Brian J. Loasby
Time, knowledge and evolutionary dynamics: why connections matter *
Journal of Evolutionary Economics
Vol. 11, 2001, 393-412
Time matters because knowledge changes. Knightian uncertainty excludes correct procedures and proven knowledge, but makes room for imagination and creativity, which drive an evolutionary process. Human cognition relies less on logic than on pattern-making; we impose connecting principles to create patterns and causal linkages between them as representations of phenomena, which are imperfect and often subject to multiple interpretations. Stable patterns provide the necessary baseline for selection. Our personal patterns are supplemented by institutional regularities, and organisations of various kinds help to shape the development of knowledge, which grows by making connections at the various margins of existing knowledge.
Economic theory has concerned itself with the sources and consequences of conduct, and has sought in this field what can be conceived as rational, what can be expressed as proportion, what is publicly and unanimously agreed, and what belongs within bounds defined by the notion of exchange in an inclusive sense.
The attractions of such a programme are evident and compelling. The cost resides in what, by its nature, it is obliged to neglect or even implicitly [to declare unimportant... the most serious of those exclusions… is the brushing aside of the question, a unity though requiring three tells to express it, of time, knowledge, and novelty… theory has chosen rationality, whole and unimpaired. Arid thus it has cut itself off from the most ascendant arid superb of human faculties. Imagination, the source of novelty, the basis of men’s claim, if they have one, to be makers and not mere executants of history, is exempted by its nature from the governance of given and delimited premises. Shackle 1972, pp. 443-444]
HHC: [bracketed] displayed on p.394 of original.
* This is a substantially amended version of a paper first presented at a meeting of the Network of Industrial Economists at the University of Reading on 19 December 2000 and then in a modified form at the DRUID Winter Conference at Klarskovgaard, Korsør, 18-20 January 2001. I am grateful to John Cantwell for providing the initial stimulus, to John Sutton for encouragement, and to participants at both meetings for their comments; I have also benefitted from interchanges with Uwe Cantner and Jason Potts. All share the credit for those parts of this paper that they agree with.
In Value and Capital, Hicks (1948, p. 115) defined “[e]conomic dynamics [as] those parts [of theory] where every quantity must be dated”. Subsequent theoretical development has shown that this is not a sufficient criterion. In the Arrow-Debreu system not only must every quantity be dated, it must also be indexed by location and state of the world; yet in a model that conforms to these specifications there is no room for dynamics, but a single equilibrium which extends over all dates, locations and contingencies. There is no arrow of time: later dates influence earlier allocations in precisely the same way as earlier dates influence later allocations; there is no sense in which one thing can lead into another. In terms of Hicks’ ( 1982b) distinction, time is incorporated into the model as an additional dimension, but the model is not ‘in time’, which would imply the need for a sequence of decisions by economic agents. Indeed, within such a model there is no scope for even a single set of decisions: an equilibrium allocation is deduced directly from the basic data, which includes a complete set of preferences but requires no algorithms of choice. The individual is of measure zero (Hahn,  1984, p. 64), not only because a single person’s preferences and endowments have no perceptible effect but because no individual is allowed to take any initiative. Everything that could possibly happen must be incorporated in the specification of one or more states of the world for each date and location; the occurrence of any novelty, either endogenous or exogenous, violates this requirement and demonstrates that the apparent equilibrium had been derived from false premises.
It is as well that all differences between dates are incorporated in a single equilibrium, because no resources are available to cope with change, having already been optimally allocated within that equilibrium. There are not even any resources available to cope with equilibration; for although the model purports to include all resources it ignores the mental resources required for human action. Its internal consistency therefore requires the exclusion of human agency, and so the process of achieving the declared equilibriumn cannot occur within a functioning Arrow-Debreu economy. As good general equilibrium theorists do not hesitate to point out, the markets - strictly a single market - in which this equilibrium allocation is represented by a complete set of contracts open once only, and close
before the economy starts operating. It is not surprising that this looks less like a model of a market system than a model of a command economy.
The requirement that all transactions are arranged outside the economy in order to exclude transaction costs from the model has significant implications. As Coase (1988, p. 15) pointed out, if there are no transaction costs there can be no problems with externalities; impacts on third parties, whether beneficial or harmful, should simply be added to the list of goods and to the preference systems of those affected by these impacts. Thus there can be no unexploited gains from trade - indeed there are no third parties - and therefore no scope for beneficial policy interventions to remedy system failures. If it is nevertheless thought appropriate for particular theoretical purposes to incorporate costs of transacting, then it is illegitimate to compute equilibria separately from the analysis of the transacting process. Moreover, these costs depend on the ways in which transactions are organised and the sequence in which agents search for the best attainable set.
It is also likely that some transactions, especially for dates at which there are many possible states of the world, will be postponed, and it may seem attractive to commit some resources to the development of systems for making and implementing later decisions. This, of course, is the basis of Coase’s explanation for the firm; it is also the basis for an explanation of markets as institutional arrangements for facilitating a series of transactions (Ménard, 1995, p. 170), and for the creation of various kinds of reserves, as Menger ( 1976) observed, in the form both of goods and of capabilities, direct arid indirect. However, none of these phenomena can be accommodated within the Arrow-Debreu system, for this system cannot accommodate the concept of change within (relatively) stable structures which make change possible - because of the particular conception of a system that underlies it, as we shall see.
If we wish to incorporate the process and costs of change within our models we need to modify our use of dates: they are now required not simply to ensure that the set of variables is complete but in order to identify the sequence in which things happen and in which knowledge and possibilities become available. This was precisely what Hicks intended to signal by his definition; in fact, as Leijonhufvd (2000, p. 111, n. 22) suggests, what matters for dynamic analysis is that decisions must be dated - because knowledge must be dated. But, as has been demonstrated in many multi-stage models, this may not make much difference to the analytical strategy, as long as it is assumed that agents correctly, if incompletely, anticipate future knowledge and future possibilities and make correct deductions from their anticipations. Hicks refused to make this assumption: the dynamics of Value and Capital are provided by a sequence of temporary equilibria, set within an unanalysed intertemporal disequilibrium - which incidentally provided Hicks with the basis for a sympathetic interpretation of Keynes’s General Theory. It is an economy in which time matters, because time changes the knowledge that is available to agents, and not only through Bayesian updating, even though time matters less in Value and Capital than in Hicks’s later writings (see especially Hicks,  1982). Like the discoverers of intertemporal equi-
librium, Lindahl arid Hayek, Hicks rejected the assumption of perfect foresight, even when extended to probability distributions, as an obstacle to modelling the working of an actual economy (Zappia, 2001), and Myrdal, Hayek, Keynes and Schumpeter all based their theories of the business cycle and unemployment on various kinds of imperfect knowledge (Loasby, 1998).
