- Email: [email protected]
Information gain by endo-observers: chances and limitations
BioSystems 46 (1998) 73 – 79
Information gain by endo-observers: chances and limitations Dieter Gernert Department of Economics, Technical Uni6ersity of Munich, Arcisstr. 21, D-80290, Mu¨nchen, Germany
Abstract Endo-observers can be described and classified on the basis of a novel two-dimensional scheme. One axis stands for the observer’s style of knowledge acquisition, whereas the other axis corresponds to the observer’s style of knowledge representation. In this way several problems concerning the relation between the endo-observer and the exo-system can be studied more systematically. Among other things, it is shown that there are two complementarity relations which restrict the observer’s capacity of information gain: a duality between structure and dynamics (an increased information gain on the structure of the endo-system implies a reduced information gain on its dynamics, and vice versa), and a duality between the predictive and the explanatory power of a model. Connections with the problem of hermeneutics are outlined. © 1998 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Endophysics; Endo-observer; Information theory; Implicit knowledge; Complementarity; Hermeneutics
1. A fundamental problem related to endo-observers Following a proposal by Matsuno, 1996 (Personal communication), we replace the clumsy term ‘internal observers’ by ‘endo-observers’; the latter expression also reminds of its endophysical context1). An endo-observer is sent into the interior of an insufficiently known system in order to analyze that system and to bring back a report
1 For endophysics see Ro¨ssler (1992), Atmanspacher and Dalenoort (1994).
afterwards. Possible types of endo-observers will be presented in Section 2. The specific problem to be discussed here is the following. Some authors are inclined to emphasize the ‘local’ character of endo-observers, whereas other ones rather like to accentuate their ‘holistic’ property. Of course, these two views do not form a pair of strict antitheses. In the sequel, it will be shown that the activities of endo-observers aiming at a gain of information can be studied within a comprehensive scheme, such that the alternatives ‘local versus global’ and ‘atomistic versus holistic’ are embedded in a natural manner.
0303-2647/98/$19.00 © 1998 Elsevier Science Ireland Ltd. All rights reserved. PII S0303-2647(97)00082-8
D. Gernert / BioSystems 46 (1998) 73–79
74
2. Various types of endo-observers Endo-observers may belong to one of the following types: human researchers or teams; technical devices; physical probes, like atoms, molecules, rays, waves etc. The reasons for considering a broad spectrum of possible types will become clear in the following discussion. It was shown earlier (Gernert, 1994) that, e.g., the methodological queries in cultural anthropology lead to fruitful analogies. The proposal to include atoms and molecules serving as probes in the study of physical systems is due to Matsuno (1985, 1989). For the sake of brevity, we equally speak about an ‘observer’ or an ‘observing system’ (also to avoid trouble with pronouns: he, she, it, or they?); the system under investigation will be called the ‘object system’.
3. Observation and measurement in a new perspective
3.1. Two ‘dimensions’ in the analysis of the obser6ation process First of all, measurement is regarded as a special case of observation, and hence only the latter term will be used. In a general framework comprehending all types endo-observers (Section 2), we must scrutinize the process in which the observer collects and stores information about the object system. This can be done on the basis of a two-dimensional coordinate system (Fig. 1). The activities and achievements of any observing system can be characterized by specifying both its style of knowledge acquisition and of knowledge representation; these terms will be explained in detail in the next sections. The shorter and established terms ‘knowledge acquisition/representation’ are used instead of composites with ‘information’; if it is accepted that ‘knowledge’ means information relevant to a recipient2, then the shorter terms are admissible. 2
The concept of pragmatic information is presupposed here
The horizontal axis stands for the observer’s style of knowledge acquisition, which can be strictly categorical, or strictly acategorical (Section 3.2), or can take on any intermediate position. The vertical axis represents the style of knowledge representation: the observer can come up with a global, holistic report, or with a mere collection of unprocessed original data at the ‘atomistic’ endpoint of the scale; any combination is possible here, too. Of course, the extreme cases have some heuristic relevance; but in many realistic situations the observing system will correspond to a point somewhere in the interior of the two-dimensional scheme.
