From Parallelism to Serialization

Riccardo Boero, Marco Castellani, Flaminio Squazzoni

GROWING BEHAVIORAL ATTITUDES, REFLEXIVE TYPIFICATION OF SOCIAL CONTEXTS AND TECHNOLOGICAL CHANGE IN A COMPUTATIONAL INDUSTRIAL DISTRICT PROTOTYPE

DSS PAPERS SOC 6-02

To be presented at the 8th International Conference of the Society for Computational Economics on “Computing in Economics and Finance”, Aix-en-Provence, France, 27-29th June 2002

1. From districts to “districtualized” firms: behavioral attitudes and social reflexivity of localized firms ...... Pag. 07

2. How ID prototype works ...... 10

3. How ID agent works ...... 17

3.1 Cognitive step 1: from information to "rough indexes" ...... 19

3.2 Cognitive step 2 : from "rough indexex" to "macro indexes" .... 22

3.3 Cognitive step 3 : indexes evaluation ...... 24

3.3.1 Behavioral attitude states ...... 25

3.3.2 Behavioral attitude morphogenetics shifts ...... 27

3.3.3 Behavioral attitude deconstruction shifts ...... 29

3.4 Cognitive step 3 : actions ...... 31

4. Indicators and Prototype Settings ...... 42

5. Analysis of outcome and emerging dynamics ...... 43

6. Conclusion: how agent-based computational models can put down stable roots in social sciences? ...... 49

References ...... 52 Abstract

Industrial districts are complex inter-organizational systems characterized by an evolutionary network of interactions amongst heterogeneous, localized, functionally integrated and complementary firms. By creating an industrial district computational prototype, that is to say simulating an archetype of industrial district through an agent- based computational model, we explore how industrial district dynamics can be conceived as byproducts of cognitive identity and social identification processes undertaken by district firms. Rather than study the district just as the effect of emerging properties of interacting firms, we try to study how firms develop over time more or less districualized behavioral attitudes based on reflexive typification of social contexts upon which firms experience and towards which they develop more or less deep “identification”. The question is: have such cognitive processes a great impact on technological learning and economic performance of firms over time?

Key Words: Agent-Based Computational Models, Industrial Districts, Cognitive Architecture, Behavioral Attitudes, Technological Change. Are industrial districts nothing but the outcome of the “complexity effect”? Or do you need to think more deeply about “district agents” and how they cognitively reflect upon such “complexity effect”? and again, is it better to speak about “industrial districts”, or to speak about dynamics of firms’ “districtualization”? Are districts nothing but the product of some emergent properties due to externalities-driven economies of proximity- related firms agglomeration, or are they as well what can be trivially called a “state of agent’s mind”? This paper rises from a first attempt to explore such kind of questions. To do this, we have created an industrial district agent-based computational prototype, that is to say we have reproduced a “theoretical idealtype” of district into a computer, simulating some relevant processes. The aim is to explore how behavioral attitudes of the industrial district firms evolve over time, how such evolution is both affected and sustained by a continuos dynamic of expansion and contraction of the social context experienced by firms, and how this has relevant impact on the technological learning undertaken by firms. The paper is organized as follows:

- the first section shows some theoretical standpoints on districts that pushed us towards the idea of exploring district phenomena using agent-based computational techniques; rather than assuming right from the start that district firms have a prototypical and homogeneous behavioral attitude, as the traditional literature on district does (i.e., automatic and natural commitment, cooperation, and trust amongst firms), we study the evolution of different behavioral attitudes, in a broad

Growing Behavioral Attitudes… 5 sense more or less “districtualized”, within firms’ paths of “day-to-day” experience and action; - the second section shows how the industrial district (ID) prototype works, from the point of view of its “structural properties”, that is to say classes of firms, division of labor amongst them, spatial localization of firms, evolution of the technology and market environment, and some different technology and market challenges upon which firms need to be able of learning; - the third section shows how ID cognitive agents work and by which kind of computational “building blocks” they are composed; they refer to what we call “information/action loop”, which shows a general framework of computational cognitive processes undertaken by district firms; - the fourth section shows some indicators we used to control the most relevant processes emerging by the prototype; - the fifth section shows the analysis of simulation outcomes, with a focus on the most relevant emerging dynamics, above all, the relation amongst changing behavioral attitudes, technological learning, and economic performance of firms over time; - finally, the sixth section shows some conclusion about how agent-based computational models can put down more stable and rich roots in social sciences and towards which kind of direction we shall develop the district prototype.

6 Growing Behavioral Attitudes… 1. From districts to “districtualized” firms: behavioral attitudes and social reflexivity of localized firms.

In recent years, the literature on industrial districts has moved several steps towards the analysis of the firm-based “cognitive mechanisms” of “individual identity” and “social (group) identification”, as a way of studying more deeply the roots of the individual action within localized social contexts. For instance, rather than taking a “system as a whole” perspective and assuming right from the start a natural, homogenous and prototypical behavioral attitude of the district firms, such analyses have focused on how “district members perceive and evaluate the district and how these perceptions affect the actual behaviors adopted by district members”, and, therefore, on how “identification” of firms towards their social contexts of experience is “an extreme form of relational modelling allowing the individual to define his/her identity in relation to the characteristics of perceived social groups” (i.e., see: Sammarra A. and Biggiero L., 2001). By taking a perspective oriented to topics such as the capacity of ID agents to develop forms of “identity”, using capabilities of “social reflexivity” and producing processes of “institutional identification”, the analysis starts to shift quietly its attention from the district as a “systemic mechanism” with specific geo-spatial boundaries sketched out by the inter- firm division of labor, to the dynamic of “districtualization” and even of “dis-districtualization” developed continuously by firms with respect to their social context of practices (i.e., see: Becattini G., 2002). From “district” as a well-defined system to the “districtualization” as a “state of mind” of locally

Growing Behavioral Attitudes… 7 interacting heterogeneous agents, this is the theoretical shift of the recent literature on district. Clearly, such analytical perspectives are nothing but a dichotomy in search of integration and synthesis. This means that we need to study the relation amongst: a) the aggregate dynamics emerging from the “bottom up” by the interaction amongst many heterogeneous localized firms, that is to say what complex system theorists call “the complexity effect” (Holland J.H., 1995); heterogeneous firms engaged in recurring patterns of interaction, but embedded within specific local context, give rise to what can be called an “aggregate composite system” (the district) (Auyang S. Y., 1998), on which they have different visions and about which none of them have neither property, nor possibility of completely control and management (i.e., see: Lane D., 2001); b) the cognitive capability of firms to develop “reflexivity” upon the fundamental characteristics of the context they experience over time; firms, even if localized, develop more or less keen antennas oriented to environment and contexts within which they move (i.e., see the cognitive and evolutionary approach suggested by Belussi F. and Gottardi G., 2000); over time, such monitoring capabilities affect their behavioral attitudes creating a continuos tension between “cognitive identification” (“districtualization” of behavior) and “cognitive distance” (“dis- districtualization” of behavior); such cognitive continuum needs to be conceived as different behavioral states of the cognitive evolutionary paths of firms, rather than a static “anthropological prototype” of the district firm, or a byproduct of the district mechanism, avoiding what

8 Growing Behavioral Attitudes… sociologists call the paradox of the over-socialized conception of agency (i.e., see: Uzzi B., 1996, 1997; Staber U., 2001); c) with this respect, the complexity of the district processes stands in the “circular loop” between emergent properties of the interaction context and evolutionary reflexivity of agents upon it.

The standpoint of our paper is to study how behavioral attitudes of firms change over time, how they are driven by the firms’ capabilities of monitoring the characteristics of their relational contexts and operational environments; how firms, as “bounded rationality” cognitive agents, develop a continuos theoretical “typification” of contexts and environments, trying an understanding of what they have done and what they need to do, and which kinds of effects on technological leaning and economic adaptation such “cognitive typification” has, too.