It was Frank Knight (1921) who first emphasised the crucial distinction between risky situations, in which there is an agreed procedure, logical or empirical, for distributing probabilities over a closed set of outcomes, and situations of uncertainty, in which there is no such procedure - and often, as Shackle was to insist, no way to ensure that all possible outcomes have been recognised. Knight pointed out that risk, as he defined it, was a calculable cost and that, since the correct method of calculating each risk was public knowledge, risk-bearing was a productive service not sharply distinguishable from any other, and therefore not a distinctive source of income. ‘Profit’ is then a misleading name for a not-very-particular kind of wage, and risk-bearing belongs in the production function, and therefore poses no threat to the homogeneity that is required for perfect competition. But since uncertainty, as defined by Knight, excludes the possibility of any method which can be shown to be correct, uncertainty-bearing cannot be treated as an input in a production function and the homogeneity of perfect competition cannot be preserved. Profit is the reward, not for risk-bearing, but for successful entrepreneurship which has coped with uncertainty by idiosyncratic means: entrepreneurs might calculate, but the correct basis for their calculations could not be deduced from the basic data - and many of them fail.
However, although the result of any attempt to cope with uncertainty must itself be uncertain, Knight did not believe that success was purely a matter of chance; nor was it simply the consequence of alertness to opportunities which, once perceived, are clearly genuine - which is the basic case of Kirzner’s (1973) theory of entrepreneurship. It was a reflection of human capabilities, and a distinctive and valuable resource. In his first published article, George Richardson (1953) drew on Knight’s analysis of the significance for economic efficiency of the deployment of distinctive entrepreneurial capabilities to argue for the importance of selection among agents and, as can now be seen, was only a thought away from the argument of his famous article of 1972.
But uncertainty prevents the closure which is essential to achieve proofs of equilibrium, and so in equilibrium theory it must be reduced to risk. Nor can modern expedients for deriving equilibria when agents have imperfect or asymmetric information claim validity unless it can be assumed that every agent draws not only from the same information set but also, though this is scarcely ever recognised, from a single system for interpreting this information.
Knight’s distinction is fundamental. It is a distinction between conceptions of knowledge, which (as so often) was best expressed by George Shackle (1972, Preface). “The economic analyst has opted for reason. He assumes that men pursue their interest by applying reason to their circumstances. And he does not ask how they know what these circumstances are”. In particular, rational choice relies on a comparison of the consequences of all available alternatives, without
deigning to explain how these consequences can be known, or even how all the alternatives can be known to be available. If we wish to do economics in the spirit of Knight and Shackle, we must do it in another way: we must switch our emphasis from closed to open systems, and from proofs to process. However, if economists in really substantial numbers are to be persuaded to change, it is important to demonstrate both that this other way is accessible, and also that it is clearly more appropriate for dealing with some issues that are widely recognised by economists to be important, such as economic growth, technological change, the scope and role of the firm, the generation of novelty, and even the co-ordination of economic activities.
Knowledge, Institutions and Evolution in Economics (Loasby, 1999) was intended to show that such a way exists, that it is already applied in other fields, that it has distinctive and significant applications within economics, and, not least, that it is already a substantial part of our heritage as economists. Economic dynamics is founded on uncertainty, because it is economics in time: this Hicks (1982a, p. 34) had begun to recognise by 1933, and this recognition grew in strength throughout his life. But though uncertainty gives rise to serious problems, not only for economic theorists but also in the conduct of economic activities at all levels from individual decision-making to the co-ordination of economic systems, it also provides abundant opportunities. For as Shackle above all continually reminded us, uncertainty is the precondition of imagination and creativity: it makes space for the growth of both theoretical and practical knowledge.
However, it also ensures that this growth must be evolutionary, because it is the result of trial and error, and the rate and directions of growth are influenced by how these processes of trial and error are organised. As Marshall (1920, p. 138) told us (but as most economists have forgotten), organisation aids knowledge, and it has many forms, each with its own particular virtues and limitations. The past cannot be changed, but it can, in part, be known; the future cannot be known, but it can be imagined, and by acting on that imagination it can, in part, be changed. Imagination is shaped - though not determined - by the interpretation of environment and experience. However, most of what is imagined turns out to be impossible; and so progress depends on both the variety of imagination and some process for selection among this variety - the essentials of evolution. It also depends on stability as a background to change.
The method of my book is to make connections, and its epistemic basis is the conception of human cognition as a connective process, in which the connections forged by logical argument are important but not primary, either in evolutionary terms or when ‘acting for good reasons’ - for neither decision premises nor the framing of problems result from logical processes. In relatively modern times this limitation of logic was established by David Hume ([1739-40] 1978): “no kind of reasoning can give rise to a new idea” (p. 164), and “reason alone can never produce any action” (p. 414). Hume also demonstrated that no amount of evidence could provide definitive proof of any general empirical proposition - a demonstration that inspired Popper’s evolutionary conception of science as a never-ending sequence of conjectures which are exposed to possible falsification.
Hume’s friendship with Adam Smith provides a foundational connection which is not explored in the book (but see Loasby, 2002): however prominence is given to the expositions by Smith, Marshall and Hayek of the theme that knowledge grows through a fallible process of making connections. The starting point is not a conception of perfect knowledge, from which one moves to risk and then to ‘uncertainty’ which is regularly assimilated to risk, but the need to construct knowledge by creating categories and imagining links between them. These two aspects of knowledge construction are precisely identified by Adam Smith ( 1980) as the ‘connecting principles’ which constitute scientific explanation.
