[Martin Taylor 2010.04.17.10.06]
[From Bruce Gregory (2010.04.17.0931 EDT)] to
[From Rick Marken (2010.04.16.2040)]
BG: One possibility that you seem to give short shrift is that there may be many questions about behavior that the control perspective does little to illuminate. PCT focusses on _how_ living systems achieve their goals. It lacks convincing models of _why_ individuals have particular goals in particular circumstances. Basically PCT seems to punt these questions by invoking upper levels in a hierarchy.
I've been watching these interchanges with some bemusement, seeing the usual positive feedback effects of conflict on both sides, but this is a comment that does deserve to be highlighted for its inaccuracy. Much of what follows comes from various readings of Bill P's writing. I can't tell how much, because a lot of it is what I thought before knowing of PCT, and over the last two decades I have read so much of Bill's writing that the two streams have become thoroughly intermingled in my mind. Maybe what follows is mostly rewriting Bill, maybe it isn't. Bill may recognize a lot of his thought in what follows, and may disagree violently with other aspects. At this point, I don't know which will be which!
PCT most definitely does NOT punt the "why" question to a higher level. There is an explicit, if purely physical, reasoning behind the "why". (The Godly might find a further reason behind the physical, but I find that to be unnecessary). The reasoning is simple: anything about a structure that increases the likelihood its continuing to exist will be more likely to be found in patterns that do exist at any moment in time than are things that decrease the likelihood of the pattern continuing to exist. It's called "survival of the fittest" or "Natural Selection".
In PCT, there is a mechanism for sustaining those "high fitness" structures. It's called "reorganization". Any living entity has many variables, typically biochemical ones, that must be maintained within reasonable bounds of some optimum or the entity dies. If it has died before propagating a reasonable facsimile of itself, its internal structures will not be found in future living entities. So, these biochemical "intrinsic" variables must have some mechanism to support their maintenance, and that mechanism must act on the world outside the entity as well as on its internal world network of chemical reactions.
Why must the mechanism act on the external world? Because every reaction involves an increase of entropy, and the entity needs an input of low entropy and a sink of higher entropy (I'm talking physics, here, so these are simply abstractions, not observations, and therefore are unconvincing to Bill P. They might fit with Bruce G's perception of the world, though). That is equivalent to saying there must be an energy flow through the entity. The energy flow must be regulated, as too much too quickly could be as damaging as too little too late. So any living entity must act on the outer world at least so far as to influence how and when the entity gets an appropriate ration of energy.
If the entity must act on the world, it could do so randomly, but in a complex world, almost all random interactions will have a negligible effect on the entity, and as many would be damaging as would be useful. So an entity that contains structures that would enhance the likelihood that the entity gets its energy input in useful amounts at the right time would be more likely to survive long enough to propagate than would a similar entity with no such structure.
Skipping over a few steps that ought to be easy for the reader to fill in, a control loop could be a "structure that would enhance the likelihood that the entity gets its energy input in useful amounts at the right time". There may be other such structures, but I don't know of them. A control loop involves sensing some state of the environment and influencing that state, and the state must be one for which the value affects either the import of energy at low entropy (food, sunlight ...) or the export of energy at high entropy (waste). It is a Perceptual Control Loop, and its reference level is determined in some unspecified way by the existing states of what PCT knows as "intrinsic variables" -- all those biochemical variables mentioned a couple of paragraphs above.
Why would the "perceptual" control loop not be sensing and acting directly upon some state of the biochemical "intrinsic" variables? How would such a control system act? It would have to act upon the external environment, and act in a way to affect the intrinsic variable in some reasonably consistent direction. But the external environment changes, and the actions that would provide a consistent effect on the intrinsic variable might be quite different from one moment to the next.
So, something must sense the external environment and act so that the effects on the intrinsic variable is more likely to drive the intrinsic variable toward its optimum value than to drive it away from its optimum. That "something" is a perceptual control system. But the control system that acts on the external environment is NOT controlling the intrinsic variable, because that's not what it is sensing. So its effects on the intrinsic variable are side-effects of its controlling some sensed state of the environment. The Perceptual Control system is partially decoupled from the intrinsic variable system.
If the Perceptual Control system is decoupled from the direct control of intrinsic variables, how can it serve to control the intrinsic variables that have to be controlled if the entity is to survive? That's where "reorganization" comes in (as does evolutionary development).
Consider the earliest, simplest entities that could be considered "alive" (whatever that means). They presumably had a much simpler biochemical network than do most entities living now, but they had the same need to act on the environment in order to keep their biochemical intrinsic variables within bounds. Some might act to control an internal representation of one environmental variable, some another. Of the myriads of possibilities, control of some "perceptions" would serve to stabilize the intrinsic variables better than would control of other possibilities, and at different times, some values of the controlled perceptions of the environmental state would be better than others. Entities that controlled "useful perceptions" at "helpful values" would, by the side effects of their perceptual control actions, tend to stabilize the intrinsic variables.