3.2. Different styles of knowledge acquisition Knowledge acquisition can be done in a categorical manner: all possible techniques of information gain, all types of relevant knowledge, all questions to be asked are fixed before the beginning of the observation process. Examples are given by an interview in which only a previously designed questionnaire may be filled in, or a space probe that can only measure certain physical variables, determined by its equipment, together with image data also of a predefined type. In the acategorical style of knowledge acquisition no such restrictions are set in before. An example may be a visitor to a foreign country who accumulates knowledge simply by living there and, finally, may become an expert for that country (‘research by walking around’). As mentioned before, a mixture of both styles is rather common. Limitations to knowledge acquisition are imposed by the categorical style itself— a space probe just programmed to perform certain measurements will not register another space vehicle—but also by the prior knowledge, perception and attitudes of human observers. If an observing system is able to learn, it can proceed to a more suitable style of knowledge (Gernert, 1996). It may sound strange to speak about knowledge in the context of atoms or molecules, but this term may be re-substituted by ‘information’, with a loss only in style. It seems that ‘knowledge’ rather corresponds to the endo-view (‘We have knowledge about…’), whilst ‘information’ better fits to the exo-view (‘They have poor information on…’).
D. Gernert / BioSystems 46 (1998) 73–79
75
Fig. 1.
acquisition; if, e.g. a questionnaire is replaced by an open interview this means a shift towards the right-hand side in the two-dimensional scheme.
3.3. Different styles in knowledge representation Any new knowledge obtained in the observation process must be recorded in one or another way, e.g. by the diffraction of a ray, technical registrations, films, protocols, drawings, or in the memories of human observers. In one of the two extreme cases, the purely atomistic style of knowledge representation, only unprocessed data are recorded, as, e.g. the original data of a series of measurements, a poll, or a census. There is not any kind of subsequent compression, aggregation or interpretation. By way of contrast, the holistic style of knowledge representation cannot be characterized so easily. Here, knowledge exists in a ‘condensed’ form, on a level above all single data; such a knowledge at least claims to give a concise, yet valid representation of the object system. A further complication comes in by the fact that there are two different types of holistically represented knowledge:
(1) Explicit knowledge can be completely expressed in a communicable form, no matter whether this is done by texts, formulas, diagrams, etc. or combinations thereof. (2) Implicit knowledge3 can be communicated at most partially and incompletely; its existence can be proved only by the fact that certain observing systems can give reasonable answers to questions about the object system’s behaviour or have a capacity of prediction (Sections 3.4 and 4). There is a variety of techniques by which an explicit holistic representation can be derived from the original data, e.g. data compression and aggregation, abstraction, derivation of rules, but also the identification of prototypes or episodes which may be suited to draw impressive and elucidating pictures of the object system (learning and teaching by examples). Here, too, the techniques for obtaining such a representation of knowledge can be refined in the course of time. The acquisition of implicit knowledge requires sufficient time and a specific process of learning— 3 For the fundamentals of implicit knowledge see Polanyi (1966).
76
D. Gernert / BioSystems 46 (1998) 73–79
in the case of humans, keywords can be personal experience, empathy, and understanding (Section 3.4). The most important fact is that implicit knowledge can be embodied both in human experts and in artificial neural networks with their striking holistic capacities, including pattern recognition and prediction (Section 4.2).
In a strict endo-physical sense, the two-dimensional scheme proposed here is an artefact, which is carried from outside into the endo-system. It takes its plausibility only from the fact that our common concepts of observation and information gain lead to the two ‘dimensions’ in a natural manner; in a totally different cultural context this possibly will be questioned.
3.4. Methodological problems 3.5. The record/report antagonism A methodological problem that cannot be circumvented is the dichotomy between endo-observers and all those who want to analyse them from outside. In how far can an ‘externalist’ really appreciate what endo-observers are doing? In the cases of categorical knowledge acquisition or atomistic representation this is easy. The full strength of the problem comes up with implicit knowledge. How can we react when an endo-observer claims to have a surplus of implicit knowledge (which, by definition, cannot be completely communicated)? The existence of implicit knowledge can be proved by the long-term striking success of many individuals in ‘fuzzy’ professions, that is in such professions in which the decisive actions can only partially be guided by rules and where there is no guarantee for the outcome (as, e.g. physicians, researchers, economists, teachers, etc.). Such a success can be understood in terms of many years’ experience. In a single case, we can identify implicit knowledge by the fact that its carrier is able, e.g., to make successful predictions about the behaviour of the endo-system, at least on a statistically significant level. Nevertheless, there is a surplus or leftover of knowledge that cannot be communicated at all. As Matsuno (1996) pointed out, also diachronic processes must be envisaged: in a dialogue an elicitation of further knowledge may be possible. Hence, the leftover can possibly be diminished, but, on account of the incongruent representation schemes of both dialogue partners (Gernert, 1994), there will still be a ‘remaining remainder’ (aside, possibly, from rather exotic situations in which a perfect assimilation of the representation schemes is reached).
