The process by which knowledge or information evolves and spreads through the economy involves changing its nature between tacit and codified forms. The process of codification incudes three aspects: model building, language creation and the writing of messages. Recent technical changes in several technologies have impinged on those three activities and have changed the costs and benefits from each of them. This in general has lowered the costs of codification. Further, technical changes have facilitated the diffusion of codified knowledge which has increased its value. Due to the temporal relations among the three aspects of codification, the ongoing process in which codification takes place may be path dependent.
JEL numbers: O31; O32; O33
keywords: knowledge; codification; diffusion; knowledge generation; innovation; path dependence
The process by which knowledge or information evolves and spreads through the economy involves changing its nature between tacit and codified forms. Typically, a piece of knowledge initially appears as purely tacit -- a person has an idea. Often, though, as the new knowledge ages, it goes through a process whereby it becomes more codified. As it is explored, used, and better understood, less of it remains idiosyncratic to a person or few people, and more of it is transformed into some systematic form that can be communicated at low costs. In some cases, as the principles underlying the piece of knowledge come to be understood, they can be written down, and the piece of information can be described as an instantiation, under certain initial conditions, of a general phenomenon. In other cases, a procedure that was developed to produce some end becomes routinized, and repeatable, which implies that it can be broken down into component pieces, each of which is sufficiently simple that it can be described verbally, or embodied in a machine. This, again, is a process in which tacit knowledge becomes codified.
Economists interested in the technological dimension of macroeconomic phenomena seem to be converging on the view that the so-called "cumulative expansion of the codified knowledge-base" has serious implications, and will change the form and structure of economic growth, modifying dramatically the organization and conduct of economic activities (Abramowitz and David, 1996; Soete, 1996; OECD, 1996; Eliasson, 1990). At the same time other authors have explored the microeconomic implications of knowledge codification as entailing changes in technological learning and in the institutional structure of innovative activities (Arora and Gambardella, 1994; Dasgupta and David, 1994; Cowan and Foray, 1995; David and Foray, 1995; Ergas, 1994). But the bridge from those microeconomic changes to macroeconomic facts is not at all clear. Even if more economists share the view that "codification is becoming the essence of economic activity" (Steinmueller, 1995), some scepticism remains with regard to the pervasiveness and the degree to which the nature of this change is revolutionary. Indeed, it is difficult to identify the actual magnitude and scope of diffusion of the tendency towards knowledge codification, who is bearing the costs and who is sharing the benefits, and what kind of characteristics (in terms of size, space, particular coordination requirements, and so on) a system must exhibit (whether it is a firm, an industry, or a scientific community) in order to increase the value of codification for the organization and conduct of its activity.
I-1 - Why is it an important issue
Knowledge codification -- the process of
conversion of knowledge into messages which can be then processed
as information -- changes some fundamental aspects of the
economics of knowledge generation and distribution. The codification
process entails high initial fixed costs but allows agents to
carry out certain operations at very low marginal costs:
* In principle, codification of knowledge can reduce the costs of knowledge acquisition. In a general sense, codification reduces the costs and improves the reliability of information storage and recall. Provided the media remain readable, and the language is not forgotten, in principle the knowledge can be stored and retrieved indefinitely. Many aspects of knowledge acquisition -- transport and transfer, reproduction, storage and even access and search, are all functions the costs of which fall dramatically with codification. Typically, knowledge that is tacit resides in people, institutions or routines. If the number of people is small, this makes tacit knowledge very difficult to transport.1 It can also make it difficult to find, since there will be often only a single instantiation of it. But codified knowledge is easy to reproduce, and thus there can be many copies of it. This will make it relatively easy to find, and easy to transport. All of this implies that codification can reduce the costs of knowledge acquisition by those who are interested in the knowledge that has been codified. (Simon, 1982, Ergas, 1991).
* A second way in which codification provides benefits is that through codification, knowledge is becoming more like a commodity. It can be more precisely described and specified in terms of content and intellectual properties, and this reduces uncertainties and information asymmetries in any transaction involving knowledge. Knowledge becomes transferable independently of the transfer of other things, such as people, in which the knowledge is embedded. This facilitates market transactions in knowledge that are difficult to enact when knowledge is tacit.
* A related issue has to do with codification reducing problems of asymmetric information. In a non-monopolistic market, asymmetric information equilibria cannot exist if it is possible (or possible at relatively low cost) for the supplier to communicate information to the consumer. Any supplier who kept a secret would be treated with suspicion by prospective consumers, and so would lose market share to his competitors. But typically, the information of interest to consumers is information that is not tacit, but codified. A consumer would like to be able to acquire, before consumption if possible, information about the properties (especially about quality) of the good. Thus the existence of a language which is public, and the codification of information in that language, will reduce asymmetric information in a market. Gunby argues that this is one of the important features (or at least potentially important features) of quality assurance standards. (Gunby, 1996, Chapter 1.) These standards constitute a language in which production processes can be described. In principle, this allows a consumer to extract information about the likely quality of the good being produced.
* A second-order effect concerns the relation between codification and the spatial organization and division of labour. This is true for economic activity in general and innovative activity in particular. The ability to codify knowledge permits the externalization of knowledge generation and allows a firm to acquire more knowledge than previously, for a given cost. Knowledge need not be developed internally, it can be purchased. This lies behind the current trend toward out-sourcing which we see in many industries. Not only is the manufacture of parts being out-sourced, but also their design. This is only possible if knowledge is commodified since design is so information or knowledge intensive.2
All of the benefits just described are of course only potential -- before they can be realized, some important conditions must be fulfilled. In particular, we will argue that getting those benefits require particular characteristics of the system of agents. Only "stable" systems, characterized by initial irreversible investment required to build a community of agents who can understand the codes, and having specific needs of communication, memory or coordination, will fully realize the potential for increased productivity through codification. Further, systems in which the knowledge environment is in flux often bear the costs of making codification possible for stable systems.
The stability of the system is not sufficient though. Some of the effects of commodification will be mitigated by the fact that some aspects of knowledge remain sticky, are not totally slippery and free-flowing. That is, there are often difficulties in transfering certain types of knowledge from one site to another. This stems from the existence of important elements of tacitness that all knowledge has. There are things that are difficult to write down and codify, for example knowledge that is embedded in routines or in individual (or collective) agents.
