Bruce, I’m pleased to hear it makes sense to you! There is clearly a neural answer to this question, but I’m particularly interested in the experiential answer. Do you think there would be any purpose of some conditions or product of this process of development of new input functions being conscious?
WM: So many of the ‘mid’ level functions are created through reorganisation too?
RM: Not the functions themselves; I think it’s the inputs to the functions that are reorganized.
BN: It’s not clear what distinction you’re making, Rick.
RM: OK, I’ll try again. Let’s take Bill’s example of an input (perceptual) function that produces a “sensation” type perceptual signal (p) that is an analog of the taste of lemonade. Bill imagines this perceptual function to be a linear combination of lower level “intensity” perceptual signals generated by sugar (s), acid (a) and other oils (o1, o2…on).
RM: So, the input function that produces a perceptual variable that could be called “lemonadeness” might look like p = k1s + k2a + k3*o1, where s, a, and o1 are the inputs. One physiologically plausible way I can imagine this perceptual function changing is in terms of values of the weighting parameters (k1,k2, k3); another physiologically plausible way is in terms of the inputs to the function; instead of s, a and o1 the inputs might change to s, a and o2 as inputs.
RM: Both of these changes – in parameters and/ or inputs – would result in “new” perceptual variables. But they would not produce a new type of perceptual variable; the variable would still be a sensation-type perceptual variable (per Bill’s model described in B:CP). What defines the type of perceptual variable is what I call the “architecture” of the perceptual input function; in this case, the architecture is a weighted sum of inputs.
RM: Since the PCT hypothesis is that all humans perceive the world in terms of the same 9 or 10 different types of perceptual variables – intensities, sensations, configurations, transitions…principles, system concepts – the implication is that the perceptual architectures that produce these different types of perceptual variables are “built in” by evolution. We can presumably learn to perceive (and control) new instances of these different types of perceptual variables (by varying the parameters and/or the inputs to perceptual functions) but we can’t learn to perceive new types of variables; we can learn to perceive and control new programs and principles, for example, but we can’t learn to experience the world in a new way that is not one of the 9 or 10 ways described in the PCT hierarchy.
Of course, this has to be tested. If we did these tests we would expect to find people developing the ability to control new perceptions but not new types of perceptions. Apparently, no one seems particularly interested in doing these tests so, to quote a great old Phil Ochs song, I guess I’ll have to do it while I’m here;-)
Rick, Some question:
Did I understand you right that you think that our types of perception – levels of perceptual hierarchy – are evolutionarily determined and by reorganization we do not and cannot create new types of perceptions? So that a child does already innately have all the (8-9) hierarchical levels of perception.
Do you think that reorganization can create new perceptions (i.e. input functions) inside these levels or do the neural structures for every possible individual perception already exist in the brain of the new born child?
The form of the input function of “sensation” type perception is p = (k1x + k2y … + kiz9), where p is the (strength of the) perceptual signal, and k1,k2, etc. are weighing parameters, and y, x, etc. are the (strengths of the) “intensity” type perceptual signals. Right?
What then is the form a “configuration” type perception, or some other type of perception? Do the types of perceptions differ according to the forms of their input functions?
I wasn’t asking about new ‘types’ of input function in the first place. So now we understand each other, please return to my original question if you want to help…
How do we get to recognition that a novel combination of perceptual inputs (concurrent or sequential) can serve a purpose?
I think to understand this we need observational data about the planning/problem-solving level above sequences. I believe that calling this a ‘program’ level has been a grossly misleading red herring. Operations at this level are not so fastidious and elegant as programmers and logicians prefer but rather are pragmatic and opportunistic hacks. The arbitrary interruptability of sequences makes it possible to understand if-then-else contingencies as merely control of perceptions that are present—if the input fits, wear it—and seeking out means to control a next step toward a goal. Take your shoe off and rap the lid of the paint can back tight with its heel, swinging it by the toe. I opened that line of insight at Manchester, and unfortunately can’t expand much on it just now. Do some reconstruction of what you are actually doing at that level to see for yourself.
We don’t have to invoke biological reorganization to see how trial and error can acquire control of a novel combination of perceptual inputs that serve our purpose. If it works well there’s a level of perception—that same planning level, I believe—which perceives that it works well and controls it in imagination (That was cool, how did I do that?), helping to establish it in long-term memory. The same kind of evaluative perception can look at a recurrent routine way of doing something, control a variation in imagination, and improve the process (or not).
Trial and error and introspective process improvement work at lower levels too, but in my experience they are carried out first at the ‘instructive’ sequence level (turn your wrist at this angle with your elbow thus relative to your shoulder and hip) and subsequently established in memory by practice as an improved forehand swing.
From Peter Corning’s article: “Control information is not a thing or a mechanism but an attribute of the relationships between things. It is defined as: the capacity (know how) to control the acquisition, disposition and utilization of matter/energy in ‘purposive’ (cybernetic) processes.” Maybe he spells this out more clearly in the body of the article. To make this statement coherent in PCT terms: The capacity to control is an ‘attribute’ of a control system that controls perceptions of things and perceptions of relationships between those things. Its doing so presents to an observer that it is acquiring, disposing, and utilizing matter and energy purposefully. The ‘control information’ that enables this is embodied in the hierarchical structure of the control system (a legitimate use of the buzz-word ‘embodied’).