Growing Behavioral Attitudes… 9 2. How ID prototype works1

In our intentions, speaking about a district prototype it means to translate a “general” and “abstract” representation of a district “archetype” into an agent-based computational architecture. According to the well known “prototype principle” suggested by Douglass R. Hofstadter (1979), almost twenty-five years ago, an observer can classify and “typify” an observed phenomenon, just synthesizing into a “hierarchy of generality” its spatial and temporal “collocation”, “specificity” and “manifestation”. Generality means the need to identify a class of phenomena and fundamental underlying mechanisms and processes, in order to be able to explain something. By using the term “computational prototype”, we reinforce the evidence both that our prototype is not a modeling of a “case- district”, and that our computational architecture has been thought as a framework to be computationally developed, both by applying it to case- study-based modeling and by further building blocks refining. We start from a very broad and accepted definition of what is an industrial district archetype: district is a decentralized complex system characterized by an evolutionary network of interactions amongst heterogeneous, localized, functionally integrated and complementary firms. Firms are embedded within an integrated geographical area, they produce goods on market according to a division of labor mechanism, they have more or less rich proximity relation, and they move within specific

1 The ID computational prototype has been created using Swarm libraries and Java programming language. Swarm is a toolkit for agent-based computational simulation developed by Santa Fe Institute (see:www.swarm.org) and used by a more and more growing community of social scientists. For descriptions and applications of Swarm to economic phenomena, see: Terna P. (1998), Luna F. and Stefansson B. (2001), and Luna F. and Perrone A. (2001).

10 Growing Behavioral Attitudes… technology and market environment. Such district structural features are not as pieces of a “social totality”, but rather as byproducts of what Giddens should call “structuring properties” of the district (Giddens A., 1984). As it is stressed by Giddens, “structure is regarded as rules and resources” not “brought into being by social actors, but continually recreated by them via the very means whereby they express themselves as actions”, in terms of recursively oriented “social practices ordered in time and space”. We think about “prototype district structural features”, such as for instance mechanisms of division (specialization), coordination and integration of labor (production chains), in a sense like this, namely as recursive time- space processes that “exhibit structuring properties” developed and reproduced by agents which develop and reproduce at the same time “conditions that make their activities possible”. Our first operation has been to translate such distinctive features in specific computational “building blocks”, designing an agent-based prototypic architecture. Right from the start, we assume that prototype agents are firms. They are 400, divided in two different classes, final firms, having functions of organizing production and selling goods on market, and sub contracted firms, having specialized functions related to the whole production process. The class of the sub contracted firms is further divided into three sub- classes, sub firms A, B and C. In order to produce a good on market, firms interact give rising to production chains. Here, final firms have a focal and innovative role because of their interstitial position at the edge of market and district (i.e., see: Albino V., Garavelli A. C., Schiuma G., 1999; Belussi F. and Gottardi G., 2000; Boari C. and Lipparini A, 1999; Lazerson M. H. and Lorenzoni G., 1999).

Growing Behavioral Attitudes… 11 We assume that a production chain must be composed by: 1 final firm + 1 sub firms A + 1 sub firm B + 1 sub firms C. Firm is located within an environment populated by other firms, namely the district, with given spatial neighborhood positions (for details, see: Boero R. and Squazzoni F., 2001). Firms have three basic features: technology (input), organizational asset (throughput), and economic performance (output). The relation amongst such three basic features is shown in figure 1, which represents the evolution of technology and market environment towards which firms need to adapt. Firms need to undergo 2000 simulation/production cycles, during which they face three phases of technological continuity and two phases of technological discontinuity. In short, over time market causes two technology breaking off (cycle 500 and 1000). Market is conceived as an “institution” collecting and distributing information about performance and technology evolution for firms. Firms need to absorb technology and to learn the way by which adapt their organizational assets trying to reach the fixed best technological practice level. We assume that technology (T1, T2, T3) implies an investment of internal organizational factors. Technology is composed by a set of four numbers (i.e., 0, 3, 7, 2). Every number can be viewed as an organizational factor, such as labor, physical capital, human capital, and information and communication internal architecture. Firms start the simulation from a combination as follows: T1 (-1, -1, -1, -1), that is to say a situation of complete ignorance about technology factors. The best technological practice level of T1 is randomly fixed at the start of the simulation, and that of T2 and T3 is randomly fixed over time. This implies that firms can improve their technological effectiveness, both decreasing or

12 Growing Behavioral Attitudes… increasing number/factors. Firms do not know the “best technological practice level” and can change number/factors just by turns. Therefore, we assume that experimental learning of firms is characterized by “path dependence”, that is to say technological innovation of firms is affected by the technological position they have. In short, when firms take technological jumps from T1 to T2, or from T2 to T3, they start to explore the new combination of number/factors by their previous combination (i.e., previous combination: T1 3, 4, 7, 8 /jump from T1 to T2/initial combination: T2 3, 4, 7, 8). According to the effectiveness of their organizational assets, in terms of distance/nearness of their combination of number/factors with respect to the “best technological practice level”, firms have specific costs and reach specific performance levels, as it is shown in Matrix 3. To adapt step-by-step their organizational asset, firms have two strategies of experimental exploration within the state of technological possibilities: radical innovation (with a possibility fixed on 80% to obtain a new value, namely a new number/factor); or imitation by exploitation of information coming from neighborhood (firms is able to look into the combination of factors of neighboring firms, to compare specific number/factors, to discover possible differences, and to imitate them) (for details, see Squazzoni F. and Boero R., 2002).

Growing Behavioral Attitudes… 13 Figure 1. Evolution of technology and market environment:  is the number of production/simulation cycles. T1, T2 and T3 are the three technological regimes impacting district firms over time. Phases of technology breaking-off are about cycles 500 and 1000. Grey areas show technological positions and related achievable performance levels of firms with respect to technology standard and market evolution. We assume that technological evolution is irreversible (from T1 to T3).

C (0.84  , T3)

B (0.59  , T2) T2

T1 A (0.34  , T1) T3

0.25  0.5  0.75  

The concept of neighborhood calls for the problem of proximity relations amongst firms. We introduce different metrics of proximity viewed as different sources of information for firms. Over time, and with respect to “behavioral attitudes” of agents described afterwards, firms develop a dynamic overlapping web of proximity relations with others, namely spatial, organizational and “social group” forms of proximity (Bellet M., Kirat T. and Largeron C., 1998; Oerlemans L. A. G., Meuss M. T. H., and Boekema F. W. M., 2001; Torre A. and Gilly J.- P., 2000). As it will be outlined in the next paragraph, proximity matters because it produces as byproducts sources of information, possibility of monitoring of the social context, and possibility of comparing individual and social context features. Proximity can be more or less spatial enlarged, more or less geography-dependent,

14 Growing Behavioral Attitudes… more or less organizational relation-dependent, or more or less “social group”-oriented. To regulate all such computational operations, we introduce three matrixes, called “Info Matrix”, “Tech Matrix”, and “Change Matrix”, where all actions are transformed in costs and values.

Matrix 1. “Change Matrix”

T1 T2 T3 Technological Change 200 400 Organizational Asset Change 50 100 200

“Change Matrix” shows costs needed to implement a new technology (first line) or to improve organizational asset, that is to say to change number/factors combination (second line). Along the column, there are all the three “technological regimes” impacting firms over time. Costs gradually increase over time with the gradual growth of market requests and performance needed.

Matrix 2. “Info Matrix”

T1 T2 T3 Technology Imitation 40 70 Organizational Asset Imitation 30 20 10 Technology Innovation 100 250 Organizational Asset Innovation 80 50 30 Best Sub on Performance or on Investment on Organizational Asset 5 5 5

“ Info Matrix” shows costs which firms must pay in order to achieve different type of information. Information concerns both technological strategies (innovation and imitation), and partnership selection mechanisms. The second case refers to different information criteria by which final firms organize their production chains, aggregating a team of sub contracted firms. Final firms need continuously information about economic,

Growing Behavioral Attitudes… 15 technology and organizational features of sub contracted firms in order to choose between stabilizing or destabilizing their inter-organizational contexts (chains).

