6 Models as resources [for power and simulation of mind]
Models are required in order to utilize information offers and exert and withstand influence. This is illustrated by contrasting the operations of two systems, one of which lacks model resources and thus is completely dependent upon external feedback. The other system may carry out internal trials through simulation, and strengthen its model power through comparisons between real and simulated outcomes. If a model-strong-actor and a model-weak-actor are coupled in an open information exchange system, the former may be expected to gradually increase his control of the other actor. Offers of information are useful only to the extent that there is model capacity for processing the information offered. Thus a successful transition in the name of democratization to a more open communication structure may freeze -- or even increase -- the influence gap.
Models as resources
Let us now make explicit the basis for the above kind of expectation. This requires reference to demonstrations of the essential function of models in regulation and interaction. Within a cybernetic frame of reference this has been led up to by Mackay (1967) and Amosov (1968), and has formally been shown by Conant and Ashby (1970). Within an interactionist frame, Mead (1934) stresses man's capacity for modelling the other as a prerequisite for significant interaction. His social-psychological theory easily lends itself to further development in systems-theoretical terms (Bråten 1972).
The importance of system component R
Consider a simple system, Beta, which seeks to realize goal states in its environment. Its available set of potential behaviour programs allows for a certain variety to be shown by its behaviour. But in order to cope with its environment, selected programs have to be tested, and possible replaced, on the basis of information fed back from the environment through external program execution. The operation of this feedback system requires, at least, equipment for obtaining data about state changes in the environment, and available criteria that may be sued in the comparisons between goal states and actual states.
Consider now another system, Alpha, which has all of Beta's resources, as well as one additional component, R. With the help of R, Alpha is able to -- prior to program execution in the external environment (Aussenwelt) -- to test the behaviour program in its own internal environment (Innenwelt). With reference to the external environment, and through operation upon data that give actual state values, R generates internally predictions about state changes as outcome of potential actions and comparable to goal criteria, and thus provides a means of correcting program selection prior to external execution.
We may now compare the two systems, Beta and Alpha. Like Beta, Alpha may obtain information feedback upon external program execution in order that goal state and actualized state may be compared as a basis for corrective actions. But unlike Beta, Alpha has the advantage of making action program tests internally through the use of R, and of gradually improving R's capacity for this kind of use on the basis of comparisons between outcomes of internal testes and outcomes of external executions.
Thus, R is used by Alpha as if R represents relations between Alpha's behaviour and environmental state changes, and thereby allows for internal experimentation which generates knowledge about potential action programs. In other words, R is used by Alpha as a model that allows for simulation of action outcomes prior to some external execution. Even if R initially may be an unfortunate model which generates misleading simulation outcomes, its possession still makes Alpha a more resourceful actor than the Beta system. While they both may adjust their behaviour on the basis of information obtained from their external environment, Alpha makes additional use of this information to adjust the model on the basis of comparisons between simulated and real behavioural outcomes.
As the predictive power of the model thereby is gradually improved, not only does Alpha develops a resource which aids in the internal testing of actions before external commitments, but also a tool for direction of search, discrimination, organization, retrieval and utilization of data.
In order to carry out and regulate its own behaviour, Beta must be able to receive and retrieve data from some short-term memory device. But even if Beta is offered access to a rich data store, this would not increase its capacity for realizing goal states more effectively, as long as the systems components are limited to those listed above. A data source is rich in potential information only to the extent that there is capacity for discriminating, structuring and processing the data offered in terms of utility, relevance and coherence. This kind of knowledge is provided by knowledge of relations between actions, events and state changes, i.e. by a model resource. As long as Beta is without such a resource, a generous data offer may have little utility value, but great nuisance value, threatening to break down the simple feedback system. Richness of information may cause critical overload.
Coupling of a model-weak and a model-strong actor
The Beta system outlined above is equipped with resources which allow for recognition of occurring goal states and of the lack of such occurrences. This means that the system is able to utilize certain data about environmental states. But without model resources, it cannot utilize these data in a manner that generates cumulative insight into relations between actions, state transitions and goal realizations. This lack prevents Beta from becoming an actor that may socially enter into symbolic interaction with some co-actor like Alpha. In order that such kind of [actor-coactor] relationship be established, model resources are required that allow for simulation of the co-actor's (re)action (Mead 1934, Bråten 1972).
Assume now that Beta' replaces Beta and is coupled to Alpha through a common environment, and that Beta' is equipped with model resources that are weak in comparison with the resources of Alpha. In this context, model strength concerns the model capacity to handle referent system variety. Thus, to say that a model resource, ri, is stronger than another model, rj, with reference to a set of possible events E, is the same as to say that each event in E which can be described or predicted by rj can be handled by ri in a similar manner, while there are events in E that can be handled by ri, but not by rj. Thus, the stronger model is able to handle a system which shows greater variety than a system which the weaker model can handle, when both referent systems are selected from the same referent domain. If they have different domains, they cannot be compared in terms of strength in the above sense.
What would happen if the model-weak Beta' and the model-strong Alpha establish an open information exchange system between them, in the sense that they give each other free access to any information possessed by any one of them?
It may be expected that the distance between the two actors in terms of model strength will increase, and not decrease, with the continuation of the open information exchange. At any point in time any datum offered by Beta' may be processed and utilized by Alpha, while data offered by Alpha may be processed and utilized only to the extent that Beta' has developed adequate model resources at that time. [...] As the open exchange continues, the model-weak Beta' -- although improving his model resources -- will gradually come under the control of the model-strong Alpha, whose rate of improvement in model capacity and data utilization will be higher. Ironically, the ultimate in control is reached if Beta' "succeeds" in adopting a model developed by Alpha at some previous stage. This gives Alpha the power of simulating even the simulations carried out by Beta'. Thus, while the intentions may be to decease the gap in model capacity a steadily increasing gap may be actualized.
1 From systems-theoretical notes prepared for the VII:th Scandinavian Sociological Congress at Helsingör, Denmark, June 1972, published as a part (pp. 98, 102-105) of the article "Model Monopoly and Communication: Systems theoretical notes on Democratization" Acta Sociologica 16, 2 (1973) pp. 98-107. Consistent with later usage, I have switched the use of the labels "alpha" and "beta", respectively, for a model-strong and a model-weak actor, and use R instead of F to denote internal representations.