Re: The nature of "internal predictive models"

[Judith Rosen <***>]

On Oct 18, 2005, at 7:35 PM, Tom Staiger wrote:
> <x-tad-smaller> In Anticipatory Systems I recalled reading that the likelihood of a system failing increased with a greater discrepancy between the system's behaviors and its models of what was actually occurring or about to occur. </x-tad-smaller>
>

This is close, but while I know that you understand the concept fine, there are a couple of things I want to clarify for the list. First off, we are talking about an "anticipatory system" when we are discussing internal predictive models linked to expressed behavior of the system. In my father's view, where naturally occurring systems are concerned, "anticipatory systems" means we are talking about organisms (living systems). Secondly, when an anticipatory system's behavior diverges from what is appropriate for environmental conditions, that's a symptom of a problem with the information encoded into its "internal predictive models." If I may take the liberty, I can use your wording to show what I mean: 

The likelihood of [an anticipatory] system failing increases with a greater discrepancy between the system's behaviors [which are based on the models' encoded information] and what is actually occurring or about to incur [in whatever actual system the models' encoded information refers to]. 

When any given model and the system it represents don't "commute" (meaning; "have the same entailment pattern") then all predictions based on the model's entailment pattern are likely to be out of synch with the behavior of the actual system. Usually, when this occurs in organisms, it is due to some radical change in the organism's native habitat's environmental conditions although there are many different aspects of context which appear to be modeled in living system organization, including internal context.  

To flesh out that description a bit: The phrase "Internal predictive models" is the label or name Robert Rosen used to describe the presence of encoded information (about various aspects of context, both internal and external) that, he said, logically must exist within the overall living-system organization. He never specified how the information gets encoded, what form the encoded information has within the system, nor how the information is able to trigger adaptive behaviors in the organism-- he didn't get that far in his own research before he died. He had more than a few thoughts on the subject/s. However, the only conclusion he felt sure enough of to actually publish was that such encoded information must be there because living organisms engage in "characteristically model-based behaviors"-- complete with characteristically model-based errors. In his investigation into what organic phenomena could be capable of causing this kind of pattern that could, in any way, be considered a model, he arrived at "encoded information".  It may sound ludicrous to empiricists or experimental quantitative scientists, etc... and they have expressed sentiments to that effect in my hearing at some of the conferences, symposia, and meetings I attended with him. Frankly, though, I don't see how anyone can refute the fact that genetic "information" is encoded information. It's abstracted in that it is separated from the aspects of the organism that the individual chromosomes are directly responsible for affecting; therefore it can be designated as "encoded information" that we know exists within living system organization. The existence of genetic material cannot be adequately explained by a purely reactive paradigm based on models of machines. However, the material nature of its existence makes its presence in living organisms an unquestioned reality and science has concentrated mostly on studying the material aspects. Luckily, this is a far easier job to explain using  reductionist methodology than the relation of those material aspects to the functional capabilities of it, or the otherwise incomprehensible fact that newborn baby girls have all their ovum, already formed-- containing the genetic results of meiosis which will be required for any future fertilization to be possible. If that isn't evidence of an anticipatory mode of system control, I don't know how it can be logically explained.

Part of the trouble with accepting the concept of anticipation in living system organization is caused by the fact that living organisms can also be reactive as well. There's no question about that. However, along with life is this other emergent capability of having more options built into the system design-- or perhaps what we recognize as "life" is actually just the organic consequences of an extremely complex organizational type. In any case, living organisms demonstrate behaviors that are both reactive and anticipatory. What intrigues me most is that, when they are engaging in anticipatory behavior, they are actually reacting to predicted events-- events predicted by their internal models. 