It is also worth pointing out that the benefits mentioned here stem largely from our ideas about the public nature of knowledge. But it is an enormous simplification to represent codified knowledge as a public good.3 On this aspect of knowledge codification is ambivalent. The process of codification can be used to share knowledge collectively and to transfer it at a minimal cost among the members of a network. On the other hand, it can also be used to maintain this knowledge in the private domain if the code cannot be understood by others. In other words, a code can serve the objective of the marketing department of a company which wants its advertising messages to be read and understood by everyone; but it can also serve the objective of the alchemist of the Middle Ages who was written a book of "secrets" -- a way of allowing himself (or perhaps a select few) and no one else access to his past discoveries. "The esoteric language of alchemy was never intended to be understood literally, but was deliberately used to protect divine secrets and to guarantee their possession by a small circle of initiates" (Eamon, 1985, p. 324). Most frequently putting knowledge into a code serves both objectives: sharing the knowledge among a certain group (of firms, of scientists) and keeping (intentionally or unintentionally) other agents out from the club. Thus, codification does not necessarily reinforce the public good character of knowledge. However, the existence of high fixed costs to produce a specific language (in order to protect the codified knowledge) should discourage any kind of strategy of private codification and should increase the incentives to make the codes public or semi-public. Thus, codification is a strategic instrument available at a certain costs for agents (or groups of agents) to pursue any strategy. That this information will be kept secret, or fully disclosed, or shared among a certain group of people is, not a feature that impinges on the basic structure of our discussion of costs and benefits. It will, of course, have implications for the relative sizes of different costs and benefits.
I-2 - The debate on codification versus
The qualification just mentioned is the reason
why the debate about the changing nature of the relation between
codified and tacit knowledge is important. Information technology
is now coming to dominate the technological paradigm to such an
extent that it seems that the codification of knowledge must be
increasing at a fantastic rate. After all, information technology
is about processing, saving, and transmitting information or codified
knowledge. This might be thought to suggest that the ratio of
codified to tacit knowledge is increasing.4 This, it is argued,
comes hand in hand with the new "knowledge-based economy"
which is built on the cumulative expansion of the base of codified
knowledge. On the other hand, some claim that the distribution
of knowledge between tacit and codified has not changed. (See
Lundvall and Johnson, 1994, Senker, 1995, Dosi, 1996, for example.)
It will always be true that tacit knowledge is needed to use codified
knowledge. Thus if there is an addition to the codified knowledge
base, there must also be an addition to the tacit knowledge base
by which agents can use the new codified knowledge and thus give
it economic value.
It is unfortunate, though possibly of necessity, that there is little empirical evidence on either side of the debate.5 There is, however, considerable "anecdotal" evidences, showing that firms and research institutions are allocating many resources to the process of codification. This is particularly true for large companies. But this tendency also concerns SME's; for example those involved process or quality standardization.
I-3 - Relations between codified and tacit
There is now a consensus that codified and tacit knowledge are complements rather than substitutes. The process of codification does not provide all of the knowledge needed to undertake an action; there will always be some tacit knowledge involved in performing any action. This is the reason that codification cannot be considered as a simple transfer of knowledge from the tacit to the codified domain. It is, rather, the construction of new ensembles of codified and tacit knowledge. In other words, codification is never complete, and some forms of tacit knowledge will always continue to play an important role.
This is not to argue that there are absolute limits to codification, as would be the case, for example, if certain knowledge, or certain types of knowledge were in principle not codifiable. The argument is, rather, that there will always be some tacit knowledge needed to use any codified knowledge -- at the very least, the knowledge of how to read the messages.6 That having been said, the scope of what can be codified seems to be continually expanding. Technical and technological advances are such that the complexity of the knowledge that we can codify continues to expand.7
The limitations to codification observed in practice are endogenous. They are generated by the costs and benefits of this action, the variations of which will affect the incentive structures to codify knowledge. For example, many factors -- such as the co-existence and survival of multiple local codes or the historical adoption of successive codes or, to take the simplest argument, the high cost of codifying a certain type of knowledge -- can decrease the incentives to go further, lowering the private rate of return on codifying knowledge. This low rate of return in turn can induce the maintenance of a large community of people possessing the tacit knowledge. In this case, there will be a labour market that can be used to store and transfer the knowledge from firm to firm. Of course, the presence of a thick labour market as a way of transfering knowledge further reduces incentives to codify.
This focus on endogenous limitations means that costs and benefits and the resulting incentive structures are pivotal in shaping the dynamics of codification. Emphasizing the role of the incentive structures by no means implies that the codification of new forms of knowledge is an instantaneous process: moving the boundaries between codified and tacit parts of the stock of knowledge is a matter of long-term technological and institutional evolution; involving changes in the incentive structures, and in the costs and benefits.
I-4 - The goal of the paper
Considering the difficulties regarding the lack of empirical evidence and available measures, we will not enter the debate about the changing nature of the relations between codified and tacit knowledge. Rather, our inquiry concerns the incentives to codify knowledge, and the conditions under which codification of knowledge has an increasing value.
More precisely, the goal of the paper is to characterize various kinds of systems (or situations), exhibiting particular classes of c/b structures. We will differentiate among:
* situations where the maintenance of a great deal of tacit knowledge is tolerable, either because the knowledge considered is too costly to codify or because the benefits expected are not significant (the tacit knowledge is easy to transfer or to memorize; there are no particular coordination needs);
* situations characterized by an excess of tacitness, because the system has specific requirements in terms of knowledge memorization, transfer, description, and so on and that cannot be achieved at the level of tacit knowledge maintained within this system; or because the lack of codification operates as a bottleneck in the realization of productivity gains from use of information technologies;
* situations where there are some excessive codification because the survival of local codes or the adoption of successive generation of codes operate as an obstacle to radical technological changes or because premature codification has stopped the exploration of technological variety.
We note finally that the process of codification is likely to be a path dependent one and this implies that efficiency is not an intrinsic attribute of the codification of a certain type of knowledge, but is rather the result of the dynamics and emergent properties of the system under consideration.
II. What is codification?
It is worth discussing exactly what we mean by "codification" as this is a term that is not always used consistently. This has led to certain internal tensions in the debate on the changing extent of codified knowledge.
II-1 - Differentiating between codification
and the dynamics of information infrastructure
At least some of the difficulty in the debate is that it has taken place before certain fundamental (terminological) difficulties have been sorted out.
A clear distinction must be made between the medium and the message. A message is expressed in a certain code (some language) and in principle can be stored, at least temporarily, on some medium, such as paper or a diskette. This does not deny that there is a relation between the types of languages, and thus the types of storable or codifiable information, and the nature of the medium. But it is worth distinguishing clearly between them in order not to conflate technological advances of very different natures. There are on the one hand for example, increases in storage capacities of a medium, or the development of new media; and on the other hand development of new languages. Running together medium and message can also conduce to the neglect of advances in codification for the reason that no changes are visible at the level of media. One example here might be the development of quality assurance standards. Currently, this is a significant case of codifying knowledge, entailing simplification and rationalization of information systems, without involving new types of media. Traditional, existing media are used to hold information about production processes in order to facilitate the transmission of this information between producers and users. [See Gunby 1996, Chapters 1 and 5.]
