Some reflections on anticipatory systems    

After reading the book of prof. Robert Rosen 
Anticipatory Systems. Phylosophy, Mathematical & Methodological Foundations 
Pergamon Press, 1985

   24 March 1989

1) System failures

A product is conceived to satisfy a certain basic need. Automobile satisfy the need of transport. However, together with the basic need there are many others more or less important needs (explicit or implicit) which the product also satisfy.

The success of a new product, or of a new model of an old product, depends therefore an how it is able to satisfy the complex whole of such needs.

Usually a new model has a lifetime. During the course of its life it might have been produced in millions of units. The experience from the markets are normally fed back into the manufacturing process so that the quality of the product increases during its lifetime as a function of the number of units produced. At a certain moment, however, the sales start to decline and the model is substituted by a new one.

This is clearly a case of system failure since the products - at the moment it is substituted - is still able to perform the mission it was designed to, and even better than the first unit produced.

In the case of automobile, the average lifetime in Europe of a new model is of approximately 10 years during which it will have gone through 2 or 3 major "restyling" and few more "face-lifting". The new model that replace the old one at the end of lifetime is usually completely different from the old one (at least if one considers it from the stand point of the production lines which are normally completely new and have required very large capital investments).

The metaphor of the anticipatory system can help in understanding such a behaviour.

The designer, when conceiving a new car, could take care explicitly of the basic needs, which goes into the list of technical specifications of the project. Often the characteristic of the project that respond to such basic needs are not changed with respect to the previous model. The designer, however, has to be able also to respond to others less explicit components of the global complex set of needs. Here is where one can distinguish (a posteriori, from the market success of the model) from excellent, good and less good designs.

The metaphor suggests that a designer should have in mind an anticipatory model (partly implicit) of the market and of the user, which includes also user's tastes and idiosyncrasies, fashion effects, prevailing culture, etc. He uses such a model when deciding on the design variables. At the end, the effect of the user's model is frozen in the design. The new model, when it first appears on the market, hopefully better responds to the user's needs and tastes than the older ones on the market. As the time pass, however, the needs or the tastes of the users change and the product less and less respond to them. To increase the product lifetime, "rejuvenation" is tried (face-lifting and restyling). When there is no more space for rejuvenating actions, the model is retired from the market and substituted with a new one. The death of a product model is therefore caused by the increased inability of the designer's anticipatory model frozen in the project to predict the user's needs and tastes.

2) Interacting qualities of the environments

It is interesting to note that in the USA (at least before the Japanese "invasion") a different approach was followed by car-manufacturers. The lifetime cycle of the new car model is planned to be much shorter (one year), but limiting the change in the "new year" model to very superficial aspects (requiring only few modifications on the production lines).

How the different approach in USA and Europe could be interpreted in term of phenotype, genome, environmental qualities?

The different approach could be explained, firstly, in term of an existing difference in the environmental qualities representing the needs to be satisfied by the product in Europe and in USA. The selection process therefore will operate on different parts of the total produce genome.

One could however question whether the different approach is really due to different environments or simply to different strategies of adaptation to environmental selection. In the latter case one should expect that the different subdivisions of the set of environmental qualities (the subset of those qualities which affect the product genome and the subset of all the others) will have different longer term effects in the two cases.

One could moreover think that it is possible to subdivide the set of environmental qualities in a more fine way: the subset of the qualities having a strong impact on the product (first order interaction), the subset of those having less strong impact (second order interaction), and so on, down to the subset of those qualities having negligible impact.

Different design strategy are then conceivable: one could decide to deal with the first subset only, or with the first and the second subsets, and so an. Accordingly, different genotype variables should be considered as affected. Correspondingly, different will be change in phenotype from one product model to the next.

With the first design strategy (considering only the first subset of environmental qualities) one could accept a shorter lifetime of the product, but with a less costly change in the phenotype. (The subsequent worries will be on how long one could defer the necessity to take into considerations also the - in the meantime built-up - effects of the second order qualities). With the choice of a different strategy (taking into consideration more than one subset of environmental qualities) one could hope to have a longer lifetime even if the change in the new product phenotype will be much more expensive.

The anticipatory mode built-in In the product in the two cases are different because they extend the predictions to different qualities of the environment. Keeping "dormant" the higher order qualities will depend from competition from other manufacturers. The Japan "invasion" in USA might have accelerated the need to extend the qualities to be included in the design strategy by USA manufacturers (however I had no occasion to examine recent market data in USA to find confirmation).

It will be interesting to know whether there are evidence of similar different cases of adaptation strategy from biology.

3) The constraints from genome architecture

The strategy that a manufacturer can adopt with respect to market has however to be compatible with internal constrains from technology. In fact a product like a car is very complex, as it is made of subsystems and components each one having different "lifetime". To give an idea, a new motor in Europe usually is designed to last at least 20-30 years (being produced in tens of millions of units before leaving place to a new motor). The basic reasons is the huge capital investments needed far manufacturing a new motor. (A second reason might be that the environment selection that operates on the whole car is less effective an the motor).

One could therefore consider that a designer is free to use only some of the genotype variables when designing a new product. The normal constrains with which the designer has to live are some time indicated referring to the existence of a technological regime. The innovation process during a prevailing regime is characterised by being more process than product oriented and in general being the summation of small steps.

