Paradigm Green: Artificial Intelligence approaches to evaluating the economic consequences of changing environmental viewpoints.

Alan J. Thomson

Pacific Forestry Centre, 506 West Burnside Road, Victoria, B. C. , Canada V8Z 1M5

ABSTRACT

Success of many projects depends on inclusion of socio-cultural viewpoints (paradigms) at the policy formulation and project planning stages. The present study proposes that semantic networks be used to represent relationships among individuals and groups to evaluate hypotheses about paradigm change in the social systems within which policies are enacted. Expert systems may then be used to determine parameter values in socio-ecological models formulated to evaluate the economic consequences of policies reflecting environmental management paradigms.

INTRODUCTION

Changing public attitudes towards environmental issues may override ecological considerations in determining resource management policy (Halleran, 1990). In managing all public policy issues, the societal background provides the context within which policy is ultimately determined (Barrows and Morris, 1989). Consideration of viewpoints is now regarded as essential in the planning process (Boshoff, 1989). However, difficulties in combining socio-cultural and bio-physical knowledge have resulted in a significant gap in knowledge necessary to formulate sound policies and programs (Lovelace, 1985a).

In this study, I first discuss the manner in which paradigms and policies interact. I then suggest ways in which AI applications, specifically semantic nets and expert systems, can help bridge that knowledge gap by guiding the setting of parameter values in economic and socio-ecological models. AI methods may also be used to evaluate the performance indicators of such models.

PARADIGMS

Peoples' views on their relationship with nature is one of the most important aspects of any strategy for human development (Colby, 1990). A five-stage evolution of such views, or paradigms, was described by Colby (1990), who provided a four-part definition of 'paradigm'.

"a criterion for choosing problems ... that can be
assumed to have solutions. Other problems are rejected
as metaphysical, as the concern of another discipline,
or sometimes as just too problematic to be worth the
time"
"the entire constellation of beliefs, values, techniques,
and so on shared by members of a given community, or one
element in that constellation, the concrete puzzle-
solutions which, employed as models or examples, can
replace explicit rules as a basis for the solution of
the remaining puzzles of normal science"
"not the same as shared rules; the existence of a
paradigm may not even imply that any full set of rules
exists (Kuhn 1970)"
"a worldview or mode of perception; a model around which
reality is organized (Berman 1981)"

The early, diametrically opposed paradigms of "frontier economics" and "deep ecology" are viewed as evolving through an "environmental protection" and "resource management" paradigm towards an "eco-development" paradigm in which "human activities ... (are) synergetic with ecosystem processes and services" (Colby 1990). Smith (1990) suggests that successful strategies for sustainable agriculture in the tropics will hinge on use of an eco-development type paradigm in which advances in scientific research are combined with traditional societies' knowledge of natural resources and their management.

THE MODEL VILLAGE

There are two basic frameworks or models of society for viewing the context within which policy is created: the 'consensus theory' resulting from a dominant set of shared values, and the 'conflict theory' based on domination and constraint (Barrows and Morris, 1989). For agrarian societies, these two models are reflected in the opposing views outlined in The moral economy of the peasant (Scott, 1976) and The Rational Peasant (Popkin, 1979).

In defining his intermediate viewpoint between the moral economic and rational economic viewpoints, Little (1988) discusses a model village with five features common among village societies in a wide variety of cultures. The model village is purely conceptual; there are no quantitative relationships of the type described by Miller (1982). The five features of Little's conceptual model are:

        1)   The model village is a relatively stable society, in that 
        its basic institutions, social relations, and ecological and 
        technical circumstances may be expected to change only very 
        gradually.
        2)   The model village is relatively isolated from outside 
        intervention and resources, whether economic or political.
        3)   There is a high level of information available to each 
        villager about the history and present activities of others.
        4)   The model village embodies a fairly high level of shared 
        values (familial, moral, religious, or political) and 
        organizations corresponding to these values (kinship 
        organizations, temple groups, political parties).
        5)   Social relations within the model village are 
        multi-stranded, in that villagers have a variety of different 
        sorts of social relationships.

AI offers a method for representing and evaluating such multi-stranded relationships. Complex interrelationships may be represented by systems of nodes and links known as semantic networks. The nodes represent objects, actions or events, and the links represent the relations among the nodes (Scown, 1985). An object can belong to more than one net. The different nets, known as "perspectives", represent the different contexts in which the object can be described. Semantic nets facilitate inheritance of values.