The totality of all new findings made by an endo-observer —no matter which styles of knowledge acquisition and representation are chosen in the individual case—can be denoted by the term ‘record’. On the other hand, it must be kept in mind that the observing system is expected to return and to deliver a report comprehending its new information. It is not at all a matter of course that a report supplied by an observer will be accepted or perceived correctly, e.g. that a society will fairly understand a report on a foreign culture delivered by a cultural anthropologist; nor is it a trivial task, e.g., to interpret molecular spectra. An earlier analysis (Gernert, 1994) can be summed up as follows: Whoever wants to understand a report on a ‘system totally different from common experience’ must be prepared to learn a novel terminology and entirely new schemes of classification and representation —it makes no sense to ask ‘immediate questions’ and to expect quick answers based on inadequate notions. There is still another aspect suggesting a ‘record/report antagonism’. Whenever an observer can anticipate already during the observation process that a report must be delivered later, then just this anticipation is likely to disturb the observation. The observer may be inclined to adapt both the strategy of observation (e.g. the selection of material) and the explicit report to the presumable interests, expectations, and capacities of understanding of the future addressees. Thus, the observer’s perception and the formulation of the report can be distorted. In extreme cases, human observers can become ‘citizens of two worlds’. Living long enough in a foreign country, rather familiar with its facts and
D. Gernert / BioSystems 46 (1998) 73–79
central ideas, they may be pressed into ‘doublethink’: they know that certain facts are real, some positions are justified, but they also know that they better should not speak thereabout at a wrong place.
77
4.2. Duality of predicti6e and explanatory power
observer entering the original data into such a network. In a first phase, the instruction phase (or learning phase), a neural network accepts a set of data (the training set), which, in the present context, describes an empirical input-output behaviour of the object system. In the subsequent recall phase (or application phase), the network is confronted to a new set of input data, and it can supply answers to questions for a hypothetical future system behaviour (questions of the type ‘what would happen if…’). Hence, the neural network has a capacity of prediction. In this case, the knowledge acquired about the object system is represented by a collection of internal states within the neural network; such models can be named implicit models. The style of knowledge representation is purely holistic. Although the success of neural networks, particularly their predictive power, may be rather suggestive, some reservations must be made. The crucial point here is the black-box character of neural networks (Tam, 1994), which do not open a way towards explicit understanding. Even if, in experimental settings, all internal parameters can be measured and registered, there is no chance for an interpretation of those data4; no rules, coupling coefficients, correlations, or other elements of our traditional thinking can be identified; no explanation is given for the way in which a certain result was derived. The conditions for the validity and time-invariance of such implicit models are unknown, and an adaptation to novel situations implies serious problems. On the other hand, models formulated by human authors regularly are compatible with traditional human thinking. They can be understood at least by experts of the discipline concerned, and such models will remain at least partially valid when the side conditions will change. These observations make it plausible that there is a duality between the explanatory power and the capacity of prediction: when two models of the same system are set up and compared under identical boundary conditions, then the explana-
An observing system can also be an artificial neural network equipped with sensors or a human
4 The interpretation of internal states within a neural network is discussed by Ossen (1990).
4. Complementarity relations as limitations to the information gain
4.1. Duality of structure and dynamics The possibilities of an information gain by an endo-observer are limited by (at least) two complementarily (or duality) relations, which are in some formal analogy with Heisenberg’s uncertainty principle. Both duality relations to be discussed here (see also Section 4.2) have in common that an improvement in the knowledge acquired about one aspect of the object system inevitably leads to a reduced knowledge about another aspect. The complementarity of structure and dynamics was first studied by Atmanspacher (1991) under the perspectives of quantum logic and temporal logic. ‘Structure’ is understood as the spatial appearance of a system in terms of the positions of its constituents, whereas ‘dynamics’ means the functional behaviour in the course of time. In the present context, the duality of structure and dynamics can be explained as follows. In the regular case, an observer has a limited ability to collect, process, and store information. There may be restrictions of time, but also limited equipment and storage capacity. A researcher in an unexplored region (with limited time and equipment) may either observe the behaviour of an unknown animal, or work like an anatomist, or choose a mixed strategy. Anyway, better knowledge about the behaviour is compensated by a reduced knowledge about the internal constitution, and vice versa. An overall breakthrough is possible only by an increase of resources.
