ARTICULATION AND APPROPRIATION OF COLLECTIVE KNOW-HOW IN THE

STEEL INDUSTRY:

EVIDENCE FROM BLAST FURNACE CONTROL IN FRANCE

Nathalie Lazaric

Lazaric@idefi.cnrs.fr LATAPSES CNRS, 250 rue A. Einstein,

06 560 Valbonne Sophia Antipolis, France

Pierre-André Mangoltep.a.mangolte@wanadoo.fr

CEPN-IDE CNRS 99 av J-B Clément,

93430 Villetaneuse, France

and

Marie-Laure Massué

Massue@reims-ms.fr;

Reims Management School, 59 rue Tattinger

51100 Reims, France

Abstract

In this article, we use the implementation of an expert system to improve blast furnace control in the French steel industry to illustrate the problem of knowledge articulation/codification. Blast furnace related knowledge still largely takes the form of empirical know-how in general and expert know-how tied to specific individuals in particular. Therefore, the articulation/codification of knowledge in this field is a difficult task requiring the identification and selection of ‘best practices’ for the purpose of codification. This process, in turn, affects daily routines and creates new forms of non exclusive generic knowledge that make use of local knowledge. These new forms of generic information reinforce the tendency to appropriate private knowledge currently prevailing in Usinor, a large French steel company, and contrast sharply with the tradition of knowledge sharing between firms that predominated until the end of the nineteenth century.

Key words : blast furnace, articulation, knowledge, codification, steel industry.

Acknowledgments : This paper has benefited from various discussions notably in EAEPE conference (Berlin, November 2000) in Nice “Knowledge workshop” (December 2000) and in Paris Sceaux “New Economy” conference (May, 2001). The authors are grateful for comments from M. Becker, W. Dolfsma, F. Tell, P. Nightingale, J. De Bandt, P.Petit and N. Greenan among others. Usual caveat apply.

1

“It will at once be evident that on this point the position will be different with respect to different

kinds of knowledge ; and the answer to our question will be there fore largely turn on the relative

importance of the different kinds of knowledge ; those more likely to be at the disposal of particular

individuals and those which we should with greater confidence expect to find in the possession of an

authority made up of suitable chosen experts. If it is today so widely assumed that the latter will be in

a better position, this is because one kind of knowledge namely, scientific knowledge, occupies now so

prominent a place in public imagination that we tend to forget that it is not the only kind that is

relevant. It may be admitted that, so far as scientific knowledge is concerned, a body of suitably

chosen experts may be in the best position to command all the best knowledge available-though this is

of course merely shifting the difficulty to the problem of selecting the experts. What I whish to point

out is that even assuming that this problem can be readily solved, it is only a small part of the wider

problem” (Hayek, 1945, p. 521)

“Experts of any kind tend to look at the world in terms of a very limited number of variables-indeed,

that is a reasonable definition of what it means to be an expert. The training and experience of

experts equip them to deal with movement along some very particular trajectories, but not others. The

old aphorism that an expert is a person who knows more and more about less and less conveys an

important truth, one that has serious implications for the understanding of technological change”

(Rosenberg, 1986, p. 23)

Introduction

Much attention has recently been paid to the articulation and codification of knowledge

(Cohendet and Steinmuller, 2000; Cowan and Foray 1997; Cowan, David and Foray, 2000;), on the

grounds that "the increase in the stock of useful knowledge and the extension of its application are the

2

essence of modern economic growth” (Kuznets, 1966). The debate, which has been very heated, is

open to a variety interpretations as indeed are its implications (see Knudsen, 2000; Nightingale,

2001). Instead of reviewing the entire debate we will attempt to illustrate the degree and ways in

which it applies to the French steel industry. We believe that sectoral differences are an important part

of the story and could help explain the degree and potential of transforming different kinds of

knowledge, be they scientific or empirical (Rosenberg, 1982; Pavitt, 1984; Balconi, 1993, 1999;

Divry and Lazaric, 1998; Saviotti, 1998).