An unintended but very welcome consequence of this method and content is that they provided me with the specific absorptive capacity to appreciate both the force and the value of Jason Potts’s (2000) book: we rely on similar ‘connecting principles’ in constructing our arguments. I believe that our books are closely complementary; reading either is a good preparation for reading the other. Which is the better sequence for a particular reader depends on the intellectual position and problem orientation from which that reader starts: this is in fact a proposition derivable from either book.
In this section I will attempt to exploit this close complementarity by adopting (and adapting) Potts’s meta-theoretical perspective on the relationship between equilibrium theorising, as generally practised, and dynamic analysis. The Arrow-Debreu system is very carefully located in integral space, where every element is directly connected to every other element, just as the Newtonian model of the solar system exists in a unified gravitational field. In the Newtonian system larger masses have bigger effects, and in the Arrow-Debreu system stronger preferences have bigger effects, but whatever the magnitude of the effect in either system it impacts directly on every other element in that system. There are no gravitational shields or specialised intermediaries to constrain interactions; ‘markets’, which provide connections that are indirect, have - and can have - no existence except as a metaphor for direct and costless transfers of property rights, and all ‘choices’ are transparent. There is no structure. Menger ( 1976), by contrast, set out to explain the structure of prices, and Hicks ( 1982b, p. 287) suggests that his condemnation of Böhmn-Bawerk’s theory as “one of the greatest errors ever committed” was a response to the debasement of time in that theory from a context for structure, as Menger had used it, to a measure of capital intensity. Menger’s is a self-organising system, which relies on the development of specific connections; it is conceptually distinct from general equilibrium.
Although the Arrow-Debreu model is no longer generally regarded as the central model of economics, nevertheless the widely-adopted principle that outcomes may be directly deduced from the data relies on the same integral conception. George Richardson (1959, p. 24) long ago pointed out that in the world of practice there is no direct link between data and outcomes, but only an indirect
link by way of beliefs and intentions; whether consciously or not, economists have recognised Richardson’s observation as a threat to the concept of a fully-connected economic system and the theoretical technique that relies on it, and have either ignored the issue or produced some notably unrigorous stories to support their practice. The rhetorical purpose of invoking rational expectations is to justify this procedure by rendering illegitimate any inquiry into the ways in which data are interpreted - thereby impoverishing business cycle theory, as well as the theory of economic growth.
Since a fully-connected equilibrium is a completed project, it is closed to further enquiry (though one may compare the equilibria that correspond to different data). However, by assuming that ‘in the beginning there was a fully-connected system’ it is possible to generate apparently well-defined analytical problems by postulating that some carefully-chosen connection is missing from a set that is otherwise complete; it is then possible to derive a local equilibrium incorporating agents’ reactions to this solitary deficiency, relying on the results of the fully-connected model to absorb that local equilibrium, and ignoring or finessing the once-powerful argument that only a complete general equilibrium analysis ensures validity. This reliance is often implicit, and sometimes appears to be unconscious.
The identification of a strictly-limited deficiency in an otherwise fully-connected system is the standard method of generating soluble problems in economic theory. There are two variants of this method: in what might loosely be called the ‘Harvard’ version the deficiency results in some welfare-reducing failure which creates a space for government intervention, while the ‘Chicago’ version demonstrates that the result is a new equilibrium in which economic agents reduce the damage to negligible proportions, whereas governments, as ‘Chicago’ economists know, can be relied on to make things much worse. ‘Harvard-style’ analysis is illustrated by ‘New Keynesians’ who produce a caricature of Keynes’s results which accepts the internal validity of new classical reasoning but makes a case for government action. An appropriate example of ‘Chicago-style’ reasoning in the context of this paper is Oliver Hart’s (1996) explanation of the firm as an optimal allocation of property rights, which may be briefly examined.
Here the problem-generating deficiency is a narrowly-specified constraint on the feasible contracting space, which is sufficient to frustrate the contractual alignment of incentives but has no other implications. The consequences, and the narrow scope, of this deficiency are so clearly defined that farsighted contracting is possible about the right to make decisions, which Hart identifies with ownership; thus the missing connection can be restored by an appropriate allocation of property rights. Since, by virtue of the background general equilibrium model, such an allocation is Pareto superior - the reduction of variety in decision-making is clearly beneficial when the correct decision is readily defined - there must be a set of contracts which make it universally acceptable; there are, of course, no obstacles to efficient contracts for property rights. The result is an analysis that combines theoretical novelty with the apparent validation of a general equilibrium theory in which allocations are derived directly from the data,
without postulating any interaction between agents. By virtue of its construction, there are no dynamics in this model; the appropriate allocation restores the connection between data and outcomes, allowing all dates and contingencies to be provided for. The standard analysis of production remains untouched. Thus the ‘firm’ which the model purports to explain is just an extension of the ‘market’; neither has any organisational or institutional existence as a particular set of connections.
Oliver Williamson’s approach looks more promising. Although he differs from Coase in insisting that the fear of opportunistic behaviour is a necessary condition for the existence of a firm, he follows Coase, and differs from Hart, in modelling the firm as a system of resource allocation by direction. This immediately suggests the possibility of a theory of organisational development, which may lead to changes over time both in the way that the firm is organised and managed and in the scope of its activities. Unlike Hart’s model, the Coase-Williamson conception implicitly defines the firm as a network of privileged connections, leading naturally to Herbert Simon’s vision of an economy in which firms, not markets, are the primary forces. Williamson, however, appears never to have appreciated the fundamental significance of this conception, for he has denied the validity of Simon’s vision (Williamson, 1996, p. 145) and has never shown much interest in what firms actually do - which is to develop and use connections, of many kinds as we shall see. In Williamson’s explanation of the firm, no less than in Hart’s, structure matters only as a means of validating the underlying theory in which structure has no role. It is therefore no accident that he has never analysed the development of firms overtime, despite Nooteboom’s (1992, p. 285) observation that his theory seems to demand a time-dimension. In terms of Hicks’ ( 1982b) distinction, time is incorporated into the model in order to differentiate the choice of governance structures from the choices that are made within the chosen structure, but the model is not itself in time: that is precisely why there are no “surprise[s], victims and the like” (Williamson, 1996, p. 46).