Matrix 3. “Tech Matrix”

T1 T2 T3 Organizational Asset A B C A B C A B C Worst 5 6 0.01 6.65 9.12 0.01 8.86 13.87 0.01 Best 7.32 10.49 0.01 9.74 15.96 0.01 12.97 24.26 0.01

“ Tech Matrix” shows data about costs and performance of firms in all the different learning steps undertaken by firms. As it is mentioned above, technology costs and economic performance gradually increase as well as the market requests over time. Column A shows technology costs, B shows levels of achievable performance, and C shows decreasing costs for the use of the same combination of number/factors for more than one simulation/production cycle. All costs and performance values are expressed by a continuum between “worst” and “best” technological practice levels, with an average calculus on the degree of distance/nearness of the combination of number/factors implemented by firms with respect to the “best” and “worst” levels.

Finally, we introduce a double metrics of the firm’s profit. Firms have their individual level or profit, due to the difference between costs and levels of economic performance, as it is shown in column A and B of “Tech Matrix”. But, at aggregate level of chains, the “total profit” is not the simple sum of the individual “profit” of interacting firms. We introduce an “extra profit” which mirrors the “technological compatibility” level of firms involved into the same chain. In order to produce quickly and to reach the possible highest level of quality of the good on market, firms need to “speak” the same technological language. In short, such “extra profit” emerging by the production-oriented aggregation of firms, is what we call in

16 Growing Behavioral Attitudes… our computational codes, “time compression” value (see for details: Squazzoni F. and Boero R., 2002). 3. How ID agent works

The foundation of the cognitive architecture of ID agents is based on the hypothesis that agents are able to process information about technology and market environment, district environment, context of partnership, organizational and economic features, and to transform information into possible courses of “appropriate” situated action. We refer to “cognitive architecture” in a very broad sense, that is to say the way by which agents perceive and process information in order to maintain or modify a particular routine of action. “Routinizing” action, agents are able to develop what can be called a “reflexivity monitoring” of “day-to-day” operations in specific fields of action (Giddens A., 1984). Computationally speaking, agents realize their tasks by means of what we call the “information/action loop” (see Figure 2). The information towards action mechanism is driven by a continuos loop which relates data to “rough” indexes, “rough” indexes to “macro indexes”, “macro indexes” to “evaluations”, and “evaluations” to actions. We set up an information set with different data concerning “day-to- day” activities of agents, as it is shown in table 1, and different cognitive steps through which agents use, monitor and transform information into decision. Such data are built both on temporal and spatial dimensions, and even on their interrelation. As in a “cycle of cognition” proceeded from Neisser essential works (Handlbauer G., 2000), the “information/action loop” causes learning characters. According to the “simonian grand theme” (Simon H. A., 1987), such learning characters rely more on “procedural” perspective about

Growing Behavioral Attitudes… 17 information processing undertaken by agents, than on “structural” perspective, as it is pointed out in various classical studies (Craik, 1980). Learning is based on the capacity of agents to typify both their social experience and their routines of action over time. Agents are able to elaborate day-to day ordered evaluations about what they have done, and day-to-day monitoring evaluations about which kind of social context they are moving in. In a sense, action has here what Emirbayer and Mische call a “practical-evaluative dimension” associated with a “relational dimension” (Emirbayer M. and Mische A., 1998). As it is stressed by Giddens, “routinization” of actions is important because of “a sense of trust in the continuity of the object world and in the fabric of social activities […] depends upon certain specifiable connections between the individual agent and the social contexts through which that agent moves in the course of day- to-day life” (1984, p. 60). As it will be outlined, individual experience, behavioral attitude development, and reflexive monitoring of social contexts by means of information processing, cognitive transformation of information into routines, practice of evaluation and monitoring are fundamentals of the cognitive architecture of ID agents. We assume that computational capabilities of agents are bounded, and that time, memory and attention of agents are finite and selective resources (March J., 1994). We assume that agents cannot act cognitively with parallel processing mechanisms, namely they cannot control, manage and face the entire set of information with the same level of “cognitive attention”. Moreover, we assume that there is a tradeoff between “width” and “depth” of the cognitive process. As it will be outlined, all the cognitive steps undertaken by agents imply an information processing activity based on approximation,

18 Growing Behavioral Attitudes… abstraction and synthesis of the “relevant attributes” belonging to the information. In conclusion, the “information/action loop” starts with a “domain specific” information and it ends, by means of specific cognitive procedural processes, with a “broad generic” information upon which the decision process of agents is based.

3.1 Cognitive step 1: from information to “rough indexes”.

The first cognitive step is the transformation of information into “rough indexes” of “attribution”. Information is about all the topics faced by firms. “Rough indexes” allow agents to assign a “positive” or “negative” judgment to information, which is expressed by a computational dichotomy of 0 and 1 values. Agents cluster, synthesize, and categorize information belonging to the same topics, transforming “numbers” into “evaluations”, even if through a “first inference” on a “rough information”. “Rough indexes” are as follows:

- “ sold” (index allows a first inference on market effectiveness of firms and their neighborhoods and a comparison amongst such values) - “ time compression” (index allows to final firms a first inference on “technological compatibility” of their production chains and their neighboring chains, and a comparison amongst such values); - “performance” (index allows an inference on the effectiveness on market and a comparison with the neighborhood); - “ number of chains” (index allows an inference on the degree of “stability” and “good relations” amongst firms);

Growing Behavioral Attitudes… 19 - “ selling firms” (index allows an inference to the effectiveness of the system as a whole);

Figure 2. From information to indexes by means of the “approximation-abstraction-synthesis mechanism”. In order to transform information data (the lightest solid), into rough indexes (the middle one) and then into macro indexes (the darkest solid), agent-based cognitive operations meet a tradeoff between increase of the degree of the three dimensions (abstraction, synthesis, approximation) and a decrease of the volume of information to be considered.

Abstraction

Synthesis

Approximation

20 Growing Behavioral Attitudes… - “ technological change” (index allows an inference to the degree of “technological instability” of the system as a whole); - “ searching for new sub firms” (index allows an inference of the instability of inter-firm relations and the tendency to the emergence of new partnership assets within the system as a whole); - “ technology” (index allows a comparison amongst the “technology level” of firms and neighborhood, standard); - “organizational asset effectiveness” (index allows a comparison between the level of effectiveness of the “organizational assets” of firms and neighborhood); - “ homogeneity of criterions for keeping sub firms” (index allows to evaluate the degree of uniformity of the “inter-organizational assets” within the neighborhood); - “ homogeneity of criterions for searching sub firms” (index allows to evaluate the degree of diffusion of changes in the “inter-organizational assets” within the neighborhood); - “profit over time” (index allows an inference on the relation of level of profit of firms and neighborhood over time, namely using an inference with temporal retrospective dimension); - “resources over time” (index allows an inference on the relation between level of resources of firms and neighborhood over time, namely using an inference with temporal retrospective dimension); - “ performance over time” (index allows an inference on the relation between performance of firms and performance of neighborhood over time, namely using an inference with temporal retrospective dimension);

Growing Behavioral Attitudes… 21 - “ investment on technology over time” (index allows to compare the average of the technology investment of firms and neighborhood over time, namely starting from data on the last 20 simulation cycles); - “investment on organizational asset over time” (index allows to compare the average of the investment on organizational assets of firms and neighborhood over time, namely starting from data on the last 20 simulation cycles).