There is a further confusion between advances in codification and advances in knowledge access and distribution through electronic network channels. In certain places, the best example perhaps being a library, the major changes over the last two decades concern not codification but rather the medium and means of access. Libraries are by definition store-houses of codified knowledge. In that sense they have not changed since Alexandria. What has changed is the technology by which the information is stored (the medium in some, though certainly not all, cases), and managed. This can be seen at two levels. The information management system -- the card catalogue -- has been made electronic. This has changed the way people inquire about and manage the universities holdings. But in addition, some of the holdings themselves have been made electronic. This has caused a similar change in the way people use the holding of libraries. The most important way in which information technology has affected libraries is not in the nature of codification but rather in the media of codification and thereby in the way in which the information is stored, retrieved and used.
Thus it is useful to differentiate between codification on the one hand and information infrastructure (by which we refer to media and networks) on the other. There is of course a strong relation and several sources of feedbacks between them. Advances in technological infrastructure will change the economic value of codification in two ways. Directly, they can change the costs of codifying certain types of knowledge, the costs of storing it once codified, and possibly the ease with which it is used in codified form. Indirectly, changes in the information infrastructure can increase the value of codification through ease of diffusion. Codified knowledge can now be transmitted over long distances and within complex networks, at very limited cost and high speed. These changes clearly increase the potential value of codified knowledge, which will make it more attractive to allocate resources to the process of codification.
Having stated what codification is not (improvements to medium and networks of storage and diffusion), we now explain what codification is.
II-2-Three aspects: messages, models, languages
There is a first, largely acknowledged, aspect
of codification which deals with the transformation of knowledge
into information. In this sense, codification is a process of
creation of messages, expressing pre-existing knowledge,
which can then be processed as information. Achieving this function,
though, implies bearing high fixed costs (both on the supply and
demand sides) in order to carry out certain operations (typically
knowledge transfer, acquisition and storage) at low costs. The
various implications of codification described at the very beginning
of the paper (decreasing costs, commodifying knowledge, reducing
information asymmetry, and so on) are concerned with this first
There is a second aspect, however, largely ignored in the literature, which deals with knowledge creation: codification is a process which typically involves the creation of models, since modelling knowledge is a prerequisite to transforming this knowledge into information. In this sense, codification is a central method for producing knowledge. The existence of this second aspect means that codification cannot be considered as a simple transfer or translation operation. There is always an aspect of creation. Codification typically entails fundamental transformations in the way knowledge is organized, so the codified knowledge-base cannot exactly cover the tacit knowledge-base for which it tries to substitute. Thus, because it involves creation, codification is an irreversible process: once knowledge is transformed into information, it is not possible to return to the original tacit state. As suggested by Hatchuel and Weil (1995, p.25), in the case of expert-systems, "the imitation of expertise, because it is a process of automation of knowledge, is possible only at the expense of the "active" transformation of this knowledge; it is hence in itself a creator of expertise".
Codification depends then, not only on the creation of messages and models, but, also on the development of some infrastructure. Infrastructural development consists largely of language development. Different types of knowledge demand different types of languages -- music, mathematics, expert systems, novels, all have different languages associated with their codification. Some languages are "generic" and can express a variety of types of knowledge (for instance, it is possible to some degree to write mathematical problems in natural language); some are very specialized. These languages must be developed before any messages can be written.
In order to codify knowledge, a language must exist, but central to any language are concepts and vocabulary. It is the creation of those two things that the modeller is doing. The existence of a vocabulary, which pre-supposes a model (so that if one does not exist it must be created) is necessary for the ability to create messages.8
References to languages mean that a minimal requirement to be a potential user of the information is that one must understand the language in which the knowledge is recorded.9 This is true whether the language is English or French, a computer language or mathematics. Knowledge is easier to codify and codified knowledge is easier to diffuse within a community made up of agents who can read the codes. (We should point out that the ability to read the codes is an important form of tacit knowledge.) Diffusion and use of codified knowledge are thus dependent upon the initial irreversible investment required to build a community of agents, a "clique" or a network the members of which can "read" the codes. This ability to produce and receive signals in a language, even a very common one, requires initial and irreversible investment (Arrow, 1974, p. 39).
The relations between the three layers are
complex: in some situations, codification processes can be carried
out on the basis of pre-existing languages and models. In other
situations, codification requires the creation of a new language
and some knowledge modelling. In most situations, there is some
degree of creation at both the level of languages and the level
of models. Costs and benefits will be very different with regard
to the specifics of each of those situations.
In the rest of this section we will consider several examples which illustrate different cases.
* The two first situations we will describe are situations where the creation of language is the central constraint on the process.
In the 16th century Agricola made a systematic attempt to codify existing knowledge about metallurgy. He decried the fact that it was made difficult due to imprecision in (and corruption of) the existing language, so planned to begin his task by "developing a uniform technical vocabulary." (Long, 1991, p. 338)
More complex cases exist. In the 1960s retailers and wholesalers saw scope for developing a universal product registration system, now known as the universal bar code (in Europe referred to as the European/International Article Number). This sounds like a trivial task but was in fact very time consuming. Standardizing the physical representation of the number, and developing the technology to read it took only a few years. But "standardization of the semantics of the numbering systems and the data interchange formats took more than two decades."10 (Kubicek and Seeger, 1992, p. 370) But until this issue was solved, the bar code, though the technology for using it was well developed, was not available as a means to codify knowledge about products, inventory and suppliers.11
* In the following situation, the constraint is placed on the aspect of knowledge modelling.
Consider a tennis player who wishes to communicate how to serve. (Suppose he has never spoken about serving before.) He must do only two things given that he plans to use an existing language for writing a book on tennis. He must create a model and create a message: He must first break up the action of serving a ball into smaller pieces; into ideas that can be spoken (the modelling phase) in the language. Then he must speak (or write) these ideas in a language that others can understand (the messaging phase). However, in the course of the process, his modelling activity may lead him to "discover" (in the sense of correct recognition of something that possibly already exists, though hidden from view) some micro-movements which cannot be described with available language. He would have to develop some jargon. Thus, he has to develop some creativity at the language level modifying the language in light of his model. This example shows that the relations between the three aspects are neither linear nor hierarchical. There is at least strong mutual causality between language and model.
* In the last two examples the codifier must deal with the three aspects; and thus bear high fixed costs.