We have seen that the basic model of standard economics ignores connections because they do not affect outcomes, but allows for a variety of special models to explore the implications of particular deficiencies; the term ‘market failure’ is a clear, if rarely recognised, indication that this is what is going on. The idea that connections are problematic in general, and should be treated as problematic, is not seriously entertained. But that is precisely what is required in evolutionary economics and in industrial dynamics - and in other less orthodox branches of economics, as Potts explains.
The obvious objection to treating connections as problematic is that whereas there is necessarily only one way in which a system can be fully connected, there are very many ways in which it may be partly connected; how then are we to know what connections to include in our model? Before we can attempt to answer that question, we should explore the fact that underlies it: the recognition that the system and our model are necessarily different. If all theoretical discussion is in terms of fully connected systems, or integral space, then it is rather easy to assume
isomorphism between the system and the theoretical model; but when theories are recognised to be simplifications then they must embody partial connections - even when the model is of a general equilibrium. Our theories, our classifications, and our ideas are not simply derivatives of reality; they exist in the space of representations, and (if we conceive them to have physical form) in the neural networks of our brain. We cannot start with a complex reality, and choose how to simplify it by removing some connections: that is a cognitive impossibility. Instead, knowledge has to be constructed by building up connections. Knightian uncertainty, though habitually treated as a rare and unimportant phenomenon, is actually the base case - and it is the basis of human cognition and human society, not only as problem but also as opportunity.
Evolutionary biology and psychology explain why human cognition should have developed, not as a general-purpose instrument for solving problems by identifying their logical structure, but as a loosely-connected cluster of context-limited categories and linkages - Smith’s connecting principles. However, the use of these cognitive skills, which are the basis of imagination, take us beyond the random mutations arid natural selection of the biological model of evolution to purposeful (but fallible) behaviour (Penrose, 1952). Wherever we start there are, in principle, very many directions in which we may look for connections (Simon, 1992, p. 21), and each move opens up a new set of possibilities; but each individual is likely to notice only a few. It therefore seems reasonable to suggest that the best way to improve knowledge is to encourage many people to imagine connections, and to try to arrange that different people will imagine different connections. The latter, of course, is the function of the division of labour, and it was Adam Smith who realised its fundamental importance - and incidentally generated the co-ordination problem which has now supposedly been solved by denying its origins. It is also necessary to have some means of deciding which products of the imagination should be preserved and developed and which discarded or amended. This is a major role of markets, but it is not the only role of markets, as I have argued previously, nor is it only markets that can do the job.
All our theoretical systems are constructed in the space of representations; but many of them are presumed to have real-world relevance. But this relevance is not ensured by applying criteria deemed appropriate to the space of representations, although this is often done - and not only in economics. There is necessarily a gap between representation and reality; and preoccupation with criteria which are internal to the representation may widen that gap. For example, refining the internal coherence of the concept of rational choice has driven choice theory away from the practice of decision making (Loasby, 1976); and the isomorphism between planning and perfect competition as represented in models of general equilibrium was a sign that these models were inappropriate for understanding either planned or market economies, because the performance of either rests on their specific structures, or patterns of connections (Richardson, 1960). Charles Suckling conceived the task of managing innovation as a careful exploration of the gap between the initial representation (typically an interesting
effect in the laboratory but sometimes a theoretical result) and the much more complex environment in which commercial success would be determined. (This is a representative example of the reasons why I dedicated my book to him.)
The problem of representation arises in many forms. How are we to know what connections to emphasise in the design of any particular organisation - for every organisation chart defines a partly connected system? How are we to know what connections should be made with other organisations? What connections are most effective in gaining customers? Who are the potential competitors? What are the factors we should take into account when designing a regulatory system for a privatised industry - and is the answer different for each industry? As economists, in what aspects of what other disciplines should we take an interest? What are the significant connections that are missing from our models of purportedly fully-connected systems? We may also ask when connections are best avoided, for example by assigning activities to distinct organisations or problems to distinct theoretical systems. A non-economist would assume, reasonably but wrongly, that the principles governing the separation of activities would be a central theme in explaining the organisation of industry; and the desirability of differentiating theoretical systems is a key question for the analysis of change, as is illustrated by the insistence of both Schumpeter (1934) and Penrose (1959) on disconnecting the theory of growth from allocative theory.
Fundamentally, every organisation, every theory, every set of expectations, every plan, and every policy privileges a very small subset of possible relationships; its applicability is therefore always problematic, and can be established only over time - and never for all time. Like the economic agents who are our nominal subjects of study, we have to work in time; why not therefore try to develop theories which are embedded in time - and therefore in uncertainty - and take seriously the selective development of connections own time as a result of fallible human action? In the remainder of this paper I will discuss three themes which seem to me central to such dynamic analysis: knowledge, institutions, and organisation. I shall argue that these themes are closely related.
Economists nowadays quite often write about information; it is a convenient way of implementing the strategy already discussed of removing a particular connection from the basic fully-connected model and thus generating a potentially-publishable paper. Information may be coarsely rather than finely partitioned, so that agents are unable to discriminate between states in which different actions are optimal; particular items of information may be missing – not only information about future actions by others; or information may be unevenly distributed. But the content of information is not itself treated as problematic; often indeed it is explicitly information about the probabilities of a closed set of possible states of the world. Underlying knowledge is complete, even if information is not. Thus even when information is dispersed and incomplete, the information sets of all
agents within a model are drawn from a single and complete set. This avoidance of Knightian uncertainty is crucial for the analytical strategies that are used, as has previously been observed.
The denial of Knightian uncertainty motivates the standard treatment of complexity. The assumption of an underlying single and complete information set ensures that all simplifications are derived from a single correct source, which provides a common basis for the analysis of transactions between agents. It is then natural to misinterpret Simon by treating bounded rationality as equivalent to a cost of information and satisficing as an optimal response, and to avoid asking how boundedly rational agents can know enough about the correct model to be certain that their simplifications, though not the whole truth, are nothing but the truth. The answer to that unasked question may be found in what I propose to call Hayek’s Impossibility Theorem: “any apparatus of classification must possess a structure of a higher degree of complexity than is possessed by the objects that it classifies; and.. therefore, the capacity of any explaining agent must be limited to objects with a structure possessing a degree of complexity lower than its own” (Hayek, 1952, p. 185).