In the steel industry, the main challenge lies in decontextualising the local knowledge

anchored in experts. Such experts often belong to different ‘communities of practice’. They therefore

tend to use localised jargons and vary widely in the way they carry out their tasks and interpret

technical phenomena. In order to shed more light on these issues we will focus on the different stages

of knowledge articulation and codification and integrate organizational dynamic and social links as

driving forces in the process and not as minor variables. The reason for this is that cognitive and

political dynamics are interlinked and the nature of their co-evolution can be crucial to the evolution

of the knowledge itself ( Coombs and Hull, 1998 ; Rochhia and Ngo Mai, 1999; Simon, 1999 ;

Cohendet and Llerena 2001 ; Dosi, Nelson and Winter, 2001). Articulation and codification

transform the way in which communities habitually represent knowledge and share it between their

members at different levels: new knowledge representations come into play at both the individual and

the collective level, while new objectives concerning knowledge accumulation and knowledge

preservation enter the organizational level.

In this article, we first try to show the ways in which the articulation and codification of

knowledge undermines traditional daily routines relating to the handling of blast furnaces. We

emphasise the fact that the knowledge associated to the workings of the blast furnace is mainly

empirical and difficult to master in its entirety. We also review the basic concepts of routine,

knowledge articulation and codification in order to clarify them.

3

Secondly, we take our argument a step further by providing an overview of the French steel

industry. Codification in this sector was directly related to its recovery. In this context, we examine

Usinor’s organizational context and economic situation in some detail. Thirdly, we discuss the firm’s

Sachem project, which saw the introduction of an important programme of knowledge articulation and

codification in a number of stages.

Fourthly, we analyse the impact of this process, placing particular emphasis on the

organisational and cognitive effects of generic knowledge implementation. The crucial role played by

the blast furnace experts becomes apparent here. Similarly, the importance of human skills and tacit

knowledge for the system in its present form but also in its evolution also become clear.

Fifthly, we turn to the debate on knowledge in order to discuss the different forms of

knowledge involved in the process and to see whether this case study may demonstrate a change of

behaviour within the steel industry involving a new and more centralized locus of cooperation.

Finally, we conclude by reviewing the sectoral dynamic our case study highlights and the

organizational and social forces involved in knowledge articulation and codification.

I) Articulation and codification of empirical know-how in the steel industry

We begin by discussing why blast furnace related knowledge is still largely empirical in its

form, thereby increasing both the difficulties associated with its generalisation and the degree of

uncertainty in process control. Any attempt to disentangle this knowledge from individual and

collective practices by articulating parts of them disturbs daily routines. Articulation paves the way

for codification and can only be achieved by making the relevant practices explicit within different

“communities of practice”. We will examine these issues at both the empirical and the analytical level

and then attempt to devise a theoretical framework with which to explain such changes.

1) Empirical know-how and the blast furnace

4

The blast furnace is used for smelting and is capable of producing different grades of steel.

The procedure involves coke, charred coal, different types of ore, hot air and gas being introduced into

the furnace and then smelted. Dross is then produced by a process of “decarburization” and

“dephosphorization”. Since the contents of the smelted scraps varies, the melted metal must be

analysed immediately in order to determine which gasses should be added to it and at what

temperatures. Many of the thermal, chemical and mechanical phenomena taking place in this kind of

large reactor are far from being well understood ( Rosenberg, 1982) .

Although some of the physical-chemical reactions inside the blast furnace have been

described by mathematical models, scientific ways of describing the relevant chemical and physical

reactions remain incomplete. However, the scope of such models is severely limited by the fact that

most of the reactions cannot be observed. To summarize briefly, we can say that the blast furnace is a

very complex structure that relies on empirical know-how and still represents a ‘black box’, in that

very few physical or mathematical models have been developed to describe what happens inside it

(Rosenberg, ibid; Steiler and Schneider, 1994). This lack of understanding enhances the importance of

human judgement, tacit knowledge and labour skills. This last point is widely recognised in the

literature:

“One might naively regard a blast furnace as a deterministic chemical system, but in fact, its

behaviour is stochastic. Many aspects of a furnace, its interior lines, the placement of tuyeres, the

quality of raw materials, the degree of scaffolding, etc –exert an elusive but consequential effect on

fuel consumption. Consequently, if one builds a taller blast furnace it is not immediately obvious

whether its coke rate indicates the systematic effect of increasing height or is distorted by unusual

random circumstances” (Allen, 1983, p. 12)

Consequently, existing knowledge of the workings of the blast furnace may be difficult to

generalise because it relates to a number of different technological artefacts. In the circumstances, any

5

attempted generalisation is confronted with the problem of learning from partial experience involving

great uncertainties- a considerable difficulty for knowledge comparisons. This also means that the

day-to-day control of a blast furnace is based on a very costly trial and error process. Let us examine

this point in more detail.