Consider the Taylorian project of codifying gestural knowledge (Mangolte, 1996). Gilbreth, a disciple of Taylor, tried to break up any elementary function (such as picking up a tool) to the point beyond which any further reduction seems impossible. He called those elementary micro-movements "therbligs" and identified seventeen basic therbligs. This leads to an "alphabet" in which any micro-movement was codifiable in symbolic terms. Gilbreth tried then, to codify any action; for instance signing a letter is a process characterized by nine therbligs. It is clear that this particular process of codification required the creation of a language, the creation of a model and the action of codification per se. And of course, the cost-benefit analysis we will propose later would show that in the particular case of what Gilbreth wanted to codify, the costs are high (Gilbreth has to produce at the same time a language, a model and a message) while the benefits are low.
Finally, consider the development of an expert system. Codifying expertise again involves the three aspects described above: the decision-making algorithm must be written in a particular computer language; which often must be improved, adjusted, extended according to the particular problems raised by the specific kind of expertise under consideration. But in this process, the most important task is often related to the analysis of the knowledge of the expert: as described in Hatchuel and Weil (1995, p.43) "our examination of the repairer's expertise progressively enabled us to take a more objective standpoint vis-à-vis conventional ways of considering the acquisition of knowledge in expert-systems. Knowledge was no longer raw matter that needed only to be memorised; it appeared rather as the product of a theoretical construction ". The modelling activity may be so important that it can involve two or three steps of codification. For instance, in the case of the codification of the expertise of a repairer, described by Hatchuel and Weil, a maintenance manual had been written. This preliminary work had already produced a description of the components and procedures of available tests; it had made it possible to clear the ambiguities that resulted from diverse definitions formulated by different actors. This first formalization was then the basis for a more advanced stage, that of the writing of dynamic interrogation rules (ibid, p.40). It is clear that a first codification process (involving the three aspects and leading to the writing of a manual) was necessary to reach the appropriate level of knowledge modelling for codifying the very expertise of the repairer.
The existing codified knowledge, the maintenance manual, serves as an input to, or the basis of the model and language of the new, more complete system. This ability to build on an existing system reduces the costs of codifying the fuller knowledge. But, it also reduces the costs of many of the users of the complete system. Any user of codified knowledge must learn the language in which it is expressed, and often must understand at least some of the aspects of the model underlying it. In this case, users familiar with the maintenance manual have already accomplished part of these tasks.
In summary, the three aspects in the activity of codifying knowledge are in principle distinct, but strongly interdependant. For the purposes of this paper, we include the three steps as part of the process of codification. The process will be constrained by one (or two) of those aspects (language, models, messages), with regard to the types of knowledge to be codified (see footnote 7), the purpose of codification, and the strategy of the codifier regarding the appropriation problem.
II- 4- Recent technological changes and
The discussion above suggests that technological
changes can, potentially, affect the economics of codification
in four ways:
- 1) through the development of new languages (in which we also include reconstructing lost languages), which might allow the codification of knowledge previously thought inherently tacit;
- 2) through changes in our ability to create models of phenomena and activities
- 3) through changes in the technologies of coding and decoding, on the basis of existing languages and models
- 4) through improvements in the technologies of storage, recording, and diffusion of messages.
Regarding the definition proposed, process 4 is not considered as affecting codification per se: Advances in technologies associated with the medium (recording, storage), as well as in technologies of distribution are complementary to the codification process. However, as discussed above, they are instrumental in increasing the economic value of codified knowledge for those agents who want to codify knowledge for the purpose of facilitating storage, access or distribution.
In this sense, advances in codification are significant but limited to processes 1, 2 and 3. In recent decades many technological changes have emerged which have had significant impacts in many parts of the economy. Some of these changes will have a direct effect on the economics of codification of knowledge. We give here a non-comprehensive list. The technologies of telecommunication have changed dramatically, with the development of digital switching, which has facilitated changes in the types of service provided and the costs of them. In addition, the technologies of transmission, for example the development and implementation of optical fibre networks, has greatly increased the carrying capacity of the physical infrastructure, which lowers the costs of transmitting data and voice. It is also commonplace that the growth in the speed of central processing units has changed the capabilities of computers. This has made computers useful in places they have not been before, and has introduced the possibility of using new techniques to address old problems. The speed of processing has, for example, made it possible to solve analytic functions symbolically, which has relaxed constraints on functional forms that scientists had previously encountered. Similarly, it is now possible, when closed form solutions are not possible in principle, to do a systematic, thorough exploration of the state space of a function or system in order to solve the problem. For dynamic systems, simulation has become possible as an analytic tool. Connected with these developments, and part of the input to the success of new modelling and solution techniques has been recent developments in applied mathematics having to do with solution and approximation techniques (Ergas, 1991). Approximation techniques that make solution times linear instead of exponential have expanded much further the scope for using computational tools as part of the modeling or analytical toolkit. Advances in the technologies used to store data have been significant, as now, for example, hard disk drives have more storage capacity than did the tape drives of a decade ago. This changes the feasibility of accessing large amounts of data in real-time systems.
In the context of or discussion, these changes have, first, improved out ability to model more complex phenomena. Advances in the development of solution algorithms, increases utility of simulations imply that we can systematize more information in such a way that its structures and relations with other structures become clearer. This ability to better model and to model new things implies that in principle we can codify more types of knowledge. In general, though, codifying new types of phenomena demands new codification tools, specifically, new languages. To a certain extent these are developed as part of the modelling exercise, and so make use of existing language structures, but there is an extent to which they must be developed anew. An example would be the recent need to develop artificial intelligence computer languages. Once these new languages exist, however, the costs of further increasing the space over which the task of codification takes place, is itself lower. This implies that the cost of codification falls with these technological developments.
In general, though, these changes are not enough to have a dramatic effect on codification in practise. Typically, as the knowledge being codified becomes more complex, or is about more complex phenomena, for the codification to be useful, it demands improved storage. More complex phenomena implies that the codified knowledge will in general contain more information, in the sense of the number of bits. In principle this is not an issue, it simply means that a larger quantity of the storage medium is required (more paper or magnetic tape perhaps). But new forms of codification, expert systems or on-line help systems for example, demand fast access to large amounts of varied information -- the stored codified knowledge. This has been made possible by the development of new and better storage media, which have resulted in a reduction of cost of storage, a reduction in cost of use, and an increase in access speed.
Thus we see that both the cost of codification has fallen and that the value of codified knowledge has risen through being less expensive to reproduce and more valuable (faster for example) to use.