The question may also be applied to those who analyse complexity in this way: how do they know that their models of complex systems are adequate representations of the systems to which they are applied? To this question also, Hayek’s Impossibility Theorem supplies the answer: they cannot know. Just as our analysis of systems should not take as its reference point a fully-connected system, which directs us to questions about specific failures arid their remedies, but start from the problem of creating and maintaining connections that are appropriate for particular purposes, so the problem of complexity is not one of simplifying a supposedly complete model - which is a fantasy - but of constructing some representation by selecting and linking elements. Both are exercises in Knightian uncertainty, for which there are no correct procedures, but the possibility of rewards for skill. Information needs to be interpreted, and the interpretation depends upon the classification systems and the connections between categories by which people attempt to make sense - for sense has to be made - of phenomena.
This is how we develop knowledge, by varying our construction systems as we “construe the replication of events” Kelly, 1963, p. 72). Knowledge is structure, in the form of categories into which phenomena or concepts may be grouped, or in the form of relationships between such categories; and structure implies a non-integral space. It is an imperfectly connected system of imperfect connections, and any of these connections may change over time, as Paul Nightingale (2000) shows in a recent analysis of pharmaceutical research strategies. The world system of knowledge is far from complete, and the knowledge possessed by - or even accessible to - any individual is a very small proportion of that world system. Nobody knows how a Boeing 737 works; and nobody knows how the Boeing Company works.
Rather than bounded rationality, which (as already noted) is usually interpreted as a particular limitation in processing knowledge, it is better to begin with bounded cognition. This has the advantage of corresponding with current
ideas about the development of human cognitive abilities. In the early stages of evolution, standard behaviours were genetically programmed; later creatures were genetically endowed with some capacity to vary behaviour by forming new linkages in their brains; performance received evolutionary priority over logical processing and neurological coding over explicit codification. Nevertheless what appeared to be appropriate could differ between individuals because of differences in the sequence of their experiences. Despite our intellectual pretensions, this is still the basic method of knowledge formation in modern humans; that is why ‘we know more than we can tell’, and in particular why we can perform many actions that we are unable to specify in detail. However, the emergence of consciousness introduced the important novel possibility of creating ideas about the future by making conjectures about new categories and relationships as yet unrecognised, leading to the possibility of taking novel actions with the intention of producing novel effects. The scope for variation between individuals was correspondingly increased, and with it the rate at which knowledge could grow. This new possibility, we should remember, is a modification of the old capabilities, which are not displaced, and relies much more on linkages than logic. [All this is portrayed in Marshall’s (1994) mental model of ‘Ye Machine’, the product of his early venture into evolutionary biology.] Indeed, as psychologists have shown, our powers of logical reasoning are still primitive in relation to the ability to make novel connections; and if uncertainty is to be gradually replaced by knowledge the latter is far more valuable.
Kelly (1963) based his theory of personality on the need to create representations of parts of a universe that he assumed to be interrelated, though not in the sense of integral space; some connections were very indirect, some were very weak, and the ultimate bond was provided by time. The need to construct knowledge, and the role of imagination in doing so, was emphasised by Adam Smith ( 1980) in his psychological theory of the emergence and growth of science as a combination of classification systems and causal links, which he illustrated by the History of Astronomy. The stimulus to imagination was provided by the failure of existing patterns of knowledge to account for newly-observed phenomena - an intrinsic motivation, beginning with unwelcome surprise and concluding with delight in creating a new pattern that worked, that appears to have had substantial survival value and still to be effective, but which is not prominent in economic theory. Since new ‘connecting principles’ led to new expectations, new activities and new observations, what began as an aid to ordinary living gradually incubated a new category of knowledge called ‘scientific’. As the psychological and practical value of this knowledge became more apparent some people came to devote particular attention to it; and as its growth accelerated it began to divide into distinctive branches, each with its own set of connections which gave rise to its own anomalies and consequent stimulus to imagination.
Having explained how the dynamics of scientific development led to specialisation which accelerated the process, Smith later transferred this analysis from science to the economy, and made the power of the division of labour to increase productivity the basis of his dynamic economic theory (Smith, 
1976b). Smith was well aware that increased specialisation had its opportunity costs in the neglect of potentially important connections; this led him to include education as an important function for government, and to give a special role to “philosophers or men of speculation” who imagined novel connections between divergent specialisms - or, in Schumpeter’s (1934) language, conceived of “new combinations”.
Knight (1921, p. 206) observed that “to live intelligently in our world we must use the principle that things similar in some respects will behave similarly in certain other respects even when they are very different in still other respects”. One class of ‘connecting principles’ serves to indicate which things should be treated as similar, despite their differences (and also which things should be treated as different, despite their similarities); and a second class of principles suggests which categories, so ordered, should be assumed to be linked, and in what way. Popper (1963, p. 44) pointed out that a perception of similarity (and also, we may add, of complementarity) always “presumes a certain point of view”. Thus the construction of knowledge is always potentially subject to interpretative ambiguity, and the boundaries of categories are likely to be differently construed by people with somewhat different histories. Now changes in knowledge systems, as Potts argues, are mainly changes to adjacent states; and Marshall expected experimentation to occur at the margins of knowledge. But for any system of any complexity there are many adjacent states; moreover, what is adjacent tends to differ between people because of the heterogeneity of their experience, and which of these possibilities is perceived also tends to differ. Thus at any time there are many margins of knowledge, arid thus the potential for a great deal of variation.
In one of his early papers, Marshall noted that “in economics every event causes permanent alterations in the conditions under which future events can occur” (Whitaker, 1975, 2, p. 163). Not the least of these alterations is the state of knowledge, which may change both directly through perception of the event or indirectly because the event prompts a search for a new interpretation (or in other words for a new connection between categories). Since, as Metcalfe (2000, p. 148) reminds us, “the supply conditions for new knowledge depend on the current state of knowledge”, the growth of knowledge is a path-dependent process. That does not mean that it is path-determined, because the conjectures that are represented by new or modified structures are subject to many different selection processes - of which selection in markets is no doubt of most interest to the majority of economists (though we must not forget that a market is a set of institutional arrangements based on knowledge structures which are themselves subject to challenge).