3.2 Cognitive step 2: from “rough indexes” to “macro indexes”.

Agents are able to extrapolate on such “rough indexes” five macro aggregated indexes, developing a level of “cognitive abstraction” even more synthetic in respect to the previous step. Such macro indexes are both external and internal ones, that is to say that they allow to monitor both technology and market environment, social context, and individual features of agents’ activity. As it is shown in Figure 3, we call “PETOE Scheme” the structure of the macro indexes. Indexes are five: “partnership”, environment”, “technology”, “organization” and “economic”. The former two refer to “external dimensions” of the firm, while the latter three refer to its “internal dimension”. Macro indexes are as follows:

- “ partnership” (index allows to synthesize “rough indexes” on “sold”, “time compression”, “performance”, and “number of chains”, in a macro- inference of the “positive” or “negative” nature of the features of the “partnership context”); - “environment” (index allows to synthesize “rough indexes” on “selling firms”, “technological change”, and “searching for new sub firms”, in a

22 Growing Behavioral Attitudes… macro-inference on the “stable” or “unstable”, or “fair” and “unfair” nature of technology and market environment); - “ technology” (index allows to synthesize “rough indexes” on “technology” and “organizational asset effectiveness” in a macro- inference on the individual degree of technological effectiveness); - “ organization” (index allows to synthesize “rough indexes” on “homogeneity of criterions for keeping sub firms”, “homogeneity of criterions for searching sub firms”, and “homogeneity of criterions for searching sub firms” in a macro-inference of the “positive” or “negative” nature of the organizational fundamentals of the firm); - “economic” (index allows to synthesize “rough indexes” on “profit over time”, “resources over time”, “investment on technology over time”, and “investment on organizational asset over time” in a macro-reference on the “positive” or “negative” nature of the economic fundamentals of the firm).

Macro indexes synthesize and cluster “rough indexes” at a higher level of cognitive abstraction. The relation between “macro indexes” and “rough indexes” conforms to a general rule as follows:

M a Wa1Ra1 Wa2 Ra2 ...Wan Ran where Ma , Wa1 , Ra1 , etc..  [0,1] and Ma represents a macro index, Wa1 is the weight of the first rough index Ra1 and so on. The shift from rough to macro indexes is based on a computational procedure called “Weighted Average” performed by agents over time. Such procedure is characterized by a heterogeneous assignment of relevance- driven attention undertaken by agents on specific indexes.

Growing Behavioral Attitudes… 23 3.3 Cognitive step 3: indexes evaluation

According to the “information/action” loop shown in figure 2, before acting, agents need to be able to evaluate such indexes. The evaluation process calls for the problem of which kind of “behavioral attitudes” agents develop over time. Attitudes are conceived as different possible “states” of agent’s behaviors emerging from a continuum between more or less degree of “districtualization”. Agents can be more or less ‘districtualized’, in the sense that their behavior can be more or less affected by the characteristics of their social context. An agent less districtualized is pushed to think more in terms of “individual centered self”. It can set aside the features of its context of interaction and social experience. Its decisions are not particularly bounded by social neighborhood influences. Otherwise, an agent more districtualized is pushed to think more in terms of “social group self”. Its attitude is characterized by a more active “identification” with other agents belonging to the same context of experience. In short, we assume that the features of the social context have a deep influence on the individual cognitive process when “social reflexivity” of agents grows over time. Thus, over time, agents develop different behavioral attitudes. They can conceive themselves as “isolated” atoms, or as “parts of a microcosm more socially enlarged”, or as members of a “social group”. Behavioral attitudes change over time with the growth of “reflexive capabilities” of agents towards the macro-characteristics of their context of experiences, by means of their “monitoring activities” concerning “global” features of their context of relations. In our perspective, such processes do not work in a kind of simple linear and irreversible progress, from the “state” of “individual self” to the “state” of “social group member”. The growth of the “reflexive

24 Growing Behavioral Attitudes… capabilities” of agents implies even their capability to “dis-districtualize” themselves, to escape from a social group, to “come back” and become once again “isolated atoms”. Agents can choose to have “rich” or “poor”, “long- time” or “one-shot”, closed” or “open” relations with others, and neighborhood relations can be more or less wide and deep. Agents can be interest in interacting with other firms just to produce goods but without founding sound and stable cooperation relations, that is to say they can act in a pure “market-like” style.

3.3.1 Behavioral attitude states

The possible behavioral attitudes of an agent are as follows: a) state 0, or the “self-centered” attitude - Agent is located in a context, it has a position within the space, that is to say it has specific neighboring agents; - Agent enacts production relations with other agents, produces and sells products, tries to increase its economic performance, its technological profile, its organizational asset, and so on; - Agent is not interested in establishing stable and rich relations with other agents, that is to say it seeks for “one-shot interactions”, focusing continuously on imperatives of the “economic performance” (Squazzoni F. and Boero R., 2002); b) state 1, or the “chain-management” attitude - Agent is interested in maintaining stable and rich relations with other interacting agents;

Growing Behavioral Attitudes… 25 - Agent thinks about chain as an “unit”, that is to say as a locus of organizational relations and relevant information, and as a source of technological learning coordination; agents start to conceive “complementarity” relations with others; - Agent enlarges the microcosm of the “state 0” to five other agents (spatial and organizational neighborhood); c) state 2, or the “clustering” attitude - Agent with “chain management attitude” meets other agents with “chain management attitude” which put “trust” on the importance of what can be called “social horizon enlargement” policies; - Agent belonging to stable chains enlarges its microcosm to other agent belonging to stable neighboring chains (spatial and organizational neighborhood plus neighboring chains); - Agent exchanges information with other agents without having direct interactions; d) state 3, or the “grouping” attitude - Agent starts to reflect upon the collective properties of the “cluster” and tries to improve the “collective effectiveness” of the “cluster”; - Agent recognizes all the other agents as member of the “cluster” and interacts with them; - Agent can exchange information and partners within all the group and makes social distribution policy of the “extra profit”.

Firms start the simulation as self-centered attitude agents (state 0). We assume that all the shifts amongst states depends on the presence of different

26 Growing Behavioral Attitudes… mechanisms of regulation of the inter-firm relations. Agents develop the perception of possible economic benefits which can emerge by the cooperation with others and are pushed to define better, and in a more stable way, their contexts of interaction. Agents start to conceive their context of interaction as a tool of learning and information. They start to explore actively the social environment because it is perceived as a source of information, comparison and mutual monitoring with other contextualized agents. Exploring the context, agents create with others a kind of “relational tie”, where information is exchanged, learning takes mutual directions, and resources are less or more shared. As it will be outlined, this implies that agents develop a cognitive representation of their tasks above all in terms of “relationship” (Bickhard M. H., 2000).

3.3.2 Behavioral attitude morphogenetic shifts

We define the agent-based process of elaboration and change of “behavioral states” from the “bottom-up” (from state 0 to state 3) over time as “morphogenetic shifts”. The shift from state 0 to state 1 depends on the emergence of a relative stability of a production chains (five cycles of recurring interactions with the same team of sub contracted firms) and on specific conditions of the macro indexes on “partnership” and “economic fundamentals” (macro index of “partnership” ≥ 0.75; macro index of “economic” ≥ 0.75). According to such conditions, agents can develop a “behavioral attitude” towards the transformation of the previous “recurring interactions” in “stable partnership relation”, loosing their previous “self- centered attitude”. We assume that agents, facing a state of good economic performance and perceiving a potential good context of interactions, are

Growing Behavioral Attitudes… 27 pushed to define, in a more binding way, their organizational relations. In a sense, agents put “trust” in their organizational neighborhood contexts. The shift from state 1 to state 2 depends on specific condition of the “partnership index” (value of 0.95). A “clustering behavioral attitude” implies the interest about, and a sharing of the information contained within the whole spatial neighboring firms. The next step is the diffusion of the “clustering behavioral attitude” within spatial neighboring chains, when similar condition of “trust” in the partnership mutually grows amongst agents. This is the mechanism which allows the diffusion of the state 2 amongst firms. This implies that spatial proximity relations start to develop cluster proximity relations. The shift from state 2 to state 3 depends on conditions as follows: if “partnership index” or “environment index”, or both ones are > 0.75, then at least two of three others (“technology”, “organization” and “economic index”) ≥ 0.75; if the spatial neighboring agents are already in the state 3. “Partnership index” and “environment index” give to agents the “trust” on the positive global state of the industry as a whole. The other indexes show a positive combination related to the individual state of the agent. We assume that, in this condition, agents are pushed to reflect in a more global way and to conceive the problem of the relation between individual effectiveness and collective effectiveness of the “group” as a whole. It is worth to notice that the growth of such “morphogenetic shifts” over time implies an expansion of the neighborhood relation of firms and an increase of information “deep on the ground”, that is to say an individual information more compared to information experienced by other agents. On the state 0, agents put “trust” on macro indexes and “antennas” which allow a broad vision of the industry as a whole. The nature of the information is

28 Growing Behavioral Attitudes… very general and not so rich, deep and “domain specific”. In short, the “morphogenetic shifts” of behaviors are evolutionary mechanism through which agents can improve their learning on what is to be done, by means of a more deep comparison both between “macro indexes” and specific actions, and between what an agents does and what others are to do.