This statement regarding the increased value of codified material must be interpreted carefully, however. Printed material, if the paper on which it is printed is relatively low in acid content, can last for many centuries. Physically, it decays slowly, and the language in which it is written evolves in such a way that interpreting it is relatively straight-forward. The same is not necessarily true of documents stored on magnetic media or even optical media. Magnetic media deteriorates relatively rapidly from a physical point of view (roughly 5 to 10 years is the expected lifetime), and with rapid changes in hardware and software technologies, the language in which it is written may disappear rapidly as well. The problem stems from the fact that with new information technologies we are storing not documents, but rather a set of instructions which must be interpreted and carried out by the relevant hardware and software combination before the information they contain is of use to humans. Currently, a reliable, relatively standardized way of archiving information that comes in these forms has not yet been implemented. Several solutions have been suggested, but none is without problems. (See Cook, 1995, pp. 48-53.) These problems exist, but are not insurmountable. Certainly costs of storing and retrieving data in the short term have fallen. If a solution to the difficulties just mentioned is found, the same can be true for longer term storage as well.
A further issue is that new information technologies have reduced the cost of diffusion of codified knowledge. One of the great benefits of codification of course is the way in which it facilitates the dissemination of knowledge. This has clearly been affected by the fall in the cost of telecommunication. This has to a very great extent made possible the explosive growth of the Internet and the World Wide Web. The greater the "market" for the codified knowledge, the greater (in principle in any case) the benefit from codification. The implication here is that any piece of codified knowledge has many more potential users. If the potential is realized, this increases the net social benefits that flow from codification.
III-Costs, Benefits and the Knowledge Environment
Evaluating the costs and benefits of codification would be a complex enterprise. This is effectively due to the interplay of the three aspects of it. In order to understand the source and magnitudes of costs and benefits, it is necessary to put them in the context of the knowledge environment.
III-1 - Have fixed costs been sunk?
A first and straightforward point is that the
incentives will depend to a very great extent on the possibility
of proceeding to codification on the basis of pre-existing languages
and models. In this kind of situation, fixed costs have generally
been sunk: languages and models have been developed by past work
and are known by codifiers and users. The only cost is the variable
one. It is, thus, useful to differentiate between contexts of
stability and contexts of change. Our definition of stability
will be, thus, very simple: stability describes situations where
codification can proceed on the basis of the existing languages
Of course stability is not stationarity. That the existing languages and models used by the agents are sufficient to proceed to the knowledge codification does not imply that there is no change to knowledge or language. There are many examples of stable contexts of innovations; where technological change operates as a standard operating procedure (Hummon, 1984), along a well-defined and predictable trajectory (see Steinmueller, 1991 on ICs technology).
III-2- Costs and Benefits in a stable context
In a stable context -- when there is a community
of people who have made the necessary initial investments to develop
a language and to maintain efficient procedures of language acquisition
for new entrants -- the transfer of messages can be assimilated
to transfer of knowledge, and storing messages means recording
On the benefit side, the efficiency of codification will be greater in very large systems having specific requirements regarding coordination among agents. We identify five classes of situations: systems including many agents and many locations, systems strongly based on recombination and re-use and which take advantage of the cumulativeness of the existing knowledge (rather than on independent innovation); systems which need memory; systems which need particular kinds of description of what (and how) the agents do; and finally systems characterized by an intensive usage of information technologies.
First, it is clear that the size of benefits will be positively related to the size of the potential audience. The more agents to whom the information diffuses, the larger the benefit. Thus, codification will provide high benefits in stable systems characterized by specific requirements of knowledge transfer and communication. This kind of requirement can result from delocalization and externalization tendencies or from the development of cooperative research, entailing a spatial distribution of activities at many places. This first effect can be appreciated without any ambiguity, for example in science: "a humble hp 9000 is radically altering the way research scientists in high energy physics swap information: each day, about 20000 e-mail messages carry to more than 60 countries the abstracts of new academic papers, which readers can develop by gaining access to the full papers. And every day, about 45000 physicists worldwide access the electronic archive to find or to contribute new items " (Mulligan, 1994). This effect operates, however, within a given "clique" or network -- that is a community which shares common codes and tacit knowledge to interpret them.
Second, in (stable) systems of innovation where advances and novelties mainly proceed from recombination, re-use and cumulativeness, benefits of codification are important.12 As claimed by Gibbs (1994), the very little progress in the productivity of software engineering is due to an excessive dependence on craft-like skills (as compared for example to chemical engineering). The schema that Gibbs has in mind is that once an algorithm is written as a piece of code, it can, in principle at least, because it is exists in codified form, be used in many applications, The difficulty in doing so arises in part because of a lack of standardization both in the way code is written and the way algorithms are employed.13 This lack of technological rationalization provided by codification impedes the full realization of the opportunities provided by the re-use and recombination model.
Third, systems which need memory (e.g. firms with long development cycles, high rates of turn-over, some particular demographic problems or institutions confronted by a big technological bifurcation) will greatly benefit from codification. In those systems, too little codification increases the risk of "accidental uninvention". MacKenzie and Spinardi (1995) showed, for example, that the specific, local and tacit knowledge was so important in the nuclear weapon design that there was, always, a risk of losing knowledge after one generation of scientists and engineers -- a risk of accidental uninvention in which much current tacit knowledge is lost. There is no reason not to extend this argument to fields where important inventions for social welfare are developed (taking the standard example of a cure for cancer); where, thus, accidental risk of uninvention must be eliminated by codification.
Fourth, systems having specific needs of describing what agents are doing (either to meet quality standards constraints, or to patent innovations, or to enter into any kind of contractual relations with a partner) will greatly benefits from codification. Here we can also include systems confronted with inefficient market transactions; where the traditional mechanisms of legal warranty, insurance, reputation and test are not efficient to mitigate the effects of information asymmetry (Gunby, 1996). Recording production practices, which is a form of codification based on recent language innovation (in the form of creating standards for record keeping) is aimed at reducing these asymmetries.
Last but not least, a sort of cross situation deals with the lack of productivity gains from the use of ITs, due to incomplete codification. Fully taking advantage of the potential productivity gains of IT typically demands not only the adoption of the technology but also organizational change. (Cowan, 1995.) But a firm undergoing organizational change does not want to lose functionality in the process. The firm must develop jointly the new technology and organizational structures that will reproduce old functions and create new ones. It is obvious that if too much of the old functionality resides in tacit knowledge, or depends heavily on it, this task will be extremely difficult. When the presence of tacit knowledge operates as a bottleneck, impeding the full realization of productivity potential, the firm can expect great benefits from codification (Baumol et al., 1989).
In all these cases, where important operations of transfer, recombination, description, memorization and adaptation (to ITs) of existing knowledge are required, it would be very costly and inefficient to keep this knowledge tacit. We will describe that situation as characterized by an "excess of tacitness".