78
D. Gernert / BioSystems 46 (1998) 73–79
tory power increases as the capacity of prediction is reduced, and vice versa. In order to improve one of both capacities without a loss in the other, it will be necessary to increase the overall effort. 5. Outlook: Towards a formalized theory of the hermeneutic process In recent years, the concept of hermeneutics (together with notions like ‘interpretation’ or ‘meaning’) attracts growing interest also in the context of natural science and general systems theory (Tsuda, 1984, 1991; Atmanspacher, 1993). Beyond a certain level of complexity the study of an object system by an observer will become a hermeneutic process. But, the other way around, also the analysis given above may contribute to an understanding of the hermeneutic process. The hermeneutic process is a recursive action. An observer has some previous information and a capacity to collect information about the object system. This information can alter the observer’s styles of knowledge acquisition and representation; in other words, the observer has won an improved capability both of extracting new information and of organizing the old and new findings. A loop can be formed consisting of three elements: (1) ability to win information; (2) new information; (3) adaptation of the styles of knowledge acquisition and/or representation. This loop can be run through recursively until a saturation takes place or the observer decides to stop (see also the distinction between synchronic and diachronic information made by Matsuno, 1996). If recursive processes, and particularly problems of convergence, are to be discussed, then a measure of distance (or dissimilarity) becomes indispensable. Questions will come in like: What is the distance between the observer’s present state and the previous one? Or between the present state and the final one after the end of the recursive process? Is a certain stragtegy of updating one’s working styles superior to another such strategy? A method to find distance functions on sets of complex objects or structures has already been published (Gernert, 1996). Along this line a formalized theory of the hermeneutic process is no more chanceless.
To sum up: It may still be legitimate to speak about local or holistic endo-observers, as far as these terms are used as short names for characteristic situations. In a more general framework, it is no more sufficient to characterize endo-observers by simply ascribing predicates like local, global, atomistic, or holistic. Rather, a two-dimensional scheme can be established such that any endo-observer can be represented by a position within this scheme; shifts in the strategies pursued by an observer correspond to motions within that scheme. If finally processes are considered in which the observer’s strategies are essentially revised in the course of time, the hermeneutic process can be represented not by a circle or a spiral in the plane, but by a spiral-like curve in three-dimensional space.
Acknowledgements The author wants to express his gratitude to Koichiro Matsuno, who set the theme and asked the decisive questions, and to two anonymous referees.
References Atmanspacher, H., 1991. Complementarity of structure and dynamics. In: Atmanspacher, H., Scheingraber, H. (Eds.), Information Dynamics. Plenum Press, New York, pp. 205 – 220. Atmanspacher, H., 1993. Die Vernunft der Metis. Metzler, Stuttgart. Atmanspacher, H., Dalenoort, G.J., 1994. In: Atmanspacher, H., Dalenoort, G.J. (Eds.), Inside Versus Outside. Springer, Berlin. Gernert, D., 1994. What can we Learn from Internal Observers?. In: Atmanspacher, H., Dalenoort G.J. (Eds.), pp. 121 – 133. Gernert, D., 1996. Pragmatic information as a unifying concept. In: Kornwachs, K., Jacoby, K. (Eds.), Information — New Questions to a Multidisciplinary Concept. Akademie – Verlag, Berlin, pp. 147 – 162. Matsuno, K., 1985. How can quantum mechanics of material evolution be possible?: symmetry and symmetry-breaking in protobiological evolution. BioSystems 17, 179 – 192. Matsuno, K., 1989. Protobiology: Physical Basis of Biology. CRC Press, Boca Raton, FL. Matsuno, K., 1996. Internalist stance and the physics of information. BioSystems 38, 111 – 118.
D. Gernert / BioSystems 46 (1998) 73–79 Ossen, A., 1990. Zur Modularisierung und Interpretierbarkeit Neuronaler Netze, Diss. TU Berlin. Polanyi, M., 1966. The Tacit Dimension. Doubleday, New York. Ro¨ssler, E.O., 1992. Endophysik. Merve, Berlin. Tam, K.Y., 1994. Neural networks for decision support. De-
.
79
cis. Support Syst. 11, 389 – 392. Tsuda, I., 1984. A hermeneutic process of the brain. Prog. Theor. Phys. 79, 241 – 259. Tsuda, I., 1991. Chaotic itinerancy as a dynamical basis of hermeneutics in brain and mind. World Futures 32, 167 – 184.