A village could be represented by a set of nodes representing individuals or groups of like-minded individuals. Each node can be defined in terms of membership of particular nets. For example, an individual could belong to the nets male, member_of_family_X, small_landholder, member_of_religious_group_Y, member_of_community_group_Z. Each individual has a set of attributes, which together make up the representation of his paradigm.

An individual's attributes that are relevant to the functioning of the net may be identified, such as a group leadership measure that could weight the final contribution of that net to the overall village decisionmaking process. The individual can inherit default values of an attribute over the network, with inheritance being possible over many levels. If family X is part of clan XX, then an individual could inherit clan attributes through his membership of the family net. Knowledge of the conditions under which default values are not inherited may be important for policy development.

Properties of a policy can be defined in terms of their relationship to the attributes of the various nets, and an the overall response of a village to these properties estimated through assumptions of the relative weightings of the networks in the village decisionmaking process. Such an approach would make explicit the value sharing and multi-stranded relationships of Little's (1988) conceptual model, in a form that permitted evaluation of alternative hypotheses.

SOCIO-ECOLOGICAL MODELS

Paradigms may not be known a priori. However, response of any individual or group to an issue is based on the paradigm of the individual or group towards that issue. Models can be formulated to include empirical specification of alternative hypotheses about the response, rather than about the underlying paradigm, and model performance can then be tested against empirical data. In this way, the paradigm of the individual or group can be inferred (Paarsch 1989).

Assumptions about group responses which reflect the underlying paradigms may be evaluated using the class of computer simulations known as socio-ecological models (Holling, 1978; Miller, 1982; Walters, 1986), which capture the interaction of human activities and the environment in the form of mathematical equations. The social component of such models includes economic aspects of the system; e.g. Miller (1982) gives the following equation for the emigration rate (EM) of salmon fishermen from a village as

        EM = UI*(c/AVFI**k)*d/log(RES)

where UI is the unemployment rate, AVFI is the average fishing income ($), and RES is the residential stability (average numbers of years of residence). Values of constants c, d and k are a reflection of the paradigm of the fishermen.

Simulation models written in procedural languages such as FORTRAN or PASCAL may include IF-THEN-ELSE structures or CASE structures to guide modification of parameter values. However, this approach does not make the knowledge base, as distinct from the data processing algorithm, as explicit as in a separate expert system. Algorithmic aspects of the system should be kept separate from the knowledge base.

Establishment of policies is the major route through which economic changes result, and Holling (1978) describes the manner in which socio-ecological models can be used to evaluate policy. As a preliminary step, all possible actions are listed, such as harvest trees, build a fish hatchery, spray insect pests or release x cubic feet of water from a reservoir. Policies are defined as the rules by which these actions are initiated, and state at what time or under what conditions actions are taken; e.g. cut all trees above a given age or release enough water from a reservoir to maintain a given minimum flow downstream. Indicators are defined as measures of system behavior in terms of meaningful and perceptible attributes; e.g. crop loss due to insects or costs of a program. Preferences are the trade-off rates between one indicator and another, and objectives are the desired goals in terms of indicators; e.g. the catch to sport fishermen to stay above 1965 levels, or the cost of management to grow at a rate less than the national budget. Decision structures are hierarchical: a goal at one level may be a policy at another (Holling 1978). Paradigms can influence all aspects of socio-ecological model development.

Policies can be coded into models by having specific sections of the program examine the current status of appropriate indicators in relation to policy-defined threshold values, and if the threshold is passed then a section of code simulating the policy-related action is invoked, changing the values of indicator values in ways defined by the equations in the program. The behavior of the indicators under different policy specifications can then be evaluated.

Socio-ecological models are often developed through an interdisciplinary workshop procedure (Holling, 1978). Rules are required to adjudicate Arguments of Principle, Arguments of Dogma and Arguments of Detail that arise during these workshops (Holling and Chambers, 1973). Arguments of principle are based on the premise that each individual has a unique weighting that he assigns to goals. Each individual should be allowed to weigh the information generated by the system in the way he feels best reflects his unique value system. Arguments of dogma emerge because of differences in background and training, and may be resolved by coupling each viewpoint with a policy which can be manipulated in runs of the model and contrasted with the results of policies reflecting other viewpoints. Arguments of detail may be resolved by running the model with conflicting hypotheses encoded separately. The rules and preferences discussed above may be dealt with explicitly using expert systems.