One dimension which can conflict with these benefits deals with the problem of the proliferation of local codes at both inter- and intra-organizational levels. This tendency creates high incompatibility costs and increases the costs of recodification, which can be considered as an important (and neglected) dimension of the productivity paradox.
Another conflicting dimension, deals with the co-existence of languages, created for the purpose of codifying the same type of knowledge. The reference here is again to incompatibility costs. We can expect a substantial decrease of the size of benefits from the survival of dialects. This is, of course, the case for natural languages (English and French). But it is also the case of very particular specialized languages, such as for example the two "multiview orthographic projections" (so-called "first angle projection" and "third angle projection"), facilitating the transfer of knowledge from engineers and architects to mechanics and builders (Belofsky, 1991). It is clear that the classical trade-off between diversity and standardization, as elaborated in the literature on technology choice applies here. The existence of more than one language raises as an issue the possibility of the loss of network externalities and a possible irreversible excess of diversity. It can be explained by using the same techniques developed in the economics of interface standards to analyse information networks and issues of compatibility among the constituent parts of information systems (David, 1987). In the same vein, any strategy used in the case of excess of technological diversity can be extended to this problem: implementing anticipatory language standards, building a meta-language and creating gateways are valuable strategies to force convergence or at least to decrease the costs of diversity.
III-3 - Costs and benefits in the context
Situations of stability in the knowledge environment
are not universal. Indeed, it is often the case that the knowledge
environment exhibits ongoing changes. Models and languages are
fluid, and the community of agents conversant with the models
and languages is itself changing. The fluidity of the language
implies that there exists a certain amount of uncertainty about
what the messages actually mean because there is uncertainty,
and perhaps change, with regard to the vocabulary in which they
are written. Even when, for example, scientific papers express
knowledge in a natural language there will be issues of jargon,
definitions of words that are specific to the community, and vocabulary
that is specific to the model being used. These can only be learned
as the model itself comes to be understood.
Codification does take place in this context -- scientific papers are written in new fields. But benefits of codification come to a larger extent from the modelling and language development part of the exercise. There may be competition among different basic models, and so among the basic tenets and vocabulary of the language. Until this competition is resolved, issues regarding diffusion discussed in the previous paragraphs will be problematic. The community of potential knowledge generators and users will have difficulty communicating, and the value of knowledge codification that arises from dissemination will be reduced. Thus the codification in this environment has some value from the content of the messages but has more value as an investment good -- a contribution to the resolution of the competition among languages.
It is in the context of change that we might find situations of excess codification. That is to say, the accumulation of successive generation of codes can prevent the development of radically new knowledge, simply because explicating and understanding it would require entirely new codes. As argued by Arrow (1974, p.56) codification entails organizational rigidity and uniformity while increasing communication and transaction efficiency: "the need for codes mutually understandable within an organization imposes a uniformity requirement on the behavior of participants. They are specialized in the information capable of being transmitted by the codes, so that they learn more in the direction of their activity and become less efficient in acquiring and transmitting information not easily fitted into the code." It is, thus, clear that codification can have unfortunate consequences on creativity and radical changes.
The second problem, then, deals with excess of inertia. There are high fixed costs to be born in this process of codification in a context of change. They concern in particular the development and learning of the language in which the new codes will be written. Roughly put, costs are born during the period when the knowledge environment is in flux, whereas benefits are received during a period of stability. During a period of change, infrastructure is developed, languages and models are built, learned and standardized, and a community of agents with shared tacit knowledge grows. All of these investments contribute to a reduction in the fluidity of the environment, and conduce to stability. When the stability is achieved, the benefits of codification can be significant. The languages and models exist, which removes one of the costs. And the network of possible users of the knowledge exists and is relatively large, which increases the potential benefits.
But if developing new languages and models allocates the fixed cost to one generation while many future generations benefit from the new infrastructure to codify knowledge, there is an intergenerational externality problem which can entail a lack of private incentives for developing languages. Solutions which can mitigate this kind of time inconsistency problem deal with the development of relevant market (which may significantly increase the benefits even for the first generation of developers) or with some kind of infinitely-lived institution, or with the existence of altruistic preferences (Konrad and Thum, 1993).
III-4 - On Percolation14
In either case, stability or change, part of
the value of codification arises from the diffusion of knowledge.
Clearly, the magnitude of that benefit will be determined in part
by the number of potential recipients of it. It will also depend
on features of the network over which the diffusion takes place.
We can imagine a network of agents and connections between them. Information enters the network at some point and has the potential to diffuse throughout. There are two parameters that determine the properties of this process. There is the probability (referred to as the node-probability or receptivity) that an agent is receptive to information. There is also a probability (referred to as the connection probability) that a given agent who has been "infected" with the information, will infect a neighbour, given that the neighbour is receptive.
Two mathematical properties of systems of this type exist. First, the probability that a piece of information entering the system will diffuse universally is an increasing function of the node and connection probabilities. Second, there are critical (non-zero) values of these two parameters such that below the critical values the probability of universal diffusion is zero, while above it the probability is strictly positive.
We can use this percolation model to analyze the diffusion of a piece of codified knowledge among the set of potential users. There are two important characteristics of a system that can influence the percolation probabilities, and thus the extent to which a piece of codified knowledge can be diffused. By extension, they affect the size of benefits of codification, as they are positively related to the size of the potential audience.
First some agents are not likely to be receptive, because they cannot understand the code. In this case, the node percolation probability will be small.
Second, some connections between agents may be closed because certain sets of agents are weakly (or not at all) connected to the rest of the network. For example, there are many libraries which still do not receive Industrial and Corporate Change! The connection percolation probability will be small.
Thus, the first percolation probability is highly dependent upon the initial irreversible investments which are necessary to be able to distinguish and understand the signal (Arrow, 1974), while the second is dependent upon the stochastic evolution and extension of the connections between agents. If we define a "clique" as a group of agents within which the two probabilities, receptivity and connectivity, are "high enough", that is, they are jointly higher than the critical values beyond which the probability of universal percolation within the group is positive, it is clear that the benefits of collective use of codified knowledge depend on the size of the clique. A large, highly cohesive group in the sense that its agents are well-connected and know the same language, will obtain larger benefits than will a loosely connected group.
IV-Codification as a path-dependent process
Diffusion of knowledge is easier in the context of stability, largely because many users know the languages and the fundamental models underlying them, and share the tacit knowledge necessary to use the codified knowledge. If many of the benefits of codification arise from diffusion, then the system within which knowledge is codified will display path dependence. Initially, as such a system develops, there will be competing languages and models. There is competition between paradigms in the sense in which Kuhn discusses them. (Kuhn 1962) But of course, as the benefits to codification increase, more people will have positive incentives to codify their knowledge. This implies that there will be changes in the incentives regarding which of the competing codes to learn; which of the paradigms to internalize. The snowballing effect present in many path dependent systems is evident (David, 1994).