However, although evolution is undoubtedly about the emergence of novelty through processes of variation and selection, it is also about stability - and necessarily so. If everything is changing, or even liable to change at any moment, then nothing can be relied on - for making decisions, interpreting information, or constructing new knowledge. Any process of variation and selection is meaningless unless both the variants and the selection environment persist for a time.
In Marshall’s (1994) mental model of a ‘machine’ the lower level maintains a collection of routines which have worked satisfactorily: and this both frees the higher level for imaginative exploration and presents clearly defined problems when an established routine fails to cope with a new situation. Penrose’s firm similarly requires both evolving resources arid an administrative structure; firms are sense-making systems which (if successful) combine the cognitive distance which supports specialisation with cognitive similarity in the dimensions which maintain focus on the objectives of the business.
Individuals develop structures of knowledge, including knowledge of how and when to perform particular actions, and how to frame sets of premises as a basis for deductive reasoning. They learn how to make sense and how to make decisions, both of which require more than logic, as Chester Barnard (1938, p. 305) emphasised. However, unless they live a purely solitary existence they do not have to do this on their own. The activities of others create a range of vicarious experiments which all individuals may use to test their own conjectures or to incite their imagination to produce new conjectures; or they may simply adopt apparently successful patterns of behaviour or satisfying ways of organising knowledge. This is how we all start as infants; and it is perhaps the basis of our willingness to accept the authority of many communications and demonstrations, not only from ‘persons of authority’ (Barnard, 1938, p. 163). It is an obvious economy, and sometimes an aesthetic pleasure, to free-ride on other people’s wisdom; that is how Smith ( 1980) explained the diffusion of new cosmological theories which appeared to resolve worrying problems, and also the adoption of rules of behaviour which appeared to conform with moral sentiments (Smith,  1976a). Ways of thinking and ways of acting that are common within a community need not be explained as solutions to co-ordination games; they may arise from individual efforts to solve individual problems.
If the sharing of patterns and routines has such origins, that helps to explain how members of a group who have been acting in parallel may converge on a particular set of procedures for managing interactions. (Smith was well aware of the importance of this sequence in making civil society possible.) What we call ‘institutions’ when they are interactive routines are not inherently different from the routines and assumptions on which people necessarily rely in order to economise on cognition for their own private purposes; they are an external supplement to the structure of internal cognition (Choi, 1993), for every person, like every firm, needs both an internal and an external organisation. Access to this external cognitive capital depends on the appropriate absorptive capacity, which, as Cohen and Levinthal (1989) reminded us, is so important in human progress. We may think of this capability as a set of receptors which can connect imported elements to internal structures: since it depends on the connecting principles which are already being used by the prospective absorbers it is not
a general ability but context-specific, and therefore embodies opportunity costs. This specificity is substantially influenced by the division of labour. The development of such capabilities is a major function of education; and studies of organisational learning have shown the importance of social interaction within and between productive organisations in facilitating such learning. In both private and interactive contexts, predominant reliance on routines is necessary in order to create space for thought; and in both contexts, variation between individuals widens the range of material about which to think. Codification is an institution which partially formalises tacit knowledge and thus provides the basis for the creation of further tacit knowledge.
An obvious but neglected application of the importance of institutions in encouraging the growth of knowledge is the emergence of markets. A market reduces the costs of making certain kinds of transactions by establishing powerful connections. Mark Casson (1982) deserves the credit for noting that the costs of continuing transactions may be reduced by appropriate investment, and identifying the entrepreneurial role of those who make such investments - though as recent events have amply demonstrated many entrepreneurs may be unfortunate or misguided. When a particular class of transactions has been substantially reduced to routine, those using that market, as buyers or sellers, no longer have to think about how to transact and are therefore free to think about what to transact, how to produce the goods or services to be transacted and how to make good use of them (Loasby, 2000). Thus the institutionalised connections provided by a market allow the formation of new connections, both in trading relationships and in the form of knowledge about both production and consumption. The emerging interest in the role of the consumer builds on an understanding of market institutions.
Institutions provide the connections which support dynamics; they also have their own dynamics, primarily of adjustments at the margin, but also of regime changes (Dopfer, 2001), which typically draw on patterns of connections from some other sphere of activity and may be treated as adjustments at the margin of a higher-level system: the idea that structure influences behaviour, for example, appears in different forms across many fields of human knowledge. Knowledge changes institutions, as institutions shape knowledge. This process drives the history of economic thought, as well as the development of productive knowledge and both managerial and entrepreneurial skills.
According to Roger Myerson (1999, p. 1068), “today economists can define their field more broadly as the analysis of incentives in all social institutions’. Economic organisation, which at one time focussed on the relationship between market structure and economic performance, is now interpreted as the organisation of incentive structures. This is certainly a broadening in one dimension, but imposes serious constraints in others. Incentives matter; but co-ordination, both
within and between firms (and for individuals too - see Kelly, 1963) is first of all a cognitive problem, as Richardson (1972) showed. Marshall (1920, p. 138) linked organisation specifically to knowledge, and half-explicitly linked different forms of organisation to different kinds of knowledge. But Williamson, in assessing the merits of different organisational arrangements, treats governance systems as protective devices against pernicious incentives and does not, like Penrose (1959, 1995), consider them as bases for the generation and application of knowledge. Williamson’s (1985, p. 48) declaration that “were it not for opportunism, all behaviour could be rule governed” ignores Knightian uncertainty and its counterpart, Shackleian imagination; opportunities, rather than opportunism, drive the growth of a Penrosian firm, and no two Penrosian firms are identical.
These opportunities result from new knowledge which is shaped by institutions that are fostered by organisational arrangements; the Penrosian process in its administrative framework combines cognitive, institutional arid organisational dynamics. The organisation of a new activity requires new connections to be made, in formal responsibility, in patterns of interaction and in individual cognition. If the activity is successful most of these connections cease to require conscious attention; a new set of institutions releases cognitive skills and organisational capabilities for other purposes (as in Marshall’s ‘machine’). This is ‘the receding managerial limit’. At the same time, the absorbed patterns of behaviour, at all levels, change the firm’s resources, which may be deployed in directions which are conjectured by the use of these cognitive skills.