However there are times when parts of the genome which was considered as fixed are changing and the designer have a larger freedom available to change the phenotype of the product. When such a time arrives one sees a larger innovation revolution in the product design. In the industrial history, "waves" of innovations have been detected related to different technological regimes.

The analogy with biology here is the recognised importance of constrains from genome architecture due to past evolutions and the "waves" of innovation corresponding with the emergence of a new biological type with respect to a new genus, family or species.

Assuming a common metaphor far biology and technology dynamics, the observation of the latter could provide interesting information because of its much shorter time scale. The amount of data collected (and evaluated referring in one way or the other to the biological evolution metaphor) is now impressive (see, e.g., C. Marchetti, Techn. Forecast. Soc. Change, 23, 3, 1983).

4) Refinement of the anticipatory system metaphor

It might therefore be interesting to try to refine the metaphor of the anticipatory system assuming that there is a "structure" not only in the environment (see the preceding section), but also in the genome.

A start in that direction is in the earlier Rosen paper on. "Dynamic similarity and the theory of biological transformations" (Bull. of Math. Biology, 40, 1978). There, a distinction is made between fundamental and derived variables in the equation of state of a given system.

The subdivision of the genomic variables into fundamental and derived ones, gives already some interesting idea far the technology case. So, to make just an example, the base materials used in a product could be considered as a fundamental genomic variable on which the product market selection does not intervene. According to the remarks In the Rosen paper, the change in the base material (e.g. composite materials in aircraft or in car) should push to change the product phenotype to maintain optimality of the design with respect to selection. Not always however this is what is done by the designer which tries instead to keep the change within the existing phenotype (material substitution in components). This might be the reason for the slowness in taking advantage of the potentiality of the microelectronics revolution in car design (limiting its use the "gadgets", without changing more drastically other product characteristics).

The importance of the above problem to better understanding the technological dynamics is such that I am very much interested to know any progress done in the refining of the anticipatory system metaphor.

On this respect it might be interesting to try to find correlations with ideas developed by W. Stegmüller (The Structure and Dynamics of Theory, Springer Verlag, 1976) in a quite different contest. In this work, Stegmüller contributes to solve the dispute between Popper and Khün an the concept of normal science. The dispute is on how a normal scientist can keep holding a theory even when "falsified" by experiments.

The solution proposed by Stegmüller is first to recognize that the concept of falsification for a formal theory is not a proper one: a formal theory has a field of truth which is intrinsic in its assiomatic definition and in its production rules. (Already here there might be a correspondence with the fact that in an anticipatory system - predicting without feedback the future of a system - it is improper to talk of errors when a difference emerge between prediction and reality).

Then Stegmüller distinguish in a formal theory of physics between a frame, a core and an expanded core. When comparing theory predictions with experiments the scientist is therefore provided with a certain degree of flexibility: he can change the expanded core of the theory without having to touch the core itself.

The work of Stegmüller is quite formal and it is difficult for me to appreciate the details of the reasoning. It might however be interesting to try to make a more detailed comparison with the model theory in the Rosen book to see if the structured flexibility of the Stegmüller theory concept could be applied to the modelling theory.

5) Definition of lifetime

I have some difficulties in understanding the concept of lifetime defined in term of divergence beyond a certain (arbitrarily chosen) value between the model predicted behaviour and the real system’s one. Could one instead, far a given system, fix intrinsically a threshold beyond which the approximation is no more acceptable so to cause the system failure?

As a matter of fact, in the case of a car new model, the lifetime is not fixed, but actually depends an the decision of the producer to stop or not production because of the reduced sales. In the biological case the process of death of a species - where there is not such a decidor - could perhaps be conceived as a fuzzy event more than a sudden catastrophe (with population decreasing an a more or less long period of time up to complete extinction).

For the individual organism however the concept is more difficult to grasp. One possible way out is the following. The environmental qualities are usually spanned continuously over certain ranges. Because of this, one could expect that the ecological niches of the species exploiting one basic environmental quality be either covering the entire range of the quality or, on the contrary, be very finely subdivided. These alternatives are however not the case both in biology and in the world of products. In fact, the market is usually subdivided into a discrete and limited number of separate segments. Some one has found a kind of rule of thumb valid far the separation of biological niches as well as far the products: the average distance of two successive niches is approximately equal to the square root of 2 (see the ratios between the radius of bicycle’s wheel).

A complex system which Is subject to environmental selection is therefore not only spanned with respect to the time variable, but also with respect to environmental variables. One should deduce from this that a complex system is realized with a certain built-in flexibility which permit to "survive" in an environment with-changing qualities up to a certain amount.
This flexibility might now define an intrinsic value of acceptability of difference between actual environment changes and the prediction of the anticipatory model built in the system. As the time pass, there is a drifting in the value of the environmental qualities. When such a drift brings the value of the environmental qualities beyond the acceptable range (due to the flexibility) the system is no more apt for survival.

At this point one could ask whether such built-in flexibility could not be enough to explain the time spanning behaviour of a complex system. What is the added value of the anticipatory model?

Possibly, the existing of such a feature acts an the system by causing the moving of the "centre of gravity" of the range of acceptable value according to the predicted average change of the environmental qualities. The lifetime is, also in this case, an abrupt event which happen when the drifted values of environmental qualities go beyond the range of acceptable values by the shifted capability range of the system.