The conversion process carried out by blast furnaces is continuous and takes up to 8 hours from

the introduction of the ores to the smelting stage. Any interruption is extremely costly and thus

prohibited unless an emergency occurs. However, a blast furnace tends to work more or less as

planned. Regular control is very important for both the smelting quality1 and the working life of the

blast furnace, which, on average, is 15 years.2 If the flow of ores is not regular, it can provoke an

above average erosion of the furnace’s internal brickwork and tank. It can also cause deficient casting

due to an iron notch. A number of problems can arise during this process, the most notorious

occurring when ores do not tap properly and flow on one side of the tank. This happens when ores are

insufficiently fluid and therefore create a kind of dome preventing gasses from moving up the

furnace’s throat. The ductility of the metal, which is affected by this problem, is a very important

quality of the final product and depends on the ability to carry out timely additions of oxygen and

other gases. If intervention is limited in any way, ores and smelt scraps can suddenly sink back and

cause a number of other problems, including obstructing the tuyeres and triggering explosions or gas

emissions. Two things must be checked continuously to ensure the smooth operation of a blast

furnace:

(1) the quality of the smelting scraps and their proper repartition inside the throat;

(2) the temperature inside the tank.

Team operators responsible for the continuous control of the blast furnace must be able to solve

problems and make quick decisions. However, this ability depends not only on the integration of

1

2

6

many pieces of articulated knowledge but also on the operators’ understanding of what makes a ‘good

process’ and what must be done to improve control. One way to achieve the latter is to reduce the

uncertainties associated with the process. To put it another way, in order to gain a better

understanding of what happens inside a blast furnace, the firm must open the “black box” and try to

analyse the causal links between the various parameters coming into play in different contexts and

situations.

2) Routinization, knowledge articulation and codification: a theoretical framework for

understanding the blast furnace

The starting point of our attempt to come to grips with the operation of a blast furnace is provided

by the concept of routine. In defining this notion, Nelson and Winter (1982) emphasize that a

“routinized” organisation can be understood by reference to specific competencies. These encompass

and embody different types of know-how, knowledge and part of the social context in which they are

embedded. Different forms of know-how are memorised using a variety of mechanisms (different

kinds of equipment, tools, procedures, data, human know-how, etc.). In order to illustrate the interplay

between various forms of know-how we use the notion of “repertoire”: just as a more or less talented

acting troupe interprets different plays in its repertoire more or less successfully, teams in the steel

industry avail of and can mobilise different kinds of know-how. These can remain inactive or, when

eventually activated, produce pig and cast iron.

An activated routine is an expression of the repertoire and can be judged on the basis of technical

indicators such as casting delivery, steel quality, raw material and energy input and process fluidity.

Collective activity involves procedures and rules and uses artefacts and know-how as well as skills.

The collective ability to solve problems and co-ordinate different incidents is highly dependent on the

existing repertoire. The co-ordination of different routines is sometimes compared to a “circuit” that

must work “smoothly” (Lazaric and Mangolte, 1999). This means that some degree of cognitive

coherence between different kinds of repertoires is necessary before they can operate successfully.

7

Effective co-ordination also depends on the relevant “motivational/relational context” (Winter, in

Cohen et al., 1995), which includes the “good will” of individuals, the prevailing interests and latent

conflicts and the discretionary aspect of behaviour within organisations.

Social relations and potential conflicts can, in fact, disturb the smooth operation of routines

and the expression of individual repertoires. Attempts to streamline individual behaviour in a

hierarchical structure through the exercise of authority or on the basis of particular incentive structures

have limited power, because forcing individuals to participate in a collective undertaking is difficult.

Individual agents have a certain amount of autonomy over their actions as well as their own way of

interpreting rules and modulating their effort, as Leibenstein (1987) put it. That is, although

individuals may have the competence to solve a problem, they may fail to do so simply because they

do not want to. In this case, a lack of individual and collective dedication can thwart cognitive co-

ordination and prevent the circuit from working smoothly.

Overall performance in the casting process is highly dependent upon:

(a) the state of cognitive repertoires, that is the accumulated stock of knowledge, and

(b) the social and relational context in which the repertoires are activated.