3.3.3 Behavioral attitude deconstruction shifts

“ Bottom-up” shifts mentioned above are not equated with linear and irreversible processes. In fact, agents can develop, change and destroy continuously their “behavioral attitudes”, over time. It is a matter of cognitive adaptation in respect to contexts and environments. Facing some “positive” cognitive configuration, agents develop a bottom-up process of elaboration of their behavioral attitudes (from 0 to 1, and so on), while facing some “negative” cognitive configuration, agents destroy their “behavioral attitudes” turning back to previous steps, in a top-down process. This “turning back” process does not works as a simple mirror of the “bottom up” in steps as follows: from state 3 to state 1 or 0, from state 2 to state 0, and from state 0 to state 1. The process of deconstruction of the behavioral attitudes conforms to a general computational rule as follows: if all the macro-indexes, both external and internal, are ≤ 0.5, then agents shift from state 1, 2, or 3 to state 0. Moreover, we assume other specific conditions for deconstructing behavioral attitudes as follows: a) deconstruction from state 3 to state 1, or the “group exit option”

Growing Behavioral Attitudes… 29 - conditions are: if “technology”, “organization” and “economic index” < 0.25, and if “partnership” and “environment index” > 0.5 b) deconstruction from state 2 to state 0, or the “cluster exit option” - conditions are: if production chain asset is broken; a production chain is broken if one of these four actions is “true”: “sold=0”, “profit t < profit t- 5”, “resources t < resources t-5”, “time compression t < time compression t-5”; if c) deconstruction from state 1 to state 0, or the “free hands option” - conditions are: if production chain asset is broken; a production chain is broken if one of these four actions is “true”: “sold=0”, “profit t < profit t- 5”, “resources t < resources t-5”, “time compression t < time compression t-5”.

The deconstruction process which pushes agents to shift from “group attitude” to “chain-management attitude” is enacted when the membership implies a deep decreasing of benefits for individual agents and external conditions of environment are perceived as “positive”. Agents are pushed to perceive their exit from the group as a source of possible benefits. Thus, agents shelter in their chain-microcosm. This is what we call the “group exit option”. The deconstruction process which push agents to shift from “clustering attitude” and “chain-management attitude” to “self-centered attitude” is based on the emergence of “negative” indexes about individual features. These are what we call “cluster exit option” and “free hands options”.

30 Growing Behavioral Attitudes… In conclusion, on one hand the dynamics of “morphogenetic shifts” hides what can be called the growing emergence of “districtualization” of agents. On the other hand, the dynamics of “deconstruction shifts” hides what can be called a “self-centered re-organization” of agents. It is a matter of an organizational schumpeterian “creative destruction” process. With this respect, because of the “structuring properties” of the prototype, it is worth to notice that final firms have a focal role. They are what Douglass R. Hofstadter (1979) calls “catalyst enzymes”, that is to say, in our words, agents endowed with the capacity to continuously grow, select, destroy and redefine the ID cognitive and relational architecture of the context within which they move.

3.4 Cognitive step 3: actions

Therefore, agents develop different “behavioral attitudes” over time, and they act into different “operation fields” having finite “action recipes”, as it is shown in table 2. The “operation fields” are what we call “technology”, “keep”, “search”, and “share”. “Technology” refers to the need of agents to exploit context-based local information to improve their “technology” and “organizational assets”. “Keep” and “Search” refer to how agents manage their partnership relations, between needs of stabilization and thrusts of de-stabilization of their relational contexts. “Share” refers to what policy of chain profit management agents do. Behavioral attitudes have the properties to relate such fields to specific “action recipes”. As it is shown in table 2, different behavioral attitudes imply the use of specific “action recipes” in specific “operation fields”.

Growing Behavioral Attitudes… 31 The principle which rules table 2 is that higher the behavioral attitude of agents towards social environment, that is to say higher the level of embeddedness of agents within the social context, then higher the “width” and the “depth” of the information it has and more developed and keen are the receptivity of their antennas oriented to environment. The differences of the “action recipes” conforms to this rules. Another principle is that agents develop routines, that is iterations of the same “action recipe” within an “operational field”. In our sense, what Giddens calls “routinization” works by means of what Douglass R. Hofstadter calls the “prototype principle”, and what Marvin Minsky calls the “analogy” mechanism, that is to say by means of capacity and tendency of agents to “represent each new thing as though it resembles to something [they] already know” (Minsky M., 1985). With this respect, figure 4 shows the “actions code” of agents. In short, at the start of the simulation, agents use a specific “recipe” assigned in a random way. Over time, they carry on using it, that is to say transform the “recipe” into routine. The routine is broken when macro indexes push the agent to change it. Thus, the agent starts a phase of trial and error process trying to define a new routine within “action recipes”. Thus, routines can be maintained or changed, and this is a focal phase of the agent’s action. The role of macro indexes and their configuration is fundamental for understanding why and how agents change or maintain their routines. Macro indexes configuration is conceived as the adaptation mechanism which force agents towards learning about routines. We set a fixed number of indexes configurations and the presence of a kind of “ringing bell” mechanism which represent the capacity of agents to perceive the presence of an unsatisfactory routine. Specific configurations of macro-indexes cause

32 Growing Behavioral Attitudes… the activation of the “ringing bell” mechanism, driving the attention of the agent on a specific topic. But, macro-indexes contain just a “synthetic” information about agent’s problem. In fact, the “ringing bell” means just that agent has some problem with its routines. According to such fixed combinations, and because of “cognitive limitations” about “memory”, “time”, “attention” and “self- monitoring”, the agent can change its routines just within a specific operation field (Technology, Keep, Search, Share), that is to say that it has limitations in proceeding to estimate the routines value (goodness) and in identifying the “critical” routine. The agent starts to perceive a problem on a specific “operational field” and develop a phase of evaluation and learning based on the exploration of other possible “action recipes” based on a “memory function” which collects data on the last “five time period” where a specific routine has been used. The agent needs to learn how to solve the problem within an “operational field” defining new routines. The “ringing bell” mechanism works as follows: a) in the case of sub firms, if “technology index” and “economic index” < 0.25, the “ringing bell” focuses on “technology”; the agent perceives the necessity of change its routine on such “operational field”; b) in the case of final firms: - if all the indexes ≤ 0.5, the agent falls in a “panic condition” and starts to change randomly its routines on one or two different “operation fields”; - otherwise, if “technology index” < 0.25, the agent starts to change its routine in the “technology operation field”; if “organization index” <0.25, the agent starts to change its routine with equal probability in

Growing Behavioral Attitudes… 33 “keep” or “search” operation fields, while if “organization index” < 0.25 ≥ 0.5, the agent starts to change its routine in “share” operation field; if “economic index” < 0.25, the agent starts to change its routine in “technology” operation field and with equal probability its routine in “keep” or “share” operation field.

The “ringing bell” mechanism allows us to computationally manage the relation between representation of contexts and environments developed by agents and the specific actions they undertake. The mechanism is based on the hypothesis that selective attention of agents is oriented towards fixed “operation fields”, and directed to specific significant areas of the ‘problem space’, by means of a sort of distinctiveness (Lockhart R. S. and Craik F. I. M., 1990). The principle of the change of routines is that agent facing the perception of a problem within an “operation field” can use its memory on past routines, that is to say the last five periods of time during which a routine has been used, to support its routine definition process. Agent can relate routines to macro-indexes in order to define “positive” or “negative” associated values. According to the “memory function”, it changes, evaluates and chooses routines. Here, the cognitive process of routine definition is based on steps as follows:

- the agent has memory of routines implemented in the past, even if concentrated upon macro-indexes and to bounded time periods (last five time cycles of implementation); - its space of possible routines is limited by its “behavioral attitude”, as it is shown in table 2;

34 Growing Behavioral Attitudes… - the agent uses continuously memory function to developing data about all routines used; - if within the space of all the possible routines, there is a routine not yet explored, the agent chooses this last one; - the agent creates an average of collected data on past routines; - in the case of complete exploration of all possible routines, using data referring to the past, the agent defines its new routine according to an evaluation about the relation between routines and macro-indexes.