.
Information gain by endo-observers: chances and limitations Dieter Gernert Department of Economics, Technical Uni6ersity of Munich, Arcisstr. 21, D-80290, Mu¨nchen, Germany
Abstract Endo-observers can be described and classified on the basis of a novel two-dimensional scheme. One axis stands for the observer’s style of knowledge acquisition, whereas the other axis corresponds to the observer’s style of knowledge representation. In this way several problems concerning the relation between the endo-observer and the exo-system can be studied more systematically. Among other things, it is shown that there are two complementarity relations which restrict the observer’s capacity of information gain: a duality between structure and dynamics (an increased information gain on the structure of the endo-system implies a reduced information gain on its dynamics, and vice versa), and a duality between the predictive and the explanatory power of a model. Connections with the problem of hermeneutics are outlined. © 1998 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Endophysics; Endo-observer; Information theory; Implicit knowledge; Complementarity; Hermeneutics
1. A fundamental problem related to endo-observers Following a proposal by Matsuno, 1996 (Personal communication), we replace the clumsy term ‘internal observers’ by ‘endo-observers’; the latter expression also reminds of its endophysical context1). An endo-observer is sent into the interior of an insufficiently known system in order to analyze that system and to bring back a report
1 For endophysics see Ro¨ssler (1992), Atmanspacher and Dalenoort (1994).
afterwards. Possible types of endo-observers will be presented in Section 2. The specific problem to be discussed here is the following. Some authors are inclined to emphasize the ‘local’ character of endo-observers, whereas other ones rather like to accentuate their ‘holistic’ property. Of course, these two views do not form a pair of strict antitheses. In the sequel, it will be shown that the activities of endo-observers aiming at a gain of information can be studied within a comprehensive scheme, such that the alternatives ‘local versus global’ and ‘atomistic versus holistic’ are embedded in a natural manner.
0303-2647/98/$19.00 © 1998 Elsevier Science Ireland Ltd. All rights reserved. PII S0303-2647(97)00082-8
D. Gernert / BioSystems 46 (1998) 73–79
74
2. Various types of endo-observers Endo-observers may belong to one of the following types: human researchers or teams; technical devices; physical probes, like atoms, molecules, rays, waves etc. The reasons for considering a broad spectrum of possible types will become clear in the following discussion. It was shown earlier (Gernert, 1994) that, e.g., the methodological queries in cultural anthropology lead to fruitful analogies. The proposal to include atoms and molecules serving as probes in the study of physical systems is due to Matsuno (1985, 1989). For the sake of brevity, we equally speak about an ‘observer’ or an ‘observing system’ (also to avoid trouble with pronouns: he, she, it, or they?); the system under investigation will be called the ‘object system’.
3. Observation and measurement in a new perspective
3.1. Two ‘dimensions’ in the analysis of the obser6ation process First of all, measurement is regarded as a special case of observation, and hence only the latter term will be used. In a general framework comprehending all types endo-observers (Section 2), we must scrutinize the process in which the observer collects and stores information about the object system. This can be done on the basis of a two-dimensional coordinate system (Fig. 1). The activities and achievements of any observing system can be characterized by specifying both its style of knowledge acquisition and of knowledge representation; these terms will be explained in detail in the next sections. The shorter and established terms ‘knowledge acquisition/representation’ are used instead of composites with ‘information’; if it is accepted that ‘knowledge’ means information relevant to a recipient2, then the shorter terms are admissible. 2
The concept of pragmatic information is presupposed here
The horizontal axis stands for the observer’s style of knowledge acquisition, which can be strictly categorical, or strictly acategorical (Section 3.2), or can take on any intermediate position. The vertical axis represents the style of knowledge representation: the observer can come up with a global, holistic report, or with a mere collection of unprocessed original data at the ‘atomistic’ endpoint of the scale; any combination is possible here, too. Of course, the extreme cases have some heuristic relevance; but in many realistic situations the observing system will correspond to a point somewhere in the interior of the two-dimensional scheme.