Most firms in this sector have looked into technical solutions, such as the implementation of

expert systems, in an attempt to improve blast furnace regularity. An expert system, however,

involves the transformation of all the knowledge stored in a particular firm, including its previous

repertoires, and therefore affects both organisational memory and routine activation. This raises a

number of questions as to where this knowledge is stored, who its carriers are, how it can be extracted

etc. As has been recently pointed out by a number of authors, this problem is far from trivial:

“It’s familiar enough that business firms and other organisations ‘know-how to do things-things like

computers to fly us from one continent to another. On second thoughts what does this mean? Is there

8

not a sense in which only a human mind can possess knowledge? If so, can this proposition somehow

be squared with the idea that organisations know-how to do things? And if organisational knowledge

is a real phenomenon, what are the principles that govern how it is acquired, maintained, extended

and sometimes lost?” (Dosi, Nelson and Winter, 2000, p. 1)

Indeed the maintenance of collective knowledge through the articulation and memorisation of

best practices is not neutral. It can generate specific assets for a firm by rendering the product of

human experience more “manageable” and by contributing to the selection of routines and practices

located within the organisational memory: “The degree of articulation of anything that is articulable is

partially controllable” (Winter, 1987, p.174). Let us examine this point further: in line with the quote,

we argue that knowledge is “articulable” and eventually “articulated” when the knowledge of some

person or some organisation is made explicit by means of language. “Articulated knowledge” can be

rendered explicit through language. Language, in this context, refers to a system of signs and

conventions that allow the reproduction and storage of knowledge in such a way that it can then be

communicated and transferred between individuals3.

The process of articulation involves the extraction of knowledge from the person carrying it and

the transformation of personal knowledge into a generic form (Winter, ibid; Mangolte, 1997).

Although some forms of knowledge can benefit from it, parts of tacit knowledge may defy articulation

and be poorly reproduced and communicated.4 In other words, only a small fraction of articulable

knowledge can in fact be articulated. Moreover, the degree to which articulation will actually be taken

up as an option may differ radically between firms, depending on the associated costs and benefits

accruing to a particular firm, on its strategic vision and the importance it places on the building of

capabilities (Teece, 1998 ; Zollo and Winter, 2000).

3

4

9

Articulation, however, is distinct from knowledge codification5 and may be considered as the

product of this preliminary step:

“Knowledge codification, in fact, is a more restrictive notion with respect to knowledge articulation

processes. The latter is required in order to achieve the former, while the opposite is obviously not

true. Actually the fact that in most cases articulated knowledge is never codified bears witness to the

argument of increasing costs to be incurred when scaling up the effort from simply sharing individual

experience to developing written reports, manuals or process-specific tools” (Zollo and Winter, 2000,

p. 17).

Following Zollo and Winter (ibid), we will take ‘routinization’, articulation and codification to

be three interlinked stages. The three forms can exist and evolve together. So, codification, while not

excluding ‘routinization’ and the presence of tacit knowledge, may well produce new useful insights

and therefore enrich old repertoires or create new ones. However, as we shall see in the case of

Usinor, this co-evolution is far from perfect because encoding has to face the problem of practice

selection in the presence of different technological visions. Moreover, relational and social aspects are

fundamental to the preservation of consensus between the different “communities of practice” co-

existing in a company6. Such communities play an active role by helping determine both the validity

and the content of the relevant knowledge and by contributing to or questioning the articulation

process. Let us discuss this point in the context of the French steel industry, an industry that has

undergone important industrial change during the last two decades.

II) The industrial and organizational contexts of knowledge articulation in Usinor

In order to understand the French steel industry in general and its recovery in particular, we start

by presenting a concise overview of the sector. The creation of the Usinor group and the associated

integration of different companies generated a number of opportunities for changing old routines. In

5

6

10

the circumstances, it was thought that the articulation and codification of old practices would be

valuable in that it could provide a common tool and lead to a shared vision of the blast furnace

process.

