- Pierpaolo Andriani
- Giuseppina Passiante
Complexity theory, as a network/multilevel theory of emerging order is ideally suited to contribute to the understanding of economic systems. Industrial networks are the collective emergent property of non linear co-evolving patterns of knowledge based transactions. Whether geographic or virtual, concepts/instruments such as ‘autocatalysis’, ‘co-evolution’, ‘self organisation’, ‘dissipative structures’ and ‘Bénard-like behaviours’ are crucial to an understanding of networks dynamics.The seminar explored some of the managerial implications of the network paradigm.
- Passiante,G. and Andriani,P. Modelling The Learning Environment of Virtual Knowledge Networks, Some Empirical Evidence: Modeling_the_Learning_Environment.pdf
Compiled For The L.S.E.by Geoffrey J.C. Higgs
Table of Contents
- The complex system
- Developing interdependency model
- The ‘people as agents’ factor
- Boundary, what boundary?
- Living on the Net
The Belousov Zhabotinsky experiment is a good example in chemistry of how reactions can catalyse each other and give rise to patterns of order at a macro level which cannot be predicted by looking at the activity going on at the micro level. Such systems at first glance appear to go against the Second Law of Thermodynamics until it is appreciated that energy or materials must be taken in from outside the system in order to sustain it.
It is also true that however much we may plan and formulate strategies in business, economic activities, often arise which are not obvious. Chemical and business systems may be many orders of complexity apart, but they give rise to the same conclusion: order can emerge out of widely distributed and often apparently unconnected activity. Mainstream economics often neglects what might be called the constraints of history; that once some event has taken place it may be the ‘seed’ on which a set of cumulative processes build. In this seminar we shall try to understand how a complex economic system comes about and describe some of its interesting features.
The complex system
The fact that a number of different activities may be occurring in some particular place or time does not necessarily make for a complex system. We have to distinguish between a mere aggregate of things and a ‘system’. For example, a number of completely different paintings might be put together as a art collection but apart from the physical space they take up there might be no other relationship between them. On the other hand if they were all Impressionist paintings then they define a certain logical space. If you can see a meaningful set of actions between people or the interacting roles which people have, then these define an environment. Every agent within that system plays a part according to the environment which they bring about. So a system is a set of actions and reactions or a set of relationships that defines a whole. And this has a ‘top down’ effect. It has ‘bottom up’ emergence and ‘top down’ constraints. Classical complexity science sees such a system as a set of components which change but whose interactions are constrained by ‘rules’. What emerges is the result of this activity.
What we have to understand is how single ‘agents’ or groups emerge which in turn drive the whole system. We might take catalysis in chemistry as an analogy. A catalyst moves a system away from equilibrium but in a system in which cycles of reaction are set up in which products increase the availability of reactants, ‘positive feedback’ occurs and the process becomes non linear. The more output you get the more input you create and so on. It is a self catalysing reaction but it will run down if it doesn’t take energy or raw materials from its external environment. In the words of Prigogine it becomes ‘dissipative’; energy is taken in in some form by the system, utilised and subsequently lost from the system in another form. When auto-catalytic loops occur in a system and interact, the system not only defines a lot of ‘design space’ but it can be defined at a number of levels. At the component level, those that will be reproduced are selected and those that are not will disappear. In the chemical system those chemicals that are not catalysed will disappear and we can start to see the emergence of pattern or order in the way that Stuart Kauffman envisaged it. When a system becomes autocatalytic and dissipative it closes in on itself and becomes an organising force in that it selects at the micro level and expands the design space at the macro level.