That such connections between resources and profitable uses are not simply deduced from the data, as in standard theories which are located in integral space, but need to be made is a clear and fundamental difference between Penrose’s theory and the standard ‘theory of the firm’, a difference emphasised by Penrose’s distinction between resources and productive services (see also Lane et al., 1996). Writers on strategy who adopt the ‘resource-based view’ often underrate the significance of this distinction. We may also think of a firm’s resources as equivalent to Lachmnann’s conception of capital: they are elements which may be substituted between uses but which in any particular use are valuable because of their specific complementarity (or connections) to certain other elements. If this complementarity produces what was once called synergy or what we now call superadditivity, the additional productivity may be attributed not to the elements but to the connections between them.
These Penrosian single-firm dynamics should be supplemented at least by the two other Marshallian categories of forms of organisation that aid knowledge: the firms within a single trade, by thinking and acting somewhat differently but in readily comprehensible ways, provide vicarious experiments and vicarious hypotheses to supplement and interact with the particular knowledge of each, and the network of complementary trades is structured on Richardsonian principles of dissimilar ways of organising knowledge to gain the advantages of the division of labour while avoiding unhelpful connections (Richardson, 1972), and linked by incremental adaptations and by speculative visions. The organisation of production is also the organisation of knowledge, and both kinds of organ-
isation change over time as the result of what happens in time. The dynamics of industrial organisation have never been better presented than by Allyn Young (1928) in a paper which rejected the applicability of equilibrium modelling to an understanding of this process of generating value as a consequence of rearranging connections by reconfiguring the internal and external boundaries of the firm, creating new connections to make new markets. Increasing returns are returns not to the elements but to the connections between them. Such an imputation is impossible in a theoretical system which assumes a fully-connected economy, but it is a natural implication of Marshall’s (1920, p. 318) definition of increasing returns as mediated by organisational change.
The concept of general equilibrium is not applicable to these dynamics, but local and temporary equilthria may serve very well to indicate the knowledge and relationships - well-established connections of various kinds - on which people may reasonably rely in order to construct useful novel connections. Innovation is always innovation in particular respects and at particular levels, and is carried by continuity, or maintained connections, in other respects and at other levels; and continuity may be expressed by an appropriate concept of equilibrium, applied to particular structures of knowledge, institutions, or organisation.
Evolutionary economics relies on differences, not only between but also within industries; the effects of these differences on behaviour, continually modifying and occasionally disrupting the environment in which firms are operating, requiring new interpretations and sometimes prompting new perceptions, provide the dynamics. These processes combine the generation of variety and the elimination of variants which do not match the criteria by which they are judged; and these criteria are themselves a proper and neglected field for analysis, for there are different criteria in different selection environments. However, there is danger in simply replacing the field theory of physics with neo-Darwinian biology, which excludes human purpose and sharply differentiates the context of variety generation from the context of selection, whereas in human brains and human organisations the contexts are often combined. It is safer to draw inspiration from Adam Smith’s ( 1980) evolutionary model, which includes complex motivation, imaginative conjecture (often driven by aesthetic considerations), selection and diffusion, falsification as a stimulus to novel conjectures, the evolution of the evolutionary process itself through increasing differentiation and the crucial importance of the division of labour. Neo-Darwinians seek to confront us with a stark choice between design and natural selection among blind mutations; standard economic theory opts decisively for design, occasionally supplemented by appeals to unanalysed selection processes to ensure that the design is optimal. Both are corner solutions in the space of theoretical strategies; evolutionary economics avoids corner solutions by choosing a sequence of ex-ante decisions and ex-post realisations that may lead to fresh decisions.
Contemporary models of economic organisation often depend on the concept of asymmetric information, which certainly corresponds to an aspect of reality; but the more important asymmetry is of interpretation and of perception, which leads some individuals and some organisations to take actions that others have dismissed, or never even thought of. Frank Knight’s theory of entrepreneurship and the firm was based on interpersonal differences in the capacity for judgement - what we might call making connections that prove to be appropriate - and of differences for each individual between fields of activity (Knight, 1921, p. 241). Shackle’s (1979, p. 26) beautiful phrase “the imagined, deemed possible” invites us to consider the stimulus and sources of imagination and why some products of the imagination are deemed possible by particular individuals while others are not.
Imagination and the assignment of possibility require the making of new connections, and often the discarding of old connections which appear to conflict with them, a process that is easier to understand in retrospect than to predict. Because new knowledge, new institutions and new organisations must all develop from connected systems (at some level) that already exist, change is always path-dependent; but this dependency may vary greatly in both degree and kind, often leaving much scope for imagination, especially if we extend Shackle’s phrase to include the imagined, deemed capable by some entrepreneur of being made possible. Since the number of connected networks that are conceivable is unimaginably greater than the number that can be handled by any human brain - or indeed by any organisation that depends on manageable interactions between human brains - it is not surprising that there will be a great variety of opinions about what will work, and what will be profitable. There will be a high rate of failure; the dynamics of evolutionary economics requires both ex-ante and ex-post selection.
This variety, and its potential, justify concluding this sketch of evolutionary dynamics by invoking George Richardson’s (1975, p. 359) principle: “Surely it is of the essence of competition that the participants hold uncertain and divergent beliefs about their chances of success”. This is competition between different ways of thinking; and the co-ordination problem within an economy is that of achieving the necessary compatability between different ways of thinking while preserving the differences. There are difficult incentive issues here, but they are not the incentive issues that dominate Myerson’s conception of economics, for they are linked to co-ordination problems at many levels, at each of which some connections are to be encouraged and others avoided. Knight (1921, p. 268) observed that “[w]ith uncertainty absent... it is doubtful whether intelligence itself would exist”. Why should we be satisfied with the analysis of rational choice when we have the opportunity to study intelligent action?