1) Concentration in the steel industry

The need to articulate and codify Usinor’s know-how emerged at the end of the 70s, a time of

serious economic difficulties for the French steel industry.7 Shrinking markets and the emergence of

new competitors made substantive improvements in productivity and product quality crucial to the

survival of the industry8. The sector, in France as well as abroad, faced a protracted period of

structural excess capacity during the first half of the 1980s. This triggered a selection process and,

later, a series of mergers amongst the survivors aimed at consolidating existing industrial positions. In

this context, Usinor, using state aid available at the time, absorbed a large part of the existing French

companies. So, for example, Sacilor, an old rival, became part of the group in 1986 (see table 1). The

French State, Usinor’s majority shareholder, progressively disengaged from Usinor following

privatisation in 1995. This change also enabled Usinor to take over foreign firms, such as Arvedi in

1998, Ekosthal and Cockeril in 1999.

As the above table shows, the creation of Usinor is a recent event. The group is the outcome of a

combination of various formerly competing organisational entities. The creation of a global strategy

for a firm of this kind was far from trivial both because local plants were very decentralised and

powerful and because top management was slow in centralising its own activity (Godelier, 1997). The

participation of the French State as principal shareholder and the introduction of new managers that

had previously been working in very different sectors of the economy also triggered some important

changes. Moreover, the firm’s staff composition changed as employee turnover inside the group

increased and a new organizational culture was introduced by Mr Levy, the company’s new CEO,

7

8

11

brought in from the petroleum industry. Over time, Usinor became increasingly centralised, as

common goals, practices and a common technological vision gradually took hold and introduced

some unity between the various plants. Creating a shared understanding of the technology and a

shared representation of the blast furnace was part of this strategy.

2) The need to create a coherent organisational entity inside Usinor

By the mid 1980s, the group had still not become an effective organizational entity and experts

using different routines had yet to develop a sufficient degree of mutual understanding. Decentralised

plants and the pervasiveness of forms of knowledge associated with the localised and sometimes

conflicting experiences of working with individual blast furnaces, led to the continued existence of

different ways of understanding the same problems. Although both articulated and tacit knowledge

were present, they varied across plants and different communities of practice co-existed.

Consequently, a variety of dialects coexisted within the company and, with them, different beliefs and

know-how about blast furnace operation. Each plant depended on the personal vision of its blast

furnace experts, the authoritative figures discussing, articulating, transmitting and validating

knowledge.9 These experts were a lot more powerful than their hierarchical grade and status inside the

plant warranted. Another crucial problem for the company was memorisation of knowledge in a

context of rapidly declining manpower.

Overall, four major reasons prompted Usinor to articulate and codify its knowledge:

1) The need to create unified capabilities and articulated knowledge across the group and to channel

all local knowledge into a common project in order to create shared semantics between experts

(“to be sure that we are talking about the same thing”). Moreover, apart from the linguistic

problem, the firm had to establish a shared tool, which would ensure problems were dealt with in

similar ways and lead to a common understanding of blast furnace operation. To summarize, the

9

12

firm had to unify the variety of dialects present in the different “communities of practice” co-

existing within Usinor.

2) The firm made 98,000 employees redundant between 1977 and 1990 (mainly through a

retirement policy). Following this period, the company hired very few new employees and

therefore, the average age of the group’s employee population rose dramatically (around 50% of

Usinor’s employees will have retired by 2010). In this context, the need to save existing

collective knowledge became crucial, given that most of its human holders were soon to

disappear!

3) The need to improve the quality of the process by introducing real time intervention, thus

ensuring the rapid resolution of problems occurring during the early stages of metal fusion (the

cost of guiding the blast furnace in a plant like Sollac represents 56% of the total costs of steel

production).

4) Usinor’s merger and acquisition activity increased the concentration of R&D centres. For

example, IRSID,10 which was meant to carry out fundamental research for all French firms in the

steel industry, became, as a matter of fact, an integrated Usinor R&D centre and a de facto a

private R&D centre. This large laboratory had extensive experience in the domain of artificial

intelligence and played an active role in the implementation of the Sachem Project.

We shall return to this latter project and examine it in more detail in the next section.

III) The Sachem project in Usinor

The background to our case studies will help us provide a good depiction of the Sachem project.

Following this preliminary stage, we will describe the gradual constitution of an ‘epistemic

community’, in charge of establishing a shared vocabulary among the ‘community of practice’. This

10

13

vocabulary plays an active role in the articulation of knowledge and helps identify the critical

practices. Codification takes the process a step further by selecting the ‘best practices’. The problem at

hand is to create models that can codify such practices in the first instance and then re-translate them

into natural language so that the proposed system can be agreed upon and socially validated by the

members of the organization.