If we think of social processes that give rise to systems it is the autocatalytic processes that define and filter the agents that come in, but once inside you have the ‘top down’ effect of the organising principle at the macro level. As soon as you have a relationship between agents ‘A’ and ‘B’ that somehow have reciprocal interactions with ‘C’ then the possibility of a complex system occurs. Order can be defined in terms of constraints. Constraints may limit the equal probability of events but can still be ‘context free’ . This needs some explaining. We can put a lot of chemicals together at a certain temperature and pressure etc. which limits possible events but puts no limit on those that might arise from a developing interdependency. We are together here today because we all have an interest in complexity but this does not define an interdependence. It’s rather like parts of a language. Letters go together in certain ways to form words. We find that certain letters are more probable in English words than others and that we wouldn’t find an English word starting with say ‘y’ followed by ‘xyz’. But that’s an example of ‘context free constraint’; it bears on the probability of use but it doesn’t limit what we can say in English.
In fact the more constraints you put on single letters in the way they can be combined together the more you expand the semantic possibilities of the words formed from them. There is a puzzle here. The more you limit freedom of agents at the micro level the more you seem to create the possibility of a complex system at a macro level. It’s an example of complex sensitive constraint. Constraining the agents creates a system of dependencies that defines the whole system. This ‘first order’ constraint gives rise to ‘Second order’ constraint which defines the boundary of the system and limits the way in which the agents can be.
But this top down aspect of complexity is often neglected. Complexity scientists often concentrate on characterising the agents and the set of interacting rules at the micro level but neglect the top down dimension. People often talk about ‘empowerment’ of people in the work-place and forget that the resulting whole system must in turn affect the way those individuals behave. If you trade your present knowledge for future collaboration with another organisation then you also set up rules by which the arrangement can be sustained. You can also alter one of the parameters under which the processes of the system occur. If you increase the temperature at which a complex chemical process occurs then you increase the rate at which it evolves. If you set employees certain time related goals then you can increase the pressure under which they work.
Once a system is created then it has a trajectory in its design space. If you’re a Christian you explore the ways in which you can live your life as one, if you’re a Muslim then that sets the stage on which you operate. In characterising a system we should look for the context free constraints such as geographical conditions. If we live in the Arctic we find certain kinds of fauna exist but that by itself doesn’t tell us the interdependence between the animals in that particular ecosystem.
Developing interdependency model
The ‘prisoners dilemma’ is an interesting example of ‘game’ development. In other words how interdependency comes about. Two criminals are captured by the police, charged with the same offence and put into separate cells. If they both confess they will get five years. If they both do not they will get eight years;. If however one agrees to testify against the other he will get say two years whilst the other gets eight. At the first stage of the game the optimum strategy for each prisoner is for them each to confess. But the game can be played over several times and the prisoners can learn from experience. In a community there are co-operators and defectors but how interdependency grows at any point depends on what has happened previously. These are context sensitive constraints.
Sometimes geographical conditions will cause the isolation of a community. But sometimes the isolation of a group occurs as the natural outcome of co-evolving agents around an evolved set of norms (rules). For example, in Bradford there is a textile business which consists of some 9 or 12000(?) small companies and there is a set of constraints or way of doing business that has grown up. There’s a lot of knowledge which is embedded and which is passively passed round and all these companies create certain patterns of operation around what we might regard as a system of catalysts. It’s the catalytic set that sustains the model. It’s the same in Silicon Valley. Engineers tend to gather round the projects and the leading edge technology. And perhaps the car manufacturer is a good example of how this can affect cost. You have the company selling the cars and they have a set of suppliers, so some fraction of the income goes to them but in turn these each have their own suppliers and so on. The bigger you are the smaller your fixed cost but economies of scale do not explain how wealth grows. The fraction of income that is transferred along the chain is a function of the size of the market and the length of the chain is related to the prosperity of a region. But there’s also a cumulative re-enforcing mechanism at work. When the export market reaches a certain size it suddenly takes off and the income of the region rockets. Of course it works the other way round as well. If the exportation goes below a certain level the industry can collapse as happened to a number of British and French silk producing areas in the nineteenth century.