3.2. Different styles of knowledge acquisition Knowledge acquisition can be done in a categorical manner: all possible techniques of information gain, all types of relevant knowledge, all questions to be asked are fixed before the beginning of the observation process. Examples are given by an interview in which only a previously designed questionnaire may be filled in, or a space probe that can only measure certain physical variables, determined by its equipment, together with image data also of a predefined type. In the acategorical style of knowledge acquisition no such restrictions are set in before. An example may be a visitor to a foreign country who accumulates knowledge simply by living there and, finally, may become an expert for that country (‘research by walking around’). As mentioned before, a mixture of both styles is rather common. Limitations to knowledge acquisition are imposed by the categorical style itself— a space probe just programmed to perform certain measurements will not register another space vehicle—but also by the prior knowledge, perception and attitudes of human observers. If an observing system is able to learn, it can proceed to a more suitable style of knowledge (Gernert, 1996). It may sound strange to speak about knowledge in the context of atoms or molecules, but this term may be re-substituted by ‘information’, with a loss only in style. It seems that ‘knowledge’ rather corresponds to the endo-view (‘We have knowledge about…’), whilst ‘information’ better fits to the exo-view (‘They have poor information on…’).
D. Gernert / BioSystems 46 (1998) 73–79
75
Fig. 1.
acquisition; if, e.g. a questionnaire is replaced by an open interview this means a shift towards the right-hand side in the two-dimensional scheme.
3.3. Different styles in knowledge representation Any new knowledge obtained in the observation process must be recorded in one or another way, e.g. by the diffraction of a ray, technical registrations, films, protocols, drawings, or in the memories of human observers. In one of the two extreme cases, the purely atomistic style of knowledge representation, only unprocessed data are recorded, as, e.g. the original data of a series of measurements, a poll, or a census. There is not any kind of subsequent compression, aggregation or interpretation. By way of contrast, the holistic style of knowledge representation cannot be characterized so easily. Here, knowledge exists in a ‘condensed’ form, on a level above all single data; such a knowledge at least claims to give a concise, yet valid representation of the object system. A further complication comes in by the fact that there are two different types of holistically represented knowledge:
(1) Explicit knowledge can be completely expressed in a communicable form, no matter whether this is done by texts, formulas, diagrams, etc. or combinations thereof. (2) Implicit knowledge3 can be communicated at most partially and incompletely; its existence can be proved only by the fact that certain observing systems can give reasonable answers to questions about the object system’s behaviour or have a capacity of prediction (Sections 3.4 and 4). There is a variety of techniques by which an explicit holistic representation can be derived from the original data, e.g. data compression and aggregation, abstraction, derivation of rules, but also the identification of prototypes or episodes which may be suited to draw impressive and elucidating pictures of the object system (learning and teaching by examples). Here, too, the techniques for obtaining such a representation of knowledge can be refined in the course of time. The acquisition of implicit knowledge requires sufficient time and a specific process of learning— 3 For the fundamentals of implicit knowledge see Polanyi (1966).
76
D. Gernert / BioSystems 46 (1998) 73–79
in the case of humans, keywords can be personal experience, empathy, and understanding (Section 3.4). The most important fact is that implicit knowledge can be embodied both in human experts and in artificial neural networks with their striking holistic capacities, including pattern recognition and prediction (Section 4.2).
In a strict endo-physical sense, the two-dimensional scheme proposed here is an artefact, which is carried from outside into the endo-system. It takes its plausibility only from the fact that our common concepts of observation and information gain lead to the two ‘dimensions’ in a natural manner; in a totally different cultural context this possibly will be questioned.
3.4. Methodological problems 3.5. The record/report antagonism A methodological problem that cannot be circumvented is the dichotomy between endo-observers and all those who want to analyse them from outside. In how far can an ‘externalist’ really appreciate what endo-observers are doing? In the cases of categorical knowledge acquisition or atomistic representation this is easy. The full strength of the problem comes up with implicit knowledge. How can we react when an endo-observer claims to have a surplus of implicit knowledge (which, by definition, cannot be completely communicated)? The existence of implicit knowledge can be proved by the long-term striking success of many individuals in ‘fuzzy’ professions, that is in such professions in which the decisive actions can only partially be guided by rules and where there is no guarantee for the outcome (as, e.g. physicians, researchers, economists, teachers, etc.). Such a success can be understood in terms of many years’ experience. In a single case, we can identify implicit knowledge by the fact that its carrier is able, e.g., to make successful predictions about the behaviour of the endo-system, at least on a statistically significant level. Nevertheless, there is a surplus or leftover of knowledge that cannot be communicated at all. As Matsuno (1996) pointed out, also diachronic processes must be envisaged: in a dialogue an elicitation of further knowledge may be possible. Hence, the leftover can possibly be diminished, but, on account of the incongruent representation schemes of both dialogue partners (Gernert, 1994), there will still be a ‘remaining remainder’ (aside, possibly, from rather exotic situations in which a perfect assimilation of the representation schemes is reached).