IV-1 Percolation again
These considerations raise again the issue
of percolation. In the previous discussion the percolation probabilities
were implicitly assumed fixed. But of course this is not generally
the case. Percolation probabilities are determined at least by
the extent to which agents have tacit knowledge and so can read
the codes written by others. But agents have control over their
acquisition of this sort of tacit knowledge. And as with any other
decision it will be made on the basis of its economic value. Here
the feedback is clear. If there is a large community of active
agents generating codified knowledge in a particular language
using particular models, it will be of value to learn that language
and those models. If there are only a few such agents, the benefits
will be much smaller. Thus in the early stages of a fluid environment,
we can think of several different percolation networks coexisting,
one conjoined with each language and its correlative models. The
languages are being used to address similar issues in codification
(experimental economics in its current state might exemplify this
feature), and the percolation networks represent the possible
agents to whom the information can be diffused, and from information
can be acquired. If an agent wishes to address these sorts of
issues he has to learn at least one of these models or languages
(or perhaps generate his own). This is clearly the sort of positive
feedback process which generates path dependence. Of course, the
decision about which of the possible models to follow and thus
which of the languages to learn is not independent of the success
of each of the paradigms in addressing "interesting"
problems. But this success too, is endogenous, as is the definition
of "interesting", and endogenous in such a way that
there are at least some elements of the positive feedback that
conduce to path dependence. (See Dasgupta and David, 1995.)
IV-2 The history of engineering drawing
Belofsky (1991, p.45), in describing the history of engineering drawing writes: "It took about 100 years from 1795 to 1895 before the two dialects became completely fixed as standards in different countries. Chance may have been a major factor. If third angle had been in use before 1795 then the invention of descriptive geometry by Monge might not have led to its mis-application to engineering drawing in first angle. Also, if Monge's theory had not been so elegant and completely formulated it might not have appealed as much to academics on both sides of the Atlantic." Belofsky's study demonstrates several aspects of codification. First, it can be heavily dependent on underlying models, in this case, Monge's model of descriptive geometry which served as the basis for first angle projection. Second, there can be a long period in which the competing paradigms co-exist before becoming entrenched. Finally, the adoption of a particular paradigm is path dependent in that it can be affected by the context in which it develops (different angles in different countries) the time at which it develops (after the invention of descriptive geometry) and the milieu: in Europe, through the application of theoretical mathematics to a practical problem within the universities, in North America, the solution of a practical problem by practitioners of the art of engineering drawing. It is also interesting to note that in this case there is, to date, no winner in the competition between two techniques. This is largely due to the fact that for much of the competition, there were two distinct venues, Europe and North America. In each venue one technique won, but in doing so each generated a large community of users who had invested human capital in acquiring a particular code.
The codification of knowledge is central to the modern processes of dissemination. While recent technical changes have rendered codification easier and more valuable in certain situations, they have not changed the fundamental structure of costs and benefits. In particular they have not changed one critical feature, namely that the costs and benefits of codification depend to a great extent on the knowledge environment in which it takes place. A stable environment tends to have low costs and high benefits, while an unstable environment tends to have high costs and lower benefits. This difference stems from the fact that there are three aspects to codification, namely building models, constructing languages and writing messages. When models and languages exist, fixed costs have been paid, and only variable costs remain. When models and languages do not exist, they must be constructed, and this can be a costly endeavour. This tripartite analysis of codification further implies that the major costs of codification are born in the process of stabilizing the knowledge environment. What makes an environment unstable is that the languages and models contain ambiguity, and perhaps exist in multiple forms. It is the elimination of the multiple forms, and the further elucidation of the models that creates the stability. Thus these costs must be born during early on. But during this period, the instability implies that codified knowledge might be difficult to use, due to the ambiguity of the language, and might have only a small number of users, due again, to the fact that there are multiple interpretations, and presumably, the potential audience will be divided among the interpretations. When these descriptions hold, the process through which codification becomes valuable and therefore common, is path dependent. The value of it increases as the number of potential readers of the codes increases, but potential readers will learn a particular dialect or model depending on how valuable it is, that is, how much codified knowledge is written in it. Thus if the ability to codify knowledge depends on there being an appropriate language, it is possible that some knowledge is made difficult to codify due to early bandwagon effects in the development of models and languages. Unlike the case of fiscal deficits, however, in the development of codification, those who make the decisions (which determine the future path that codification can take) also bear the costs.
Abramowitz, M. and David, P.A. (1996) "Technological change and the rise of intangible investments: the US economy's growth-path in the twentieth century", in Foray and Lundvall (eds), Employment and growth in the knowledge-based economy, OECD
Arora, A. (1995) "Appropriating rents from innovation: a historical look at the chemical industry", in Albach and Rosenkranz (eds.) Intellectual property Rights and Global Competition, Berlin: ed.Sigma
Arora, A. and Gambardella, A. (1995), "The changing technology of technological change -- general and abstract knowledge and the division of innovative labour", Research Policy, vol.23-5
Arrow, K. (1974), The limits of organization, Norton
Baumol, W., Batey Blackman, S. and Wolff, E. (1989), Productivity and American leadership: the long view, Cambridge: MIT Press
Belofsky, H. (1991), "Engineering drawing -- a universal language in two dialects", Technology and Culture
Cook, T. (1995), "It's 10 O'clock: do you know where your data are?" Technology Review, January
Cowan, R. (1995). "The informatization of government as an economic opportunity"STI Review, n°16
Cowan, R. and Foray, D. (1995), The changing economics of technological learning, WP 95-39, IIASA
Dasgupta, P. and David P.A. (1995), "Towards a new economics of science", Research Policy, vol.23-5
David, P.A. (1987), "New standards for the economics of standardization in the information age", in Dasgupta and Stoneman (eds.) Economic Theory and technology Policy, Cambridge University Press
David, P.A. (1994), "Why are institutions "the carriers" of history? Path-dependence and the evolution of conventions, organizations and institutions", Structural Change and Economic Dynamics, vol.5, n°2
David, P.A. and Foray, D. (1994), "Dynamics of competitive technology diffusion through local network structures", in Leydesdorff and Van den Besselaar (eds.), Evolutionary Economics and Chaos Theory, Pinter
David, P.A. and Foray, D. (1995), "Accessing and expanding the science and technology knowledge-base", STI Review, n°16
Dosi, G. (1996), "The contribution of economic theory in the understanding of a knowledge-based economy", in Foray and Lundvall (eds), Employment and growth in the knowledge-based economy, OECD
Eamon, A. (1985), "From the secrets of nature to public knowledge: the origin of the concept of openness in science", Minerva, 23(3)
Eliasson, G. (1990), The Knowledge-base information economy, Almquist & Wiksell International, Stockholm
Ergas, H. (1991), The new face of technological change and some of its consequences, mimeo
Feldman, M. and Lichtenberg, F. (1996) Consequences and determinants of the geographic distribution of R&D: cross country evidence from the European Community's R&D information survey, paper prepared for the International Conference on Economics and Econometrics of Innovation, Strasbourg
Foray, D. and Lundvall, B.A. (1996), "From the economics of knowledge to the learning economy", in Foray and Lundvall (eds), Employment and growth in the knowledge-based economy, OECD
Gibbs (1994), "The crisis in software", Scientific American
Gunby, P. (1996), Explaining adoption patterns of process standards, The University of Western Ontario, Dept. of Economics, PhD Dissertation
Hatchuel, A. and Weil.B. (1995), Experts in organizations, de Gruyter, New York
Hummon, N. (1984), "Organizational aspects of technological change", in Laudan (ed.), The nature of technological knowledge. Are models of scientific change relevant?, Reidel Publ.