So the sequence of events seems to be that you have an aggregate in which an organising principle arises which leads to the closure of an auto-catalytic set. The closure results in the evolution of the system by the interaction of its components which occurs in the presence of both internal and external pressures . At certain stages of development the system undergoes phase transitions. But what seems important to take account of in an economic region is the relationship between the income of the region an the market diversity within it. Krugman relates the size of the market to the ‘multiplier’ or how wealth increases in a non linear way but if there’s a set of re-enforcing mechanisms at work then relating x and y on the graph is only part of the analysis.
Kauffman said that ‘diversity begets diversity’. Companies tend to shed risk by ‘out-sourcing’ which is often a reaction against environmental uncertainty. Hollywood, for example, was extremly integrated in the 1940’s, but now individuals work for themselves or small companies. The process of vertical disintegration increases the number of components in the system but you still need to increase the rate of integration between agents by say focusing on the technology. If what was inside the company needs to go outside the company you need to manage the extent of the resources and interact more with the suppliers. Interaction at the micro-level starts internalisation within the region creating an economic ecosystem. In a biological ecosystem species competing in a territory for the same resources show a high rate of co-evolution. In an economic system this means that organisations compete on technological innovation. They don’t compete on price. This is a kind of niche suppression effect which increases the variety within a technological paradigm. If you decentralise production you need to control the process of innovation and here the information network becomes paramount. The more you can exchange existing knowledge the more you tend to increase production. However centralisation can also increase production if you put the components of the system together in a different way.
Innovation may be co-ordinated by a particular agents in a system but a company’s life involves an ongoing strategy which should be geared to evolving in the economic climate. As an organisation grows there is an increasing move towards specialisation of function which means increasing variety and ideally increasing interactivity. If innovation is radical in that you invent a new paradigm you have to be sure you have a means of co-ordinating resources which are usually dictated by the technology and the geography of a region. In a geographic cluster people know each other and interact socially and this results in emerging co-ordination at the macro level. The Kauffman model where S is the number of agencies and C is the number of ways of interacting is relevant here. It may be possible to fine tune S and C to an optimum. If agencies are competing for the same resources then they move onto something else the way species do in nature. So for example if you have 40,000 developers all over the world working on software for the Linux organisation you can expect rapid developments.
The influence of rules is an area of complexity that has still to be fully explored. Ant colonies provide some useful parallels. Ants set out from the nest in a random way, find food and take it back to the nest. We may find decreasing pheromone gradients and can suggest simple rules at the level of interacting agents. We can view the situation as pseudo-rational in that it seems as if there is a conductor telling every individual ant what to do. And though we often model the ‘bottom up’ traditional complexity view we find it difficult to model the ‘top down’. What we cannot yet grasp is how the simple rules between the interacting agents give rise to the apparent purposeness at the higher level. If we build a business model we define behaviour nodes (agencies) and how they compete, but it’s difficult to describe how the model influences the agents at the micro level.
If you have an aggregate it might evolve but it won’t necessarily co-evolve. On the other hand even if you have segregation of co-operatives at the setting up stage, as soon as one agency starts co-operating with another you get a primitive closure. Systems may remain ‘open’ in that they do not develop interdependencies though such a description would be dependent on the level of explanation. ‘Closed’ systems are by this definition co-evolutionary ones. We might set up an organisation by thinking about positioning and basing a strategy on it. In doing so we define what we are and we supposedly position ourselves against an environment which is decoupled from us. This is Michael Porter’s kind of strategy but it doesn’t work because you can’t define your boundary. The environment is really nothing more than a set of constraints which affect the behaviour of the components s in the system. In fact we can expand the concept of an ‘open’ system to one which has some islands of correlation and some of chaos in which some evolution and co-evolution occurs.