I've noticed that the word "predicted" is the red flag for quantitative science. It  certainly seems to be the sticking point for many of the aforementioned scientists, because as they have complained after some of my father's talks; no single-cellular life form can "think" much less consult a model and decide on behavior based on it. How true: there is no "thought process" involved, no choice in whether or not to engage in the behaviors predicated on the encoded information.  Similarly, a single-celled life form is engaging in a whole lot of behaviors that are dictated by the genetic information which is encoded into its organization. Again; no choice and no thought process involved.  Furthermore, even in the area of genetic inheritance, we have overwhelming evidence that there is a great deal more to the behavior of organisms than can be completely explained by the genetic information in the genome. So there have to be more modalities by which information can be encoded, stored, and expressed within the organization than simply the chemical one geneticists have been focusing on. 

Another problem with the notion of anticipation, as my father described it, apparently has to do with Physics-based assumptions about the nature of time. Specifically, time is assumed to be a unidirectional and linear constant, such that we can only progress in one direction, at a constant rate, from past toward future, with our current state space on the time line labeled "the present." However, the behavior of living organisms reveals a pattern of engaging in change BEFORE such changes will become necessary for survival/optimality. The sorts of changes which are engaged in via an anticipatory mode are all based on cycles of time; repeating patterns or rhythms which recur with regularity... like clockwork (as the saying goes...)(sorry, I couldn't resist). So, I can predict that the sun will become visible in the eastern sky a certain number of hours after it disappears on the western horizon-- it's not that I'm prescient; it's that this is a pattern that is so regular as to be predictable. It's a rhythm that is actually part of our physiology. How? I'd sure like to know!

Out of sensitivity to those who have a problem with the word "predictive"... we can easily reword my first description, above, to remove the words based on "predict" and still retain the accuracy of the description. I'll give it a shot: 

Anticipatory behavior describes a characteristic pattern of behavior in a living organism that is caused when the organism reacts to entailments generated via relational interactions between aspects of encoded information within its own organization.  If we decide to collectively name this encoded information  "a SET OF MODELS" then it is logical to say that such behavior is "model-based". 

In order for encoded information to trigger systemic behaviors, whereby the organism is reacting to events which have not yet happened but are (according to the models) about to happen, my father was really forced by the evidence to conclude that these models must incorporate information about multiple aspects of time as well as information about internal and external contextual behaviors, AND about relations between all. That's a very sophisticated collection of information. It would also be logical to conclude that this encoded information must be passed from generation to generation, through reproduction within species, because behavior patterns also hold within geographically co-evolved individuals of a species and begin to diverge when a group is geographically separated from the rest of their species. We can also demonstrate that organism species are not able to change the encodings of the information in their internal models in the short term span of "real time," however they are somehow able to be changed over evolutionary time. This is absolutely consistent with what we know about genetic information.. In fact, genetic information is part of the internal predictive models.

I have said before that concepts such as survival and optimality are also anticipatory.  Organisms seem to be organized around certain built-in potentials: The potential for continuation through time (survival) and the potential for constant change requiring constant rebalancing (optimality). If the encoded information refers to aspects of context that we would identify as "the environment" for example, then any behavioral changes based on that information will only be optimal when the information "commutes" with the environmental system's actual behaviors. If the information encoded into the models generates entailments (predictions entailed by the information) which turn out to be different enough from the behavior of the actual environment, obviously the course of action the models trigger in the organism has a greater chance of being disastrously inappropriate for actual conditions.  

I've noted before that global climate change is likely to give humanity a crash course in what happens to organisms/species when things like the local seasonal cycles behave differently from internal encoded information about those cycles. The consequences will likely be all over the map: some inappropriate behaviors do no harm, some may even provide some sort of survival advantage-- purely by luck of the evolutionary draw... but the vast majority of inappropriate behaviors will be suboptimal at best and lethal at worst. When it proves to be lethal to some of the keystone species of the main systemic aspects of our planet, such as food chain or atmospheric composition ratios (phytoplankton comes to mind) it becomes obvious that this is a relational chain reaction that we don't want to have to try and figure out how to live through. If I could find a way to get that idea through to Mr. Bush Jr. I would be a whole lot more hopeful about averting that particular entailed future.


Judith
Web address: http://www.rosen-enterprises.com
BioTheory: An electronic journal of general science based on the Relational (Rosennean) Complexity Paradigm