1) The methodological approach

In order to collect consistent information about Usinor’s Sachem project, we had to carry out a

series of face-to-face interviews during the course of a year (in fact, between August 1997 and June

1998). During this first stage, we tried to collect information about Sachem from as many

representatives of the sets of actors involved as possible. We therefore conducted interviews with:

- two blast furnace experts,

- the Sachem manager,

- a knowledge manager,

- a R&D department manager, two engineers and a researcher from IRSID,

- a representative individual from Sachem’s user group (composed mainly of operators and

technicians), and

- a retired chief executive of the human resources department.

Each interview took a total of 4 hours on average: a first long face-to-face session (between

two and three hours long) was supplemented by a further discussion at a later stage aimed at

completing the previous interview (mostly face-to-face but also via telephone in some cases).

It is important to note that all interviews were summarised and incorporated into a general

research report. This was then discussed with the project’s manager in order to validate its general

vision and technical aspects, but also in order to avoid the dissemination of confidential information

14

contained in the Sachem methodology. Following the collection of the qualitative information, we

also gathered general information from newspapers and books in order to enhance our understanding

and knowledge. In October 1999, a member of our team finalised important details with Sachem’s

manager. Discussions at this stage were informal and focused on specific issues on which we required

detailed documentation (for information we felt was incomplete, notably training policies).

Let us now turn to the Sachem project and trace it from its infancy to its maturity during the

1980s in Usinor.

2) Knowledge articulation in Usinor

The tradition of memorising long standing practices in order to improve processes is not new in

the steel industry. In France, many books provide descriptions of the empirical practices associated

with the blast furnace.11 In the case of Usinor, knowledge was articulated through the launch of

‘Xperdoc’ in 1989. This established a common word-list and was aimed at creating a shared

vocabulary among blast furnace experts. Shared semantics were seen as the first step in the process of

harmonising different practices, beliefs and repertoires and were meant to constitute a common

reference for the group. This “hand-book”12 was used by the experts and created an “epistemic

community”, which, in turn, generated discussion and the mutual acknowledgement of different

practices.13 “Xperdoc” had the following objectives:

(1) To bring the experts together by giving them the opportunity to meet (“Xperdoc” in this way

gave them a chance to exchange views and discuss their practices);

(2) To establish an acknowledgment of different practices and help institute personal trust

between experts.

11

12

13

15

This latter point was far from trivial: the creation of a ‘common reference’ depended on a

prior selection of best practices, which was difficult to implement. The firm needed to establish a

climate of trust before it could achieve a real acknowledgement of practices among experts. As

Steinmuller recognises:

“The availability of collective memory within an epistemic community also allows the reduction of

remembering solutions to problems that have been previously discovered and documented. This

permits a reduction in effort devoted to re-invention. For such gains to be realised the codification of

solutions need not to be complete. Instead the purpose of the collective memory is to ‘signal’ the

availability of the previous work and a solution that might arise again within the community. The

gains from this process are both collective and individual and therefore are incentive compatible with

efforts to contribute to the group memory project. Note that it is important for the group to have some

degree of trust and mutual regard for such system to work. Otherwise individuals will not regard the

‘solutions’ or ‘ideas’ of others to be worth the time it takes to become aware of and evaluate them in

a new context” (Steinmuller, 2000, p.367).

‘Xperdoc’ brought changes to the usual ways of doing things and was understood in different

ways by the experts who had shaped their personal vision through their own experience of the

workings of the blast furnace. Moreover, Xperdoc produced a radical change in organisational

memory and in the knowledge activated daily in the firm, by questioning old repertoires and social

links surrounding the activation of these repertoires.

3) Identification of ‘best practices’

In 1987, the Sachem initiative came to light mainly due to the efforts of a part of the staff who

believed that artificial intelligence could help memorise a large part of the knowledge held by experts.

This idea was only implemented in 1990 under the supervision of Francis Mer, the company’s top

16

manager, who was particularly interested in this new tool in his efforts to improve Usinor’s

productivity.