The ‘people as agents’ factor
Applying complexity theory to physical systems which involve a large number of components is one thing. Applying it to people organisations is another. Nature has it’s own rules which though operative are not our concern when talking about people. There seems to be a constant danger in drawing analogies between the two which goes to the heart of the relationship between social and natural sciences. It is impossible to distinguish between them in a rigid way and philosophy informs both. We use the same principles and it is disconcerting to not be able to define the boundaries of what natural science is and what social science is. On the other hand one of the goals of complexity is to demonstrate that there are laws that don’t depend on what kind of systems are under investigation and apply equally to social systems and natural systems. If you take a vase and smash it against a wall, collect all the fragments and count the number of fragments at particular sizes you get the typical random distribution curve. This configuration is the same as that obtained by taking the number of towns in the U.S. or Britain or France with a particular number of people. The same applies to the number of earthquakes at a certain size or the number of extinction involving a certain number of species. There appears to be something behind the technical factors and the specificity of the systems that behaves in a certain way. It’s across social and natural systems and nobody knows why.
It is difficult to draw parallels between physical systems and human organisations. Human organisations in the past have tended to be hierarchical and the number of component parts much more limited though the rules governing the interactions between individuals or agencies much more complicated. One might ask the question ‘ why should we expect to see emergent properties arising from auto-catalytic processes in social economic systems? In the case the vase there are clearly observable results of the fracturing process. It starts off as an integrated whole which becomes fragmented and the challenge is to find some sort of organising principle in the fracturing process that tells you why it behaves in the way that it does and why the particular distribution of particles occurs. In the case of a socio-economic system with people and decision makers it could be asked what warrants us to take such a reductionist approach? The fact that we might get beautiful mathematical diagrams is great but what entitles us to define the regimes and autocatalytic processes that we imagine are there?
Boundary, what boundary?
Classical reductionism defines a system’s boundary and identifies the components within it. With a socio-economic system we can mention the properties but defining the boundary is difficult. In reality each of us are mutually interacting auto-catalytic sets which are self organising and this makes defining the boundary difficult. Saying what a system’s boundary is, is a very subjective interpretation where human agents are involved. Moreover a social system has different configurations in terms of the number of effective participating components. The same emergent properties can be observed with varying numbers of people. The critical density for phase change may vary widely. We are still a long way from understanding a relationship. It ultimately raises the possibility that recognising it is a cognitive process and it begs the question of what is a simple thing that we can recognise and what isn’t?
Of course now people understand that their personal interaction with others contributes to the whole system in a way that they didn’t fully appreciate before. Twenty years ago people tended to think that they were just improving their own position and behaving like the rest of the individuals. People didn’t appreciate that there is a boundary around their behaviour patterns. Of course there is a difference between knowing that you are part of a system and being told you are part of a system. The very act of telling people they are an integral part of a system may cause a system to change. Social systems are reflexive whereas natural systems are not reflexive in quite the same way. Whereas physical systems can be explained perhaps ultimately in terms of simple rules, people react and anticipate. Of course things go on which we are not aware of and sometimes we might be aware of things in an intuitive way and be reflexive as a result of that. A lot of diversity is exchanged at a passive level of interaction.
Living on the Net
Billions of transactions take place on the Internet in terms of business to business, business to supplier and business to client interactions. Services to those who use the Net are constantly improving in terms of cost, time and quality of product. Whereas in the past business models tended to be based on financial transactions we now have an information based model. In the shares market, for example, there is an increasing number of consultants or IFA’s who help their clients choose the funds and company shares in which to put their money. This enormous flow of information creates clusters of shareholders which change rapidly in makeup. The same applies to other fields but in other ways. There is no reason why a large organisation like a Utility should be fully integrated by having all its functions ‘ in house’. The glue that holds components together is informational asymmetry and if contracting outside is cheaper than maintaining different departments then there’s no need for such integration. So there is often vertical disintegration dependent on function and new companies emerge around all these functions.