It is worth to notice that the link between memory and information processing has a procedural cognitive nature, focused on a “frugal” design by which cognitive limitations of agents imply a restricted possibility of items recalling. Agents explore, maintain and change routines of actions by means of a step-by-step adjustment mechanism by which agents develop information according to specific search rules within specific behavioral attitudes (Gigerenzer G. and Selten R., 2001). In short, as it is shown in table 2, behavioral attitudes imply search rules towards “operational fields”- oriented problem solving activities which are monitored by agents. Search rules act within different “repertoires of routines”, which are composed by different “action recipes”, according to “behavioral attitude” developed by agents over time. In conclusion, the cognitive architecture of ID agents is based on “cognitive typification” activities which relate continuously individual experience and social contexts. Macro-indexes evaluation is a cognitive step through which agents try to incorporate information and to develop “attribution” about state of technology and market environment, characteristics of their relational social contexts and control of their own

Growing Behavioral Attitudes… 35 individual features, in order to find “appropriate” strategies of technological learning. “Reflexive typification” works through the capacity of agents to assign “objective” characteristics both on their experiences, their social context, and their operation environment.

Table 1: Information Processing of ID Agents

Information Rough Indexes Macro Indexes Selling Result Sold Index Neighbors Selling Result Time Compression Time Compression Index Neighbors Time Compression Performance Partnership Index Performance Index Neighbors Performance Number of Stable Production Number of Chains Index Chains in the Industry Percentage of Firms Selling to Selling Firms Index the Market Percentage of Technological Technological Changes Environment Jumps Index Index Percentage of Brand New Searching for New Subs Production Chains Index Technology Level Technology Index Neighbors Technology Level Organizational Asset Technology Index Effectiveness Level Organizational Asset Neighbors Organizational Effectiveness Index Asset Effectiveness Level Percentage of Extra-Profit Homogeneity of Extra- Organization Neighbors Percentage of Profit Sharing Policies Index Extra-Profit Index Routine for Choosing if Keep Actual Sub Firms Homogeneity of Criterions Neighbors Routine for for Keeping Sub Firms Choosing if Keep Actual Sub Index Producers Routine for Choosing New Homogeneity of Criterions

36 Growing Behavioral Attitudes… Sub Firms for Searching New Sub Neighbors Routine for Firms Index Choosing New Sub Firms  Profit Profit over Time Index Neighbors  Profit  Resources Resources over Time Neighbors  Resources Index  Performance Performance over Time Neighbors  Performance Index (Investment on Investment on Technology Economic Index Technology )/20 t,..,t-20 over time Neighbors (Investment on Index Technologyt,..,t-20)/20 (Investment on

Organizational Assett,..,t-5)/5 Investment on Neighbors (Investment on Organizational Asset Index

Organizational Assett,..,t-5)/5

Figure 2: Information/Action Loop

information actions

rough indexes indexes evaluation

macro indexes

Growing Behavioral Attitudes… 37 Figure 3 The “PETOE” Scheme: Macro Indexes Refer to Different Topics and Environments

World Partnership I. Environment I.

Firm/Group

Technology I. Organization I. Economic I.

Table 2: Relations amongst “operation fields”, “behavioral attitudes” and “action recipes”.

Action Recipes Behavior Operation At tit ud es look at the first agent with different Self Centered Technology technology/organizational asset you meet look at the first agent with different (imitation in technology/organizational asset you meet, which has the sub- sold its product

38 Growing Behavioral Attitudes… look at the agent with different technology/organizational asset you meet, which has a percentage of extra-profit better than yours and the highest available look at the agent with different technology/organizational asset you meet, which has a behavioral attitude higher than yours and the highest available look at the agent with different technology/organizational asset you meet, which has Chain a level of resources better than yours and the highest Management available Clustering look at the agent with different technology/organizational asset you meet, which has a level of profit better than yours and the highest available look at the agent with different fields of technology/organizational asset you meet, which has technology a level of cost higher than yours and the highest and available organization look at the agent with different asset) technology/organizational asset you meet, which has a level of effectiveness of organizational asset better than yours and the highest available look at the agent with different Grouping technology/organizational asset you meet, which has a level of investment on technology/organizational asset better than yours and the highest available look at the first agent with different technology/organizational asset you meet, which has a level of performance better than yours and the highest available keep your team of sub firms if time compression t, Self Centered Keep t-1 >= 0

Growing Behavioral Attitudes… 39 keep your team of sub firms if profit t, t-1 >= 0 keep your team of sub firms if resources t, t-1 >= 0 keep your team of sub firms if you have sold your (strategy of product Chain partnership keep your team of sub firms if time compression t, Management stabilization) t-5 >= 0 Clustering keep your team of sub firms if profit t, t-5 >= 0 keep your team of sub firms if resources t, t-5 >= 0 Grouping search for a new team of sub firms randomly search for a new team of sub firms focusing on who Search Self Centered has the highest investment on organizational asset Chain search for a new team of sub firms focusing on who (strategy of Management has the highest performance partnership search for a new team of sub firms focusing on who Clustering definition) has the most similar technology and organizational Grouping asset configuration give to your partners the 0% extra-profit give the 5% extra-profit to each partner Self Centered give the 10% extra-profit to each partner Share give the 13.3% extra-profit to each partner Chain give the 16.6% extra-profit to each partner (policy of Management give the 20% extra-profit to each partner chain extra- Clustering give the 23.3% extra-profit to each partner profit

give 25% extra-profit to each partner management give the 70% extra-profit to partners, distributed and Grouping proportionally according to their needs distribution) distribute proportionally the 100% extra-profit according to the needs of each member of the chain

01 11 01 00 Figure 4: Actions Code of the ID Agent (according to structuring properties of the prototype, just final firms have the complete actions code) Routine for choosing neighbor from which imitate technology and organizational asset (“true” for all the firms).

Routine for choosing if sub firms need to be changed

Only for Routine for finding new sub 40 final firms Growing Behavioral Attitudes… firms.

Routine for sharing extra-profit. Growing Behavioral Attitudes… 41 4. Indicators and Prototype Settings

To test the prototype, we use several indicators. By observing them, it is possible to grasp fundamental dynamics emerging by the prototype. We create also different experimental settings in order to reinforce evidences about how behavioral attitudes and typology of social contexts affect performance of firms (running the prototype it is possible to choose all the different combinations of behavioral states, right from the start of the simulation). Indicators we use here are as follows (running the prototype, it is possible to observe and produce other kinds of indicators):

- final firms matching market requests over time; - final firms performance and behavioral attitudes over time; - final firms performance in different prototype settings running separately with behavioral attitudes state 0, state 0 and 1, and with complete behavioral states; - weight of the different macro indexes over time; - dimension of the neighborhood relations over time.