During the course of year, over a hundred artificial intelligence applications were tested in

collaboration with Usinor’s R&D centre, IRSID. The implementation of artificial intelligence had

benefited from European subsidies and a 40-strong team spent 5 years working towards the creation of

systems and their diffusion inside the group. Following this first stage, 17 technical solutions were

selected, among which the Sachem project.14 This latter was first implemented in October 1996,

following a long period of discussion within the company which focused mainly on the following

crucial questions:

- What kinds of knowledge have to be articulated and stored?

- How should practices be selected?

- How can such practices be transposed into a new tool?

- How can losing tacit knowledge be avoided as the importance of articulated knowledge increases?

In order to identify the ‘best practices’ and key ‘know-how’, 13 experts were chosen among

those who had co-operated in the writing of “Xperdoc”. Experts were selected according to know-how

and location in order to ensure a fair representation of the various types of knowledge prevailing in

the different plants (Sollac Fos, Dunkerque, Lorfonte Patural, etc.). The team worked with six

“knowledge engineers” in order to extract the “core know-how” and articulate it (400 interviews were

conducted).

Before the extraction process took place, an interpretative model15 was created in order to gain

an understanding of expert know-how. Blast-furnace experts based this interpretative model on a

collection of representative case studies. The samples exemplified the different ways in which experts

14

15

17

understood the blast furnace and the ways in which they solved problems. This preliminary work took

10 months (from September 1991 to June 1992) and allowed identification of the crucial know-how

activated daily. Consequently, the perimeter of expertise was delimited and the relevant expertise was

diffused among the experts. This process of disclosure also entailed a selection of knowledge from the

activated daily know-how. Two important conditions had to be fulfilled before knowledge could be

selected, validated and eventually codified.

First, knowledge had to be generic, that is to say, it had to be sufficiently broad to remain

meaningful once it had been dissociated from its local context and specific use. Variety amongst

experts and their associated expertise, which stemmed from different plants, helped identify the

broader strands of knowledge.

Second, the knowledge had to be identified and acknowledged as ‘true’ by all the experts

involved. The knowledge had to be deemed useful to and employable by operators before it was

codified. A knowledge manager clearly recognised this point: “the automation of knowledge is the

transformation of true expertise into useful one”. Consequently, at this stage, parts of the pre-existing

knowledge were lost while others were accentuated (validation reinforces the parts collectively judged

as crucial to the detriment of those deemed too tacit or local to be selected).

4) Selection and validation of ‘best practices’

The selected know-how was classified and indexed between April 1992 and May 1995.

During this stage, called AED,16 key concepts and their articulation were registered in order to

introduce consistent problem solving. For example, 150 potential blast furnace anomalies were

identified and classified according to their position in the steel making process (quality, gas, wall,

temperature decline, permeability, etc.). On the basis of this knowledge, specific problems (notably an

unusual temperature increase) could be identified systematically, operators alerted and specific

recommendations made on the actions to be implemented.

16

18

Following the process of translating words into codes, the stage of knowledge

acknowledgement and validation was launched.17 Blast furnace experts had to recognise their codified

know-how, which had been radically transformed by computation. This stage was crucial, as it

allowed experts to verify whether the codes did in fact represent what they had intended to articulate

in the first place. Knowledge validation was a long and difficult stage as consensus had to be reached

before it was concluded. Individual meetings, which saw experts having one-to-one discussions with

knowledge engineers, and a collective one, including all the experts, took place. Local know-how had

been radically transformed as the “knowledge engineers” had changed the way experts represented

their own expertise. As it happened, parts of the general knowledge codified in the expert system had

ceased to be meaningful to some of the experts. As a result, the knowledge engineers designed a

linguistic model (in natural language) in order to translate the code. The experts were thus able to

recognise their own expertise and acknowledge it collectively. Similarly to the interpretative model,

which had translated the experts’ articulated knowledge into codes, the linguistic model converted

codes into words. Different models had to be created because different levels of abstraction and local

knowledge were required depending on their use.

The different levels of knowledge created by Sachem are depicted in the following table (table

2), which also shows the break introduced by the automation of knowledge.18 The different stages of

the Sachem project are summarised in the table 3.

IV) Cognitive and organisational consequences of the Sachem project

The consequences of articulation and codification were very important because actual

knowledge content changed drastically, thus forcing some blast furnace experts to modify long

standing beliefs and their usual interpretation of technical phenomena. One would expect some degree

17

18

19

of resistance to this process. In fact, co-operation in the knowledge identification and validation

processes proved crucial and was successful because the blast furnace experts were acknowledged as

the organizational knowledge carriers and their existing expertise was validated inside the ‘epistemic

community’. The company also allowed the experts to play an active role in the adaptation and update

of the system in order to integrate minor changes and improve its daily performance.