A ‘virtual cluster’ is a network of interrelated explicitly or implicitly linked businesses. Each business can be seen as an agent or ‘actor’. On the ‘Net’ there are many alliances of businesses that make up virtual clusters. For example Java started through the co-operation of four main actors. Large actors can become part of different clusters in which there is co-operation for a particular idea. Microsoft, for example is in a number of virtual clusters. In some cases it is the technological partner, in others it is one of the supporters of the language, whilst in others it is involved in collecting information about the community. In each the rules are different. Linux used an infrastructure that was already present. They didn’t buy or invest or go into partnership but used the Internet as cyberspace in which to inform the community. As a company building an extranet it simply used the technology that was already there. On the other hand if the technology is the object of the enterprise then you have to invest money in it. In joining a cluster you need to know the business and culture of the cluster. What identity it has and what its constraints are. In any cluster there may be some actors who concentrate on increasing efficiency, who will set targets. Other actors focus on improving the business by change. They seek to change the context or the rules by which people operate. Some idea of how this kind of positive change can be brought about can be understood by considering ‘Bénard’s cells’. This is a physical model consisting of a film of liquid between two plates of glass. You control the temperature difference between the two plates and convection currents are set up. By modifying the constraints on the system the patterns of convection currents can be moved from chaotic to ordered and back to chaotic again. By adjusting one of the parameters such as the temperature the system can be induced to assume different states. One of the states, between stable pattern and chaos shows different emerging patterns of various order. This realm or state is referred to as ‘the edge of chaos’.
In searching for a definition of a virtual cluster it may be useful to look at the National Health Service. G.P.s do not work for the N.H.S. as employees but are under independent contracts. The patient is the focus of everything that the N.H.S. does and there is a lot of emphasis now being placed on clinical governance. There is a National Commission for Clinical Excellence etc. The drive over many years has been to push the boundaries of care and the Internet can be used for the rapid exchange of information. But the emphasis is on establishing an extended community, involving local government in primary care, acute care, drug companies and so on. The aim is the excellent use of resources which is an exciting prospect. There are a lot of divisions in terms of professions and the way they position themselves and the way the care is delivered. We could view this as a virtual cluster but how do we view the power of government to actually initiate it? The previous government created a free market that didn’t bring it about. And in fact a lot of joined up thinking and collaboration was needed to bring it about. The ‘top down’ effect is very much in evidence here in that the macro structures and patterns influence the behaviour of individuals in the system. Some of the groupings like the Clinical Governance will be driven by a commission and this will influence the role of the G.P. and of course new regulations will hopefully expose situations like the ‘Shipman’ case quickly.
People bring problems to the network for which they want consultancy. The more consultants you bring in the large the consultancy. The flexibility, speed and to some extent the anonymity of the Internet gives rise to new ways of working and it may qualitatively change the kind of help that you can give to clients and their problems and the range of problems that can be addressed. Changing rules of contact constitutes an emergent phenomenon that raises the quality of solution to a higher level.
Clusters on the Internet are independent of physical space. But whilst it may be relatively easy for a medical consultant to join a cluster, businesses concerned with production of goods or technological innovation may have to make quite dramatic changes in terms of investment for transformation supply and order processing. The consolidation of a cluster may take some time based on the decision making of prospective participants. There is always some risk in terms of allowing other partners in the cluster free access to your operation and some large companies decide not to participate in a virtual cluster but to optimise their operation by influencing the role of existing partners.
This is outside the technical definition of a virtual cluster on the Network since the participants are not free to become part of another cluster. The large company is imposing the constraints and standards. One of the perhaps less obvious hazards of virtual clusters based on the Internet is that change can be so fast that participants cannot cope with it and a kind of paralysis occurs. Staff then become demotivated and go to other organisations which they view as being more stable. People as the actors in a system have their limitations. When the system changes faster than the participants can cope with it then chaos ensues. We need a means by which we can understand when we lose a sense of coherence. We need to know at what point we’re like Alice, running as fast as we can merely to stay in the same place and what would happen if we stopped. we need to know whether we should try to patch something up or whether we should abandon it.