EXPERT SYSTEMS

The scientific paradigm shift from data through information towards knowledge processing is reflected in changes in the computing technology used for resource management (Grey 1990). Use of expert systems or knowledge-based systems is increasing in many areas of resource management (Kourtz, 1990; Starfield and Bleloch, 1983; Noble, 1987; Mills, 1987; McKinion and Lemmon, 1985; Schmoldt and Martin, 1986; Rauscher and Hacker, 1989; Davis and Clark, 1989). System output may be in natural language format, with the information content and format geared to the paradigm of the output recipient as perceived by the system developers (Thomson and Taylor 1990).

An expert system is a computer program that solves problems in a specific field at a level comparable to human experts (Feigenbaum 1979), and can deal with uncertain or incomplete information. The most widely applied type of knowledge representation is the rule-based form (Buchanan and Duda, 1983), in which knowledge is encoded as a set of condition/action links of the form :

             "IF (condition(s)) 
              THEN (consequence_set_1)
              ELSE (consequence_set_2)".

While the domain knowledge in an expert system may be captured in the above format, to form the knowledge base of the system, a separate part of the system, the inference engine, processes the information in the knowledge base. The inference engine attempts to confirm each condition in the 'IF' section, and if all conditions are confirmed, then all the assertions of the 'THEN' section are inferred to be true. If one or more conditions in the 'IF' section cannot be confirmed, then all the assertions of the 'ELSE' section are inferred to be true. A condition in the 'IF' section of one rule may be in the consequence set of another rule, so a number of rules may be examined, in a process known as 'chaining', in processing a particular rule. Probabilities or confidence levels may be assigned to the inferences of the consequence sets.

Note that the language of expert systems includes many of the terms, such as "rules", "rules-of-thumb", and "knowledge" which abound in studies such as Colby (1990) and Barrows and Morris (1989), indicating that the proposed expert systems methodology is compatible with the philosophical basis of the topic.

An example of a paradigm-related rule from Colfer et al. (1989) is

        RULE NUMBER: 11

          IF:   Ethnicity is Javanese transmigrant

        THEN:   Landowner is normally considered to be a male
                household head
         and    Land is viewed as very limited
         and    Rights to land are traditionally certified and
                private
         and    Women's agricultural labor is recognized as
                necessary but not preferred
         and    Ethnicity is symbolized by farming and small-
                scale female trade
         and    World view is hierarchical and authoritarian
         and    Domestic animals may include <2 cows and goats
                and chickens and 2 or more cattle
         and    Most crops planted probably require intensive
                management
         and    People value fertilizer and hoeing and cattle

Subsequent rules could then make inferences based on the knowledge, for example, that the world view of the subject group is hierarchical and authoritarian. It is evident, therefore, that expert systems offer a means of dealing with the qualitative and subjective aspects of paradigm-related hypotheses.

Expert systems and socio-ecological models can interact in a number of ways, reflecting the above issues. An expert system could be used to define paradigm-specific parameters for relationships in the model, or to determine actions under a given policy given the current state of the model. Note that as well as an expert system running the model, the model may itself run an expert system; e.g. the model could write a status file of all indicators for the current iteration then run an expert system to evaluate the indicators. Results of the evaluation could then be passed back to the model for processing in subsequent iterations. Again, the separation of the knowledge-based processing from the algorithmic processing is emphasized. Finally, an expert system could be used to evaluate the performance of a model embodying a particular philosophy, and may entail storage of output from simulation runs with all policies for comparison at this stage. This last type of expert system could be linked with utility analysis (Keeney and Raiffa, 1976).

DISCUSSION

In the absence of adequate studies, perceptions of paradigms may be inaccurate to varying degrees. The extent to which the paradigms of a group being studied are inferred correctly and the extent to which they are misunderstood has a critical effect on policy formulation, especially when it is governments perception of a paradigm (Dove, 1985).

The AI approaches discussed here would amply meet the requirements for systems models of human ecology identified by Lovelace (1985a), and the need for methodologies for evaluating socioeconomic factors influencing acceptability to rural people of social forestry interventions (Rao et al., 1985). Uses could include identification of potential difficulties where policy makers and planners view an issue as a problem while certain rural groups view it as an opportunity (Lovelace, 1985b). While the importance of socioeconomic aspects of resource management is well established in underdeveloped countries, there are also many potential applications in North America, such as in forest management programs on Indian lands, where the issue is not numbers or dollars and cents (although these are important), but rather the revitalization of a way of life (Hopwood, 1988).