So I argue that Perceptual Control is an intrinsic feature of life, and that its existence is the only defining feature of life. If an entity can be shown to have perceptual control, it is thereby shown to be living. If it does not, its continued existence is due either to material structural strength (an iron bar, a rock, the skeleton of a skyscraper) or to self-organized structures in an energy flow (a set of which living things are a subset), such as a whirlpool in a river. Whirlpools often spawn "child" whirlpools in the form of downstream eddies, but cannot be reasonably called alive. They control nothing.
Notice some of what is stated above. I have described a perceptual control structure with a reference value. What perception the control structure controls is not predetermined, but depends on the nature of the environment. Only a control action that affects the intrinsic variables will long survive evolution or reorganization, and since the control system is controlling a representation of the environment and NOT a representation of an intrinsic variable, the only way its actions can influence an intrinsic variable is through side effects of its actions on the environment. It is blind to its effects on the intrinsic variables.
Do the intrinsic variables provide the reference values for the perceptual control structures whose side effects influence them? It seems that they must but they cannot. They must because it is only when appropriate perceptions are controlled at appropriate values (appropriate for the environmental conditions) that the side effects will influence the intrinsic variables toward their optimum values. They cannot, because the side effects of controlling any one perception to a particular reference value may be vastly different in different environmental circumstances. Furthermore, it may be that the same side effects result from the control actions of controlling a variety of different perceptions of the environment. It is also likely that the side-effects of any one perceptual control will influence the values of several intrinsic variables.
How is this paradox of "must but cannot" resolved in living systems? The answer was given above: Evolution for conditions in which the environmental feedback paths are stable enough to allow the side-effects of controlling certain perceptions to do the right thing over time scales of several generations, reorganization for conditions where this is false. Perceptual control takes care of rapid environmental state fluctuation (but not too rapid) in either case.
At this point, we are beginning to get to Bruce's question of "_why_ individuals have particular goals in particular circumstances." The base answer is that to survive, individuals (of any living species) must be controlling some "perceptual" representation of a state of the environment, and probably many more than one such representation. At this level of answer, the answer is "To survive". (To ward off a possible criticism that this would eliminate self-sacrifice for the good of the group, I suppose it must be pointed out that what is to survive is the structure, and that may not always be best served by the survival of the individual entity. Martyrdom is a very powerful way for the conceptual structure of a religion to survive, for example).
The generic answer "To survive" does not cover "particular goals in particular circumstances", but I think the form of the more complete answer should be clear. Any one individual has the perceptual control structures that have been developed by the individual's evolutionary history. Those structures almost certainly determine what sensors are available to evaluate states of the environment. In some species, they may determine quite a lot about how those sensor data streams are combined to form controllable perceptions, whereas in other species evolution may provide only a skeleton of the structure of control within the individual, "reorganization" providing the rest. The latter species are likely to be more adaptable to varying environments than the former.
Different members of an adaptable species may have the same set of intrinsic variables, but are likely to differ considerably in some aspects of what specific perceptual functions they develop and of those, which perceptions they control, and of those, what (possibly changeable) reference values serve to maintain their intrinsic variables near their optimum values. Finally, individuals differ greatly in how effectively and for how long they actually manage to keep their intrinsic variables within non-lethal ranges of values. These differences in part depend on the random (i.e. unpredictable by the perceptual control structure) influences of the environment that are beyond the power of the structure to control (the states of teh environment), and depend also on how well evolution and reorganization have adjusted the perceptions to be controlled (according to the environmental feedback paths available -- the environmental affordances) and the top-level reference values (those with no input signals) for some of the controlled perceptions.
According to the view expounded here (I hope it is pretty close to what Bill P has been expounding, though I take responsibilities for errors in the parts where I diverge from his views), different people will have similar structures of control systems, but will control very different top-level perceptions, and those who control similar top-level perceptions may well control them at very different reference values. In any control system at any level, the answer to "why control this at that value" is physical. The reference signal input determines the reference value at all but the top level, and the perceptual input function determines what it is that is controlled at this level.
While thinking of humans in terms of their goals can very helpful, the mechanisms by which they attempt to achieve those goals is often less so. Controlling a perception of having more money may well describe a Wall Street Banker's motivation, but does not illuminate the workings of a collateralized debt obligation in any way that I can see. But perhaps I am just missing something both obvious and important.
You are talking here about the environmental affordances available for perceptual control. Those are features of the environment, and, I think, worthy of another essay, since the question can be illustrated metaphorically: "If one wants to cross a narrow but deep and fast stream, and there is a strong plank lying nearby, why would one place the plank across the stream?"
Think about that one in the context of environmental affordances. One control system controls for the position of the plank, but what is changed is the environmental affordances available to a quite independent control system controlling for the person's location. The change allows the person to walk where walking had not been possible. It's not enough to say that plank placement control is a lower-level control system under the perceived location control system. There are quite a few ramifications, including touching on the question of the internal observer. I don't think this is the place or time to go into them, but they are worth thinking about before we get into a serious discussion of them.
Martin