The Internet is a new communication tool. When telegraphy was founded you had new communities that used this new channel of communication to redefine the boundary of geographical community. A virtual cluster is only an extension or analogy of a geographical cluster. But some groups of individuals like say the farmers are a consolidation but not a virtual cluster in the sense we have been considering. They don’t learn a great deal from the collectivity because the interaction is limited. The G.P.s on the other hand need to constantly spread ‘best practice’. When governments put regulatory constraints in place they push the system towards more co-ordination but this can also have a negative effect by increasing inflexibility. Access to knowledge is a very important factor. No individual or community holds the knowledge that’s on the Internet; it’s dispersed in the nodes and the relationship of the system. The knowledge base in a company moves from the top of the system at a macro level to the micro level of the individual. Moreover individuals or groups only ever hold only part of the available knowledge. The virtual cluster like the geographic one produces a final product or service as the result of its whole dynamic.
Companies entering the net have to adapt and they have to evolve. There are four key words which sum up the important characteristics of this process: improvisation, flexible structure, real time communication (question/response time) and co-operation. In order to cooperate within a cluster you have to know the focus or target of it. Each participant or actor has to play a particular role. You have to know your role and the role of others at a particular time of operation. Evolving towards the future requires ‘regeneration’ in that you have to blend the best part of the past into the future and ‘variety’ which means experimenting with different solutions. The learning process must be developed at all levels.
The drive for each enterprise is the desire to satisfy the client. If an enterprise asks itself ‘how can we bring to bear services that will satisfy the clients needs’ it opens up an even larger requirement for diverse knowledge. Along with the exchange of explicit knowledge there is a vast exchange of implicit knowledge and though supply chain management still plays an important part, we are almost in the era of the one to one market with each provider trying to customise the product or service for a single client.
We are also moving away from company dominated careers to the situation where individuals have to be responsible for their own and the work is project centred instead of corporate. Nevertheless it is still necessary to provide the right framework of support for individuals in an organisation and if change is necessary it should lead to individuals being healthier and happier.
Business organisations in the past tended to have very rigid managerial structures which were hierarchical; the ‘brains’ of the business at the top and the managers and workers underneath. This kind of organisation is inadequate in dealing with the fast changing world of Internet business. What an actor responsible for increasing efficiency must do is to change a central intelligence structure to a more distributed one. You can do this by changing the rules or constraints by which the company operates. McKelvey talks about ‘adaptive tension’ which leads to new forms of structure. The goal of a CEO is not to tell the organisation what to do but to set the constraints, to create the adaptive tension, to clarify a vision and to set a target. If for example, you change the physical environment of the office you change the patterns of staff interaction. If on the Internet cluster you set knew targets and specify different ways in which information is to be obtained and sent then you alter the constraints by which the business operates.
In the past it was possible to plan by focusing on the functions or talents of individuals.Now you have to think as much as possible in terms of what the cluster offers and it may change constantly in terms of makeup or individual workers. Planning is therefore an ongoing process. We have to respond to the changing environment. The actions that we carry out today may be different tomorrow and because of the interconnectivity with others we cannot say ‘in the future I will not do this or that’. The ‘top down’ effect impinges much on the style of management required. Decisions are still influenced by finance, or by the decisions by mainstream clients but optimisation can be achieved in several different ways. If the number of constraints on a system is too high to build a realistic model then Kauffman’s idea of dividing the system into ‘patches’ may help. You divide an area into units around which there is a perceived boundary and you tell its agents to organise their own working process regardless of what is happening in the rest of the organisation. This leads to self organisation in which the number of constraints is much lower and this increases the amount of diversity in the system which improves the fitness. The logic is the same in dealing with a state which is too centralised or federalistic. You let communities decide on the basis of their own fitness irrespective of the fitness of the whole. Strategy is more concerned with trying to identify the little ‘butterflies’ of today that will transform the business in the future.