For many issues, knowledge of the pertinent alternative paradigms may reside in a discipline such as anthropology or sociology where the ability or willingness of an individual to frame his knowledge in a mathematical or statistical formulation may be limited. Encapsulation of such knowledge in the English-like format of expert system rules is a method of knowledge representation more acceptable to such disciplines and has been used to capture anthropological knowledge of paradigms in an expert system for advising on agricultural research policy (Colfer et al. 1989), which fits well with part (1) of Colby's definition of 'paradigm'.

In other domains, paradigms of terrorist groups have been the subject of an expert system to combat international terrorism (Waterman and Jenkins, 1986), and paradigms of litigants have been captured in expert systems for legal decisionmaking (Waterman and Peterson 1986). The present study proposes an expert system -based methodology for evaluating the economic consequences of policies reflecting environmental management paradigms. In addition, use of another Artificial Intelligence (AI) approach, semantic nets, is proposed to define the social systems within which the policies are enacted.

Table 1 lists a number of questions indicated by Popkin (1979) as being answerable by the political economy approach and not by the moral economy theory. Answers to all these questions are implicit in village decisions such as the number of animals each family can pasture, or when fields are intermingled and unfenced, when planting, harvesting or grazing of each field should occur, or when fields are irrigated, apportionment of work on the irrigation system. Little (1988), however, argues that Popkin's negative case for the moral economy view is too sweeping, and proposes an intermediate model based on collective action by rational peasants.

Many of the issues of Table 1 are suitable topics for expert systems treatment. For example, development of a 'needy' scale is a categorization problem. Categories are regarded as central to a person's picture of reality and to knowledge representation and reasoning in AI (Thompson and Thompson, 1991). Fuzzy logic is an AI approach designed to handle such concepts that are relative and approximate (Mishkoff, 1986), and promotes constructs such as "moderately needy" or "very needy", descriptions that cannot be handled by formal logic, but which are analogous to human thinking.

REFERENCES

Barrows, D.S., and Morris, S. 1989.  Managing public policy 
     issues.  Long Range Planning 22: 66-73.

Berman, M. 1981.  "The Reenchantment of the World".   Cornell 
     University Press, Ithaca, NY.

Boshoff,H. 1989.  Testing plans against alternative futures.   
     Long Range Planning 22: 69-75.

Buchanan, B.G., and Duda, R.O. 1983.  Principles of rule-based 
     expert systems.  In: M.C. Yovits (Editor) "Advances in 
     Computers" Vol. 2.  Academic Press, New York, pp. 163-216.

Colby, M.E. 1990.  Environmental management in development.   
     The evolution of paradigms.  World Bank Discussion Papers 
     80, 39p.

Colfer, C.J.P., Yost, R., Agus, F., and Evensen, S.  1989.   
     Expert systems: A possible link from field work to policy 
     in farming systems.   AI Applications in Natural Resource 
     Management 3: 31-40.

Davis, J.R., and Clark, J.L. 1989.  A selective bibliography of 
     expert systems in natural resource management.  AI 
     Applications in Natural Resource Management 3(3): 1-18.

Dove, M.R. 1985.  Government perceptions of traditional social 
     forestry in Indonesia: the history, causes and implications 
     of state policy on swidden agriculture.   In: Rao, Y.S., 
     Vergara, N.T., and Lovelace, G.W. (editors) Community 
     forestry: socio-economic aspects.   FAO, Regional Office 
     for Asia and the Pacific, Bangkok.  pp. 173-197.

Feigenbaum, E.A. 1979.  Themes and case studies of knowledge 
     engineering.  In: D. Mitchie (Editor),"Expert systems in 
     the micro electronics age."  Edinburgh University Press, 
     Edinburgh, pp. 3-25.

Grey, D.C. 1990.  Knowledge management in forestry.   South 
     African Forestry Journal 153: 36-40.

Halleran, M. 1990.  Forest vegetation management and the 
     politics of the environment.   For. Chron. 66: 369-371.

Holling, C.S. (editor) 1978.  Adaptive environmental assessment 
     and management.  John Wiley & Sons.  377 pp.

Holling, C.S., and Chambers, A.D. 1973.  Resource science: the 
     nurture of an infant.  Bioscience 23: 13-20.

Hopwood, A. 1988.  The social and economic returns from 
     investments in forest management programs on Indian lands - 
     two case studies.  Canada-British Columbia Forest Research 
     and Development Agreement Report 049, Forestry Canada.  
     56pp.