42 Growing Behavioral Attitudes… 5. Analysis of outcome and emerging dynamics

Before observing the outcome of the simulation on the prototype, we have set some question to test: Is there a positive relation between development of behavioral attitudes more oriented to social contexts and technological leaning of firms, that is to say are agents able of technological learning just by means of the enlargement of their social contexts? or are agents facing technological breaking-off more oriented to perceive social contexts as adaptation “individual constraints”? in short, have social grouping and cognitive identification activities towards social contexts a positive or negative impact on economic performance of district firms? is it better to districtualize during phase of technological continuity and dis-districtualize during phases of technological discontinuity? If we observe the simulation outcome, we can sketch several inferences. Technological breaking-off phases imply a selection of the district firms, even if with different dynamics. As it is shown in figure 6, the first discontinuity phase (about cycle 500) is absorbed by the 88% of firms, while the second phase causes a strong oscillation in the firms’ performance, but without implying a further exit of firms from market. This is due to the fact that firms over time are more effective in technological learning, despite the growth of costs and request of technological quality of their goods marked by market. The evolution of behavioral attitudes states over time shows that firms facing technological discontinuity and increasing market pressure phases develop different strategies of response over time, while firms facing

Growing Behavioral Attitudes… 43 technological continuity tend to stabilize their “behavioral attitudes”. As it is shown in figure 6, the first phase of technology and market stability (until 500 cycle) shows a tendency of agents to lock-in their behavioral attitudes with a lot of them in state 2 (“clustering attitude”) and few of them in state 0 (“self centered attitude”). The 10% of agents are quickly able to develop the “grouping attitude” (state 3), while another 10% of agents lock-in their behavioral attitude right from the start in the state 0. Such stabilization of the behavioral attitudes goes on until the first technology breaking-off (around cycle 500). In this phase, agents in more critical technology and market conditions try to develop their behavioral attitudes, above all shifting from state 0 to state 2, but without success. They are the first and the only victims of the market selection. The behavioral state 1 (“chain-management attitude”) is just a shelter in times of technology and market deeper challenge, along all the simulation time. Just as before, the second phase of technological stability shows a “long durée” settlement of behavioral attitudes of agents. Certainly, the second phase of technological instability is more interesting than the previous one (around cycle 1000). As it is shown in figure 5, here district firms live a deep but quick phase of market crisis. How firms face such crisis, from the point of view of their behavioral attitudes? As it is possible to observe comparing figure 5 and 6 around cycle 1000, firms involved in such crisis are above all those in state 2 (“clustering attitude”). They do not simply destroy their behavioral state, for instance passing from state 2 to state 0, but someone develops a like-state 3 “behavioral attitude”. It is worth to notice that the so called “grouping” behavioral attitude of firms stands up despite the two technological and market crisis, and even strengthens over time.

44 Growing Behavioral Attitudes… Such strengthening-effect of the “identity towards identification” attitudes over time can be as well confirmed if one observes figure 7, where data about dynamics of the average dimension of neighborhood proximity relations allow to observe how much large is the context of relations enacted by agents over time. Such enlargement of the social horizons of agents not only grows over time, but rather it grows during phase of technology and market adaptation challenges. Such data tell us that “identification” dynamics developed by agents are not only a fundamental tool of technological learning and economic performance for firms, but rather that district firms develop a polarization of “grouping attitudes” over time, and that they are over time more enhanced and reinforced, when agents face technology and market adaptation needs.

Figure 5. Final firms matching market requests over time.

Growing Behavioral Attitudes… 45 Figure 6. Final firms matching market requests over time at variance of behavioral attitudes.

Figure 7. Average dimension of neighborhood proximity relation.

If we observe figure 8, where data on the relevance of the different “macro indexes” over time are shown, it is possible to outline that at the

46 Growing Behavioral Attitudes… start of the simulation firms develop too much “optimistic confidence” on the “fair nature” of the technology and market environment. The problem perceived by agents during the technology breaking-off phase is just the redefinition of routine towards environment, and the perception of an “environment complication” is let off on problems on the “technology indexes”. More than problems on operation fields of “economic” and “organization” indexes, agents perceive problems on “technology” features. Finally, we create different prototype settings by changing mechanisms of behavioral attitudes development. As it is shown in figure 9, we set a prototype running just with behavioral attitude state 0, and running with state 0 and 1. The outcome of set with state 0 and 1 confirms that behavioral attitude state 1 (“chain management attitude”) causes the loss of the advantage of market-oriented “self centered attitudes” without generating the advantage of information source and processing typical of a wide “relational context”, as in the state 2 and 3. Figure 9 shows that at the end of simulation cycles, levels of firms still on market are as follows: 88% in the complete set, 76% in state 0, and 66% in state 0 and 1.

Figure 8. Relevance of the different “macro indexes” over time.

Growing Behavioral Attitudes… 47 Figure 9. Final firms matching market requests on different experimental settings. From the left to the right, outcomes of state “complete”, state 0 and 1, and state 0.

In conclusion, the simulation outcome of the prototype shows that district agents are able to develop different behavioral attitudes over time, such attitude development has an positive effect on long-time period learning of agents, and social relational context and “districtualized” behavioral attitude are more deeply developed during phases of technology and market challenge. As in the case of technology breaking-off phases, if we compare figure 7, 8 and 9, it is possible to outline that social projection of agents and “districtualization” of firms within their contexts of action are conceived not as a “constraint” upon the individual economic imperative, but rather as a source of information and learning about environment challenge.

48 Growing Behavioral Attitudes… 6. Conclusion: how agent-based computational models can put down stable roots in social sciences?

But, how to further develop the district prototype? In this paper, our intentions were to move some steps towards an useful and integrated way of using agent-based computational models in social sciences. In fact, our opinion is that social phenomena need to be studied not only from a bottom- up emergent properties perspective, as in the traditional literature on agent- based computational models, but also from a perspective focused on “reflexive capabilities” of “human and social agents” towards the social contexts which they live in. Such perspective has been suggested some years ago by Nigel Gilbert, Rosaria Conte and Cristiano Castelfranchi (1996), and more recently again by Rosaria Conte (2000). Sociological perspective on computational models starts by the awareness that agent-based model mainstream has underestimated the difference between modeling reactive/simple which respond to environment signals and proactive/purposeful agents which are able to reflect upon the global characteristics of contexts within which they move:

“ some computer simulations may have oversimplified important characteristics of specifically human societies, because the actors (agents) in these societies are capable of, and do routinely reason about the emergent properties of their own societies. This adds a degree of reflexivity to action which is not present (for the most part) in societies made up of simpler agents, and in particular is not a feature of most current computer simulations” (Gilbert N., 1996; see also: Caldas J. C. and Coelho H., 1999; Chattoe E., 1998).

Growing Behavioral Attitudes… 49 As it is stressed by Nigel Gilbert, “the complication in the social world is that individuals can recognise, reason about and react to the institutions that their actions have created. Understanding this feature of human society, variously known as second-order emergence, reflexivity, and the double hermeneutics, is an area where computational modelling shows promise” (Gilbert N., 1996; Gilbert N., Terna P., 2000)…“not only can we as social scientists distinguish patterns of collective action, but the agents themselves cal also do so and therefore their actions can be affected by the existence of these patterns” (Gilbert N. e Troitzsch K. G:, 1999). The famous question about the relation between individuals and aggregate posed some years ago by Thomas Schelling (1979) and the question about “how does the heterogeneous micro-world of individual behavior generate the global macroscopic regularities of the society?”, more recently confirmed by Joshua Epstein and Robert Axtell (1996), are issues towards which agent-based computational social scientists have been usually interested. Otherwise, issues towards which agent-based computational social scientists need to be more interest are about the other side of the question: “how such emerging regularities are monitored, reflected, experienced and changed even “intentionally” by heterogeneous agents?”. Clearly, institutions emerge by interactions amongst heterogeneous agents, according to distributed artificial intelligence mechanisms, but agents also reflect upon them, perceive their fundamental features, monitor, typify and “intentionally” change them. In a top-down way, institutions are produced and have reinforcing effects, even dysfunctional ones. In conclusion, to integrate more actively agent-based computational models in the estate of social sciences, social computational scientists need

50 Growing Behavioral Attitudes… not only to show that computational models allow to formalize complex social phenomena in a way that traditional mathematical and statistical tools can not do (Hanneman R., 1995). They also need to address typical social science theory problems, as it is stressed by so called “socionists” (Malsh T., 2001). To do this, they can not simply pursue a “bottom-up reductionism” way, comparing human agent intelligence to ant “swarm intelligence” (Bonabeau E., Dorigo M., Theraulaz G., 1999). According to such ideas, our intentions are to develop the district prototype towards two interlaced directions:

- introducing a meta-cognitive level in the cognitive architecture of district agents, by which agents should be able to direct their own actions considering specific relations between the “ringing bell” mechanism and the operation fields, not only by means of a strategy based on a preset rule of searching alternatives; the way to symbolize a meta-cognitive frame is to show how agent could act as a result of a learning process which allow them to set up different link amongst specific information sources and operation fields they choose to control and manage;

- introducing a formal institutional level in the cognitive architecture of district agents, that is to say the capability of agents to grow formalized institutional contexts able to reinforce some “behavioral attitudes” of agents, redefining the cognitive process of agents by means of a set of macro “cognitive urgency” impacting macro indexes, or the antennas of agents; by looping top-down and bottom-up levels, what could happen is what Nigel Gilbert call “second order emergence” (1996).