The totality of all new findings made by an endo-observer —no matter which styles of knowledge acquisition and representation are chosen in the individual case—can be denoted by the term ‘record’. On the other hand, it must be kept in mind that the observing system is expected to return and to deliver a report comprehending its new information. It is not at all a matter of course that a report supplied by an observer will be accepted or perceived correctly, e.g. that a society will fairly understand a report on a foreign culture delivered by a cultural anthropologist; nor is it a trivial task, e.g., to interpret molecular spectra. An earlier analysis (Gernert, 1994) can be summed up as follows: Whoever wants to understand a report on a ‘system totally different from common experience’ must be prepared to learn a novel terminology and entirely new schemes of classification and representation —it makes no sense to ask ‘immediate questions’ and to expect quick answers based on inadequate notions. There is still another aspect suggesting a ‘record/report antagonism’. Whenever an observer can anticipate already during the observation process that a report must be delivered later, then just this anticipation is likely to disturb the observation. The observer may be inclined to adapt both the strategy of observation (e.g. the selection of material) and the explicit report to the presumable interests, expectations, and capacities of understanding of the future addressees. Thus, the observer’s perception and the formulation of the report can be distorted. In extreme cases, human observers can become ‘citizens of two worlds’. Living long enough in a foreign country, rather familiar with its facts and
D. Gernert / BioSystems 46 (1998) 73–79
central ideas, they may be pressed into ‘doublethink’: they know that certain facts are real, some positions are justified, but they also know that they better should not speak thereabout at a wrong place.
77
4.2. Duality of predicti6e and explanatory power
observer entering the original data into such a network. In a first phase, the instruction phase (or learning phase), a neural network accepts a set of data (the training set), which, in the present context, describes an empirical input-output behaviour of the object system. In the subsequent recall phase (or application phase), the network is confronted to a new set of input data, and it can supply answers to questions for a hypothetical future system behaviour (questions of the type ‘what would happen if…’). Hence, the neural network has a capacity of prediction. In this case, the knowledge acquired about the object system is represented by a collection of internal states within the neural network; such models can be named implicit models. The style of knowledge representation is purely holistic. Although the success of neural networks, particularly their predictive power, may be rather suggestive, some reservations must be made. The crucial point here is the black-box character of neural networks (Tam, 1994), which do not open a way towards explicit understanding. Even if, in experimental settings, all internal parameters can be measured and registered, there is no chance for an interpretation of those data4; no rules, coupling coefficients, correlations, or other elements of our traditional thinking can be identified; no explanation is given for the way in which a certain result was derived. The conditions for the validity and time-invariance of such implicit models are unknown, and an adaptation to novel situations implies serious problems. On the other hand, models formulated by human authors regularly are compatible with traditional human thinking. They can be understood at least by experts of the discipline concerned, and such models will remain at least partially valid when the side conditions will change. These observations make it plausible that there is a duality between the explanatory power and the capacity of prediction: when two models of the same system are set up and compared under identical boundary conditions, then the explana-
An observing system can also be an artificial neural network equipped with sensors or a human
4 The interpretation of internal states within a neural network is discussed by Ossen (1990).
4. Complementarity relations as limitations to the information gain
4.1. Duality of structure and dynamics The possibilities of an information gain by an endo-observer are limited by (at least) two complementarily (or duality) relations, which are in some formal analogy with Heisenberg’s uncertainty principle. Both duality relations to be discussed here (see also Section 4.2) have in common that an improvement in the knowledge acquired about one aspect of the object system inevitably leads to a reduced knowledge about another aspect. The complementarity of structure and dynamics was first studied by Atmanspacher (1991) under the perspectives of quantum logic and temporal logic. ‘Structure’ is understood as the spatial appearance of a system in terms of the positions of its constituents, whereas ‘dynamics’ means the functional behaviour in the course of time. In the present context, the duality of structure and dynamics can be explained as follows. In the regular case, an observer has a limited ability to collect, process, and store information. There may be restrictions of time, but also limited equipment and storage capacity. A researcher in an unexplored region (with limited time and equipment) may either observe the behaviour of an unknown animal, or work like an anatomist, or choose a mixed strategy. Anyway, better knowledge about the behaviour is compensated by a reduced knowledge about the internal constitution, and vice versa. An overall breakthrough is possible only by an increase of resources.