Konrad, K. and Thum, M. (1993), "Fundamental standards and time consistency", Kyklos, vol.46
Kubicek,H. and Seeger, P. (1992) "The negotiation of data standards: A comparative analysis of EAN and EFT/POS systems"in M. Dierkes and U. Hoffman eds, New Technology at the Outset: Social forces in the shaping of technological innovation, Campus Main
Kuhn, T. (1962) The structure of scientific revolutions, University of Chicago Press
Long, P. O. (1991), "The openness of knowledge: an ideal and its context in 16th-century writings on mining and metallurgy", Technology and Culture
Lundvall, B.A. and Johnson, B. (1994), " The learning economy", Journal of Industry Studies, vol.1(2)
Mac Kenzie, D. and Spinardi, G.(1995), "Tacit knowledge, weapons design and the uninvention of nuclear weapons", American Journal of Sociology , vol.101, n°1
Mangolte, P.A. (1996), Mémoire organisationnelle ou mémoires dans et hors de l'organisation?, draft, CREI, Université Paris Nord
van Meijl, H. (1996) Measuring intersectoral spillovers from IT and non-IT sectors: French evidence, paper prepared for the International Conference on Economics and Econometrics of Innovation, Strasbourg
Mulligan, M. (1994), "Speeding up the appliance of science", Financial Times
Nelson, R. (1992) "What is "commercial" and what is "public" about technology and what should be?", in Rosenberg, Landau and Mowery (eds.), Technology and the Wealth of Nations, Stanford University Press
OECD (1996), Technology, productivity and job creation: Analytical report, DSTI 95(4)
Putnam, Hilary (19 ) The Many Faces of realism Open Court Press
Rothenberg, J. (1995), "Ensuring the longevity of digital documents, Scientific American, January
Senker, J. (1995), "Tacit knowledge and models of innovation", Industrial and Corporate Change, 4(2)
Simon, H. (1982), "Programs as factors of production", in Models of Bounded Rationality: Behavioral Economics and Business Organization, vol.2, The MIT Press
Soete, L. (1996), "Globalisation, employment and the knowledge-based economy", in Foray et Lundvall (eds), Employment and growth in the knowledge-based economy, OECD
Steinmueller, E. (1991), The economics of alternative integrated circuit manufacturing technology: a framework and appraisal, CEPR publication, n°253, Stanford University
Steinmueller, E. (1995), "Neglected dimensions of the productivity paradox: users, complementarities and infrastructure", Toronto Conference, OECD
1 If many people possess the tacit knowledge, there will be a labour market that can be used to transfer the knowledge from firm to firm.
2 For a discussion of this phenomenon in the automobile industry, see Fortune, September 5, 1994, pp. 53-60.
3 It would equally be oversimplified to follow Nelson (1992) in arguing that only generic knowledge can be codified.
4 Given the obvious difficulties of measuring how much knowledge exists, no matter whether codified or tacit, and particularly since it is not obvious that a measure could exist even in principle, it is not clear exactly what such a claim would actually mean or imply.
5 We should note some preliminary attempts on the measurement side. For example, Van Meijl (1996) proposes to relate the importance of spillovers effects to the increased codification. Feldman and Lichtenberg (1996) test the effect of the codifiability of results of research programs on geographic concentration. Although promising, these approaches are still very preliminary.
6 This can be seen as an application of Gödel's incompleteness theorem.
7 It is perhaps useful here to use the taxonomy of knowledge-type elaborated in Lundvall and Johnson. They differentiate between know-what (referring to knowledge about facts); know-why (referring to principles and laws); know-how (referring to skills); and know-who (referring to the knowledge supporting indirect access to knowledge). Using this taxonomy, we can identify different logics and trends of codification (see Foray and Lundvall, 1996). One can make a further distinction within the know-how category (see Hatchuel and Weil, 1995), between doing know-how (referring to the generation of consistent pre-established prescriptions for action in specific contexts), understanding know-how (referring to the generation of plausible interpretations for some questions); and combining know-how (referring to the generation of a sequence of prescriptions which allow certain desired outcomes to be achieved). If the current codification movement on know-how takes the generic form of expert-systems, the type of codification will be specific to the particular kind of know-how to be codified. This variety in turn entails a diversity of codification forms, which are difficult to encapsulate within a general taxonomy. There are thus aspects of knowledge that are more readily codifiable because these are most closely identifiable with engineering principles and physical and chemical "laws" (Arora, 1995).
8 For a discussion of this and related arguments about the philosophy of science, and the philosophy of reference see the works of Hilary Putnam, for example, 1987.
9 The word "recorded" is deliberately ambiguous. Codifiers must understand the language through which the knowledge is systematized and stored. Users must understand the language in which messages are written. So, for example, developers of a medical diagnosis expert system must understand Lisp, but users of it must understand something like a dialect of a natural language.
10 There was a long list of potentially useful information to be included in the number. The constraint was that it was desirable to have few digits, since for the bar code to be useful it needed to be connected to a data base linking numbers and information. Clearly, the more information, and thus the longer the number, the more costly it would be to provide enough memory in the data base.
11 Of course, there were other technologies with which this information was codified, but the point here is the difficulty and cost of writing what appears to be a very simple language, able to express only very simple messages.
12 On the model of recombination and re-use, see David and Foray, 1995.
13 One notable exception exists in books of "recipes" for such things as random number generators with certain properties, search or sorting routines and so on.
section is drawn from David and Foray (1994).