Growing Behavioral Attitudes… 51 References

Albino V., Garavelli A. C., Schiuma G. (1999) Knowledge Transfer and Inter-Firm Relationships in Industrial Districts: The Role of the Leader Firm, Technovation, 19: 53-63.

Auyang S. Y. (1998) Foundations of Complex-System Theories in Economics, Evolutionary Biology and Statistical Physics, Cambridge University Press.

Bellet M., Kirat T. and Largeron C. (Eds.) (1998) Approaches Multiforms de la Proximité, Hermes Science Publications, Paris.

Belussi F. and Gottardi G. (Eds.) (2000) Evolutionary Patterns of Local Industrial Systems: Towards a Cognitive Approach to the Industrial District, Ashgate, Aldershot Brookfield Singapore Sydney.

Bickhard M. H. (2000) Information and representation in autonomous agents, Journal of Cognitive System Research, 1.

Boari C. and Lipparini A. (1999) Networks within Industrial Districts: Organizing Knowledge Creation and Transfer by means of Moderate Hierarchy, Journal of Management and Governance, 3: 339-360.

Boero R. and Squazzoni F. (2001) A Computational Prototype of Industrial District, forthcoming, .

Bonabeau E., Dorigo M., Theraulaz G. (1999) Swarm Intelligence. From Natural to Artificial Systems, SFI Studies in the Sciences of Complexity, Oxford University Press, New York.

52 Growing Behavioral Attitudes… Caldas J. C. e Coelho H. (1999) The Origin of Institutions: Socio-economic Processes, Choice, Norms and Conventions, Journal of Artificial Societies and Social Simulation, vol. 2, n. 2,

Chattoe E. (1998) Just How (Un)realistic are Evolutionary Algorithms as Representations of Social Processes?, Journal of Artificial Societies and Social Simulation, 1, 3.

Conte R. (2000) The Necessity of Intelligent Agents in Social Simulation, in Ballot G., Weisbuch G. (Eds.), Applications of Simulation to Social Sciences, Hermes Science Publishing Ltd, Oxford: 19-38.

Conte R. and Castelfranchi C. (1996) Simulating Multi-Agent Interdependencies: A Two-Way Approach to the Micro-Macro Link, in (Eds.), Troitzsch K. G, Mueller U., Gilbert N. e Doran J., Social Science Microsimulation, Springer-Verlag, Berlin: 394-415.

Craik F. I. M. (1980) Cognitive views of human memory, Washington D.C., American Psychological Association.

Emirbayer M. and Mische A. (1998) What is Agency?, American Journal of Sociology, 103, 4: 962-1023.

Epstein J. M., Axtell R. (1996) Growing Artificial Societies. Social Science from the Bottom-Up, MIT Press, Cambridge, Massachusetts.

Gigerenzer G. and Selten R. (2001) Rethinking Rationality, in Gigerenzer G. and Selten R. (Eds.), Bounded Rationality - The Adaptive Toolbox, MIT Press, Cambridge, Massachusetts.

Giddens A. (1984) The Constitution of Society. Outline of a Theory of Structuration, University of California Press, Berkeley and Los Angeles, 1986.

Growing Behavioral Attitudes… 53 Gilbert N. (1996) Holism, Individualism and Emergent Properties. An Approach from the Perspective of Simulation, in Hegselmann R., Mueller U. and Troitzsch K. G. (Eds.), Modelling and Simulation in the Social Sciences from the Philosophy of Science Point of View, Kluwer Academic Publishers, Dordrecht/Boston/London: 1-27.

Gilbert N. and Terna P. (2000) How to Build and Use Agent-Based Models in Social Sciences, Mind & Society, I, 1: 57-72.

Gilbert N., Troitzsch K. G. (1999) Simulation for the Social Scientists, Open University Press, Buckingham Philadelphia.

Handlbauer G. (2000) Decision-Making and Institutionalised Cognition, in Streit M.E., Mummert U. and Kiwit D. (Eds.) Cognition, Rationality, and Institutions, Springer Verlag, Berlin.

Hanneman R. (1995) Simulation Modeling and Theoretical Analysis in Sociology, Sociological Perspectives, vol. 38, n. 4: 457-462.

Hofstadter D. R. (1979) Gödel, Escher, Bach: An Eternal Golden Braid, Basic Books, New York, 1999.

Holland J. H. (1995) Hidden Order. How Adaptation Builds Complexity, Perseus Books, Cambridge, Massachusetts.

Lane D. (2001) Complexity and Local Interactions: Towards a Theory of Industrial Districts, Paper presented to the conference “Complexity and Industrial Districts”, Montedison Foundation, Milan, Italy, June 2001.

Lazerson M. H. and Lorenzoni G. (1999) The Firms that Feed Industrial Districts: A Return to the Italian Source, Industrial and Corporate Change, vol. 8, N. 2.

54 Growing Behavioral Attitudes… Lockhart R. S. and Craik F. I. M. (1990) Levels of processing: A retrospective commentary on a framework for memory research, Canadian Journal of Psychology, 44.

Luna F., Stefansson B. (Eds.) (2000) Economic Simulation in Swarm: Agent-Based Modelling and Object Oriented Programming, Kluwer Academic Publishers, Boston/Dordrecht/London.

Luna F., Perrone A. (Eds.) (2001) Agent-Based Methods in Economic and Finance: Simulations in Swarm, Kluwer Academic Publishers, Boston/Dordrecht/London.

Malsh T. (2001) Naming the Unnamable: Socionics or the Sociological Turn of/to Distributed Artificial Intelligence, Autonomous Agents and Multi-Agent Systems, 4: 155-186.

March J. G. (1994) A Primer on Decision Making, The Free Press, New York.

Minsky M. (1985) The Society of Mind, Simon & Schuster, New York, 1988.

Oerlemans L. A. G., Meuss M. T. H., and Boekema F. W. M. (2001) Innovation, Organizational and Spatial Embeddedness: An Explanation of Determinants and Effects, Eindhoven Center of Innovation Studies, Working Paper 01-03.

Sammarra A. and Biggiero L. (2001) Identity and Identification in Industrial Districts, Journal of Management and Governance, 5: 61-82.

Schelling T. C. (1978) Micromotives and Macrobehavior, W. W. Norton & Company, New York.

Simon H. A. (1987) Rationality in Psychology and Economics, in Hogarth R.M. and Reder M.W. (Eds.), Rational Choice, Chicago University Press, Chicago.

Growing Behavioral Attitudes… 55 Squazzoni F. and Boero R. (2002) Economic Performance, Inter-Firm Relations and Local Institutional Engineering in a Computational Prototype of Industrial Districts, Journal of Artificial Societies and Social Simulation, vol.5, No.1, .

Staber U. (2001) The Structure of Networks in Industrial Districts, International Journal of Urban and Regional Research, Vol.25, No.3: 537-552.

Terna P. (1998) Simulation Tools for Social Scientists: Building Agent-Based Models with Swarm, Journal of Artificial Societies and Social Simulation, vol.1, No.2, .

Torre A. and Gilly J.- P. (2000) On the Analytical Dimensions of Proximity Dynamics, Regional Studies, Vol.34, 2: 169-180.

Uzzi B. (1996) The Sources and Consequences of Embeddedness for the Economic Performance of Organizations: The Network Effect, American Sociological Review, Vol.61: 674-698.

Uzzi B. (1997) Social Structure and Competition in Inter-Firm Networks: The Paradox of Embeddedness, Administrative Science Quarterly, 42: 35-67.

56 Growing Behavioral Attitudes…