78
D. Gernert / BioSystems 46 (1998) 73–79
tory power increases as the capacity of prediction is reduced, and vice versa. In order to improve one of both capacities without a loss in the other, it will be necessary to increase the overall effort. 5. Outlook: Towards a formalized theory of the hermeneutic process In recent years, the concept of hermeneutics (together with notions like ‘interpretation’ or ‘meaning’) attracts growing interest also in the context of natural science and general systems theory (Tsuda, 1984, 1991; Atmanspacher, 1993). Beyond a certain level of complexity the study of an object system by an observer will become a hermeneutic process. But, the other way around, also the analysis given above may contribute to an understanding of the hermeneutic process. The hermeneutic process is a recursive action. An observer has some previous information and a capacity to collect information about the object system. This information can alter the observer’s styles of knowledge acquisition and representation; in other words, the observer has won an improved capability both of extracting new information and of organizing the old and new findings. A loop can be formed consisting of three elements: (1) ability to win information; (2) new information; (3) adaptation of the styles of knowledge acquisition and/or representation. This loop can be run through recursively until a saturation takes place or the observer decides to stop (see also the distinction between synchronic and diachronic information made by Matsuno, 1996). If recursive processes, and particularly problems of convergence, are to be discussed, then a measure of distance (or dissimilarity) becomes indispensable. Questions will come in like: What is the distance between the observer’s present state and the previous one? Or between the present state and the final one after the end of the recursive process? Is a certain stragtegy of updating one’s working styles superior to another such strategy? A method to find distance functions on sets of complex objects or structures has already been published (Gernert, 1996). Along this line a formalized theory of the hermeneutic process is no more chanceless.
To sum up: It may still be legitimate to speak about local or holistic endo-observers, as far as these terms are used as short names for characteristic situations. In a more general framework, it is no more sufficient to characterize endo-observers by simply ascribing predicates like local, global, atomistic, or holistic. Rather, a two-dimensional scheme can be established such that any endo-observer can be represented by a position within this scheme; shifts in the strategies pursued by an observer correspond to motions within that scheme. If finally processes are considered in which the observer’s strategies are essentially revised in the course of time, the hermeneutic process can be represented not by a circle or a spiral in the plane, but by a spiral-like curve in three-dimensional space.
Acknowledgements The author wants to express his gratitude to Koichiro Matsuno, who set the theme and asked the decisive questions, and to two anonymous referees.
References Atmanspacher, H., 1991. Complementarity of structure and dynamics. In: Atmanspacher, H., Scheingraber, H. (Eds.), Information Dynamics. Plenum Press, New York, pp. 205 – 220. Atmanspacher, H., 1993. Die Vernunft der Metis. Metzler, Stuttgart. Atmanspacher, H., Dalenoort, G.J., 1994. In: Atmanspacher, H., Dalenoort, G.J. (Eds.), Inside Versus Outside. Springer, Berlin. Gernert, D., 1994. What can we Learn from Internal Observers?. In: Atmanspacher, H., Dalenoort G.J. (Eds.), pp. 121 – 133. Gernert, D., 1996. Pragmatic information as a unifying concept. In: Kornwachs, K., Jacoby, K. (Eds.), Information — New Questions to a Multidisciplinary Concept. Akademie – Verlag, Berlin, pp. 147 – 162. Matsuno, K., 1985. How can quantum mechanics of material evolution be possible?: symmetry and symmetry-breaking in protobiological evolution. BioSystems 17, 179 – 192. Matsuno, K., 1989. Protobiology: Physical Basis of Biology. CRC Press, Boca Raton, FL. Matsuno, K., 1996. Internalist stance and the physics of information. BioSystems 38, 111 – 118.
D. Gernert / BioSystems 46 (1998) 73–79 Ossen, A., 1990. Zur Modularisierung und Interpretierbarkeit Neuronaler Netze, Diss. TU Berlin. Polanyi, M., 1966. The Tacit Dimension. Doubleday, New York. Ro¨ssler, E.O., 1992. Endophysik. Merve, Berlin. Tam, K.Y., 1994. Neural networks for decision support. De-
.
79
cis. Support Syst. 11, 389 – 392. Tsuda, I., 1984. A hermeneutic process of the brain. Prog. Theor. Phys. 79, 241 – 259. Tsuda, I., 1991. Chaotic itinerancy as a dynamical basis of hermeneutics in brain and mind. World Futures 32, 167 – 184.
.