EP: Rick, Some question:
EP: Did I understand you right that you think that our types of perception – levels of perceptual hierarchy – are evolutionarily determined and by reorganization we do not and cannot create new types of perceptions? So that a child does already innately have all the (8-9) hierarchical levels of perception.
RM: Yes. I say this because, if the model is right, then it implies that all humans experience the world in terms of the same 9 or 10 different types of perceptual variables: intensities, sensations, configurations, transitions, event/sequence, relationship, category, program., principle, system concept. Given the complexity of the neural networks involved in constructing such perceptual variables – particularly those above sensations – it seems to me that it is very unlikely that all humans would be able to develop such complex neural networks, let alone neural networks that always construct the same kinds of perceptual variable, in the short period during which a person develops the ability to perceive and control the world in terms of these different percpetual varia
EP: Do you think that reorganization can create new perceptions (i.e. input functions) inside these levels or do the neural structures for every possible individual perception already exist in the brain of the new born child?
RM: I would say the former. But then I realized that the latter seems to be true of intensities and sensations; we seem to come into the world pre-wired to experience every possible sensory intensity (all brightness levels, loudness levels, etc) and every possible sensation (all colors, pitches, tastes, etc). So I am willing to consider the possibility that this is true for all types of perceptual variables in the hierarchy.
RM: Perhaps what we experience as learning to perceive in a new way doesn’t involve learning new input functions at all. Perhaps we are just learning to notice perceptual variables that were always there; or learning to control perceptual variables that were always available to be controlled but hadn’t been controlled before (as in the Robertson/Glines experiment and Plooij studies described in the Learning chapter of my book The Study of Living Control Systems.
RM: My confusion about this – whether instances of perceptual types are learned or built in or both) is why I prefer to take a “phenomena phirst” approach to understanding behavior. At least we take this approach, which means talking about this stuff in terms of actual observations of behavior – we know that our questions are about something that is happening in our experience rather than just in our imagination.
EP: The form of the input function of “sensation” type perception is p = (k1x + k2y … + kiz9), where p is the (strength of the) perceptual signal, and k1,k2, etc. are weighing parameters, and y, x, etc. are the (strengths of the) “intensity” type perceptual signals. Right?
RM: Yes, that’s the theory.
EP: What then is the form a “configuration” type perception, or some other type of perception? Do the types of perceptions differ according to the forms of their input functions?
RM: Bill never proposed a model of the form of the neural input functions that produce configuration type perceptions. There are a lot of theories in the computational perception literature about how this might be done; computer scientists have gotten pretty good at configuration perception (I’m able to reply to this post because my computer recognized my face and allowed me to get on); but I don’t think computer configuration perception is carried out in anything like the way like it is carried out by the nervous system.
RM: We did discuss the model Bill proposed (in B:CP) of the form of a neural input function that can produce sequence type perceptual variables. I think he just proposed it to show readers that a neural perceptual function for a fairly complex controlled variable was feasible.
WM: I wasn’t asking about new ‘types’ of input function in the first place. So now we understand each other, please return to my original question if you want to help…
RM: I’m still not sure we do create new input functions, of any kind. As I just said to Eetu, it seems to me that many of the observations I can think of that could be seen as examples of “creating new input functions” could also be seen as learning to notice perceptual variables that were always there or learning to control perceptual variables that were always available to be controlled but hadn’t been controlled before.
RM: I would love to help you answer your question about how new input functions are created. But I think it would help if you first described the observations that lead you to believe that new input functions are created and why they do.
BN: How do we get to recognition that a novel combination of perceptual inputs (concurrent or sequential) can serve a purpose?
BN: I think to understand this we need observational data about the planning/problem-solving level above sequences.
BN: I believe that calling this a ‘program’ level has been a grossly misleading red herring.
RM: I’m pretty sure Bill was never in the business of selling red herrings;-). Bill defined a program as a “network of contingencies”. My program control demo shows that the ability to control of a program (such as “if the shape is circle, the next color is blue; else, the next color is red”) requires that the elements of the program (the shapes and colors, in this case) be presented at a much slower rate than that at which it is possible to control a sequence (such as “small”, “medium”, “large”) of the same components.
RM: I don’t understand what you have against the idea the we control program perceptions but the fact is that we can and we do, as demonstrated by the program control demo. I believe that you tried to dismiss this fact by claiming that program control is actually sequence control where what you are controlling are “sub-sequences”, not programs. If this were the case then it should be possible to control the sub-sequences at the same presentation rate as that at which one can control a sequence.
RM: But control of these “sub-sequences” (which are the contingencies in the program) can only be done when the rate of presentation of the components of the program is much slower than that at which one can control a sequence. This is very clear evidence that programs are not sequences of sequence perceptions.
RM: Programs are a different type of perception than sequences and the program control demo shows that program perceptions are controlled at a higher level than sequence perceptions in a control hierarchy. This is demonstrated by the fact that a program can only be controlled when its components are presented at a slower rate than that at which it is possible to control a sequence.
OK thanks Rick, let me work on this a while!