Barnard CI (1938) The functions of the executive. Harvard University Press, Cambridge, MA
Casson M (1982) The entrepreneur: an economic theory. Martin Robertson, Oxford
Choi YB (1993) Paradigms and conventions: uncertainty, decision making and entrepreneurship. University of Michigan Press, Ann Arbor
Coase RH (1988) The firm, the market and the law. University of Chicago Press, Chicago
Cohen WM and Levinthal O (1989) Innovation and leaning: the two faces of R & D. Economic Journal 99: 569-596
Dopfer K (2001) History-friendly theories in economics: reconciling universality and context in evolutionary analyses. In: Foster J, Metcalfe JS (eds) Frontiers of evolutionary economics: competition, self-organization and innovation policy. Edward Elgar, Cheltenham Northampton, MA
Hahn FH ( 1984) On the notion of equilibrium in economics. Cambridge University Press, Cambridge. Reprinted in: Equilibrium and macroeconomics, pp 43-71. Basil Blackwell, Oxford
Hart OD(1996) Firms, contracts and financial structure. Clarendon Press, Oxford
Hayek F A (1952) The sensory order. University of Chicago Press, Chicago
Hicks JR (1948) Value and capital, 2nd edn. Oxford University Press, Oxford
Hicks JR (1982a) Money, interest and wages: collected essays in economic theory, Vol 2. Basil Blackwell, Oxford
Hicks JR ( 1982b) Time in economics. In: Hicks (l982a) pp 282-300
Hume D ([1730-40] 1978) In: Selby-Bigge LA, Nidditch PH (eds) A treatise on human nature, 2nd edn. Clarendon Press, Oxford
Kelly GA (1963) A theory of personality. Norton, New York
Kirzner IM (1973) Competition and entrepreneurship. University of Chicago Press, Chicago
Knight FH (1021) Risk, uncertainty and profit. Houghton Mifflin, Boston
Lane R, Malerba F, Maxfield R, Orsenigo, L. (1996) Choice and action. Journal of Evolutionary Economics 6(l): 43-76
Leijonhufvud A (2000) Macroeconomic instability and co-ordination: selected essays of Axel Leijonhufvud. Edward Elgar, Cheltenham Northampton, MA
Loasby BJ (1976) Choice, complexity and ignorance: an enquiry into economic theory and the practice of decision making. Cambridge University Press, Cambridge
Loasby BJ (1998) Co-ordination failure in economic theory: Economists in the 1930s. In: Jolink A, Fontaine P (eds) Historical perspectives on macroeconomics: 60 years after the General Theory, pp 72-87. Routledge, London
Loasby BJ (2000) Knowledge, institutions and evolution in economics. Routledge, London
Loasby BJ (2000) Market institutions and economic evolution. Journal of Evolutionary Economics
Loasby BJ (2002) Content and method: an epistemic perspective on some historical episodes. European Journal of the History of Economic Thought 9 (1)
Marshall A (1920) Principles of economics. Macmillan, London
Marshall A (1994) Ye machine. Research in the history of economic thought and methodology, Archival Suppl 4, 116-132. JAI Press, Greenwich, CT
Ménard C (1995) Markets as institutions versus organizations as markets? Disentangling some fundamental concepts. Journal of Economic Behavior and Organization 28: 161-182
Menger C ( 1976) Principles of economics; translated by: Dingwall J, Hoselitz BF, New York University Press, New York
Metcalfe JS (2000) Evolutionary approaches to population thinking and the problem of growth and development. In: Dopfer K (ed) Evolutionary economics: program and scope, pp 141-164. Kluwer, Boston Dordecht London
Myerson RB (1999) Nash equilibrium and the history of economic theory. Journal of Economic Literature 37: 1067-1082
Nightingale P (2000) Economies of scale in experimentation: knowledge and technology in pharmaceutical R&D. Industrial and Corporate Change 9: 315-359
Nooteboom B (1992) Towards a dynamic theory of transactions. Journal of Evolutionary Economics
Pemrose PT (1052) Biological analogies in the theory of the firm. American Economic Review 42:
Pemrose ET (1959, 1995) The theory of the growth of the firm. Basil Blackwell, Oxford (1959); 3rd edn: Oxford University Press, Oxford (1005)
Popper KR (1963) Conjectures and refutations. Routledge and Kegan Paul London
Potts J (2000) The new evolutionary microeconomics: complexity, competence and adaptive behaviour. Edward Elgar, Cheltenham Northampton, MA
Richardson GB (1953) Imperfect knowledge and economic efficiency. Oxford Economic Papers 5(2):
116-156; reprinted in: Richardson GB (1998), pp 1-21
Richardson GB (1959) Equilibrium, expectations and information. Economic Journal 69: 223-237; reprinted in: Richardson GB (1998) pp 51-65
Richardson GB (1960) Information and investment. Oxford University Press, Oxford
Richardson GB (1972) The organisation of industry. Economic Journal 82: 883-896; reprinted in:
Richardson GB (1998) pp 143-156
Richardson, GB (1975) Adam Smith on competition and increasing returns. In: Skinner AS, Witson TW (eds) Essays on Adam Smith, pp 350-360. Clarendon Press, Oxford; reprinted in: Richardson GB (1998) pp 157-167
Richardson GB (1998) The economics of imperfect knowledge. Edward Elgar, Aldershot Northampton, MA
Schumpeter JA (1934) The theory of economic development. Harvard University Press, Cambridge, MA
Shackle GLS (1972) Epistemics and Economics. Cambridge University Press, Cambridge
Shackle GLS (1979) Imagination and the nature of choice. Edinburgh University Press, Edinburgh
Simon HA, Egidi M, Mains R, Viale R (1992) Economics, bounded rationality and the cognitive revolution. Edward Elgar, Aldershot
Smith A ( 1976a) In: Raphael DR Macfie AL (eds) The theory of moral sentiments. Oxford University Press, Oxford
Smith, Adam ( 1976b) In: Campbell RH, Skinner AS, Todd WB (eds) An inquiry into the nature and causes of the wealth of nation, 2 vol. Oxford University Press, Oxford
Smith, A ( 1980) The principles which lead and direct philosophical enquiries: illustrated by the history of astronomy. In: Wightman WPD (ed) Essays on philosophical subjects, pp 33-105. Oxford University Press, Oxford
Whitaker JK (1975) The early economic writings of Alfred Marshall 1867-1890. Cambridge University Press, Cambridge
Williamson OE (1985) The economic institutions of capitalism: firm; markets and relational contracting. Free Press, New York
Williamson OE (1996) The mechanisms of governance. Oxford University Press, Oxford
Young AA (1928) Increasing returns and economic progress. Economic Journal 38: 527-542
Zappia C (2001) Equilibrium and disequilibrium dynamics in the 1930s. Journal of the History of Economic Thought 23(l): 55-75