Keeney, R.L., and Raiffa, H. 1976. Decisions with multiple 
     objectives.  John Wiley, New York.

Kourtz, P.  1990.   Artificial intelligence: a new tool for 
     forest management.  Can. J. For. Res. 20: 428-437.

Kuhn, T.S. 1970.  "The Structure of Scientific Revolutions".  
     2nd edition.   University of Chicago Press, Chicago.

Little, D. 1988.  Collective action and the traditional village.  
     Journal of Agricultural Ethics 1: 41-58.

Lovelace, G.W.  1985a.  Human ecology: a conceptual framework 
     for research and development in community and social 
     forestry.   In: Rao, Y.S., Vergara, N.T., and Lovelace, 
     G.W. (editors) Community forestry: socio-economic aspects.   
     FAO, Regional Office for Asia and the Pacific, Bangkok.  
     pp. 18-28.

Lovelace, G.W.  1985b.  Some thoughts on the relationship of 
     cultural values and  farmers' attitudes to rural 
     development programmes.   In: Rao, Y.S., Vergara, N.T., and 
     Lovelace, G.W. (editors) Community forestry: socio-economic 
     aspects.   FAO, Regional Office for Asia and the Pacific, 
     Bangkok.  pp. 29-39.

McKinion, J.M., and Lemmon, H.E. 1985.  Expert systems for 
     agriculture.  Computers and Electronics in Agriculture 1: 
     31-40.

Miller, P.C.  1982.  Simulation of socio-ecological impacts: 
     Modeling a fishing village.  Environmental Management 6: 
     123-144.

Mills, W.L. 1987.  Expert systems: applications in the forest 
     products industry.  Forest Products J. 37(9): 40-44.

Mishkoff, H.C. 1986.  Understanding Artificial Intelligence.   
     Texas Instruments, Inc.

Noble, I.R. 1987.  The role of expert systems in vegetation 
     science.  Vegetatio 69: 115-121.

Paarsch, H.J.  1989.  Deciding between the common and private 
     value paradigms in empirical models of auctions.   
     University of British Columbia, Forest Economics and Policy 
     Analysis Research Unit Working Paper 135.  29p.

Popkin, S.L. 1979.  The Rational Peasant.  University of 
     California, Berkeley.

Rao, Y.S., Vergara, N.T., and Lovelace, G.W. (editors).  1985.  
     Community forestry: socio-economic aspects.   FAO, Regional 
     Office for Asia and the Pacific, Bangkok.

Rauscher, H.M., and Hacker, R. 1989.  Overview of AI 
     applications in natural resource management.  Proceedings 
     IUFRO S4.02: "AI and growth models for forest management 
     decisions."  IUFRO, Vienna, Austria, September 18-22, 1989.  
     202-222.

Schmoldt, D.L., and Martin, G.L. 1986.  Expert systems in 
     forestry: utilizing information and expertise for decision 
     making.  Computers and Electronics in Agriculture 1: 
     233-250.

Scott, J. 1976.  The moral economy of the peasant.  Yale 
     University Press, New Haven.

Scown, S.J. 1985.  The artificial intelligence experience: an 
     introduction.   Digital Equipment Corporation.

Smith, N.J.H. 1990.  Strategies for sustainable agriculture in 
     the tropics.   Ecological Economics 2:  311-323.

Starfield, A.M., and Bleloch, A.L. 1983.  Expert systems: an 
     approach to problems in ecological management that are 
     difficult to quantify.  Journal of Environmental Management 
     16: 261-268.

Thompson, B, and Thompson, B.  1991.   Overturning the category 
     bucket.  Byte.  January 1991, p. 249-256.

Thomson, A.J., and Taylor, C.M.A. 1990.  An expert system for 
     diagnosis and treatment of nutrient deficiencies of Sitka 
     spruce in Great Britain.  AI Applications in Natural 
     Resource Management 4(1): 44-52

Walters, C. 1986.  Adaptive management of renewable resources.  
     Macmillan Publishing Company, New York.  374 pp.

Waterman, D.A., and Jenkins, B.M. 1986.  Developing expert 
     systems to combat international terrorism.  In: P. Klahr 
     and D.A. Waterman (eds.)  Expert systems.   The Rand 
     Corporation.  p 95-134.

Waterman, D.A., and Peterson, M.A. 1986.  Developing expert 
     systems to combat international terrorism.  In: P. Klahr 
     and D.A. Waterman (eds.)  Expert systems.   The Rand 
     Corporation.  p 135-185.