1) Deep transformation of the content of knowledge

Articulation and codification entails a radical change in knowledge because it involves the

selection of parts of all available know-how. Moreover, its affects the content of knowledge, as, in

practice, the traditional expertise anchored in an expert’s routines is alive. A first transformation

occurs when experts put their practices and parts of their tacit know-how into words. This

‘explicitation’ creates articulated knowledge, which entails a first selection of know-how. Parts of

know-how are too dependent on practices prevailing in local plants: these cannot be articulated and

resist extraction because of their ambiguity and fragility (deeply tacit knowledge and personal

judgement may defy codification by virtue of being too personal). As Dreyfus emphasizes knowledge

may be difficult to disembody :

“Experts generally know what to do because they have a mature and practiced understanding. When

deeply involved in coping with their environment, they do not see problems in some detached way and

consciously work at solving them. The skills of experts have become so much a part of them that they

need be no more aware of them than they are of their own bodies” ( H. and S. Dreyfus, 1986, p. 44).

A second transformation takes place when articulated knowledge is turned into code, as

technicians in the area it affects have their own ways of representing and selecting knowledge: parts of

knowledge may be deemed useful simply because of the nature of particular technical parameters

embedded in the expert system. In other words, the nature of the container is far from being neutral

and can in fact change the knowledge content itself by including unnecessary bits of know-how while

20

excluding others. Consequently, the outcome is not a simple translation of existing knowledge into

code but also a reformulation produced by the knowledge engineers and validated by the experts. Each

stage of the transformation, which takes place in the handover from live expertise and activated

knowledge to the memory of an outsider and from one outsider to another, entails a change in the

preserved knowledge. This is neither a perfect equivalent nor a total substitute of the knowledge

carried in the different memories. As a knowledge engineer recognised:

“It is the same relation that the painter has with the nature he observes. He magnifies it. The

knowledge engineer describes the know-how of the expert; his art is to capture live expertise in order

to reveal it and make it consistent outside of him”.

Hatchuel and Weill (1992) brilliantly describe this deep transformation of knowledge and

argue that the container transforms knowledge content because each language has its very own ways

of representing things. Repeated transmission through a variety of languages will always involve

some losses as codes differ radically across languages. Moreover, articulation and codification are in a

way unpredictable because they are largely based on individuals’ willingness to participate in a

process that is likely to depart from their initial experience (most implemented codes differ

substantially from the original individual representations of particular technical problems). This is

precisely the reason why the translation into natural language and the validation by experts that took

place following the codification process was crucial to the Sachem project: it prevented experts from

feeling they lost their original know-how once they had passed it on to the knowledge engineer.19

2) Revision of prior beliefs and reinforcement of empirical know-how

An important contribution of the Sachem project was that it brought about a change in the

way operators and experts themselves understood the blast furnace process. As we have already

emphasised, the blast-furnace experts’ know-how is generally empirical and has not yet be captured in

19

21

its entirety by a scientific model. Implementation of the Sachem codes meant that some parts of pre-

existing know-how were validated and acknowledged, whereas other prior beliefs were discussed and

collectively rejected. In this way, the newly validated know-how, a subset of that previously

employed, in itself contributed to the improvement of the company’s empirical knowledge.

For example, before Sachem was set up, a phenomenon called ‘fluidisation’, involving an

increase in temperature above the ores was often observed. This was connected to the suspension of

coke iron ore and limestone flux and was associated with a fall in different ores. In the ‘Fos sur Mer’

plant two beliefs prevailed about this phenomenon:

Belief 1: Experts believed that fluidisation never occurred in ‘Fos sur Mer’

Belief 2: When descending ores and fluidisation were detected, experts did not associate it with

fluidisation and the temperature increase was attributed to a different phenomenon.

After Sachem came into use, prior beliefs were revised, notably:

Belief 1: Fluidisation did take place in ‘Fos sur Mer’

Belief 2: When fluidisation occurred, fluidisation had preceded the descent of ores by an hour.

In practice, the system provided a robust tool against which empirical know-how could be

tested in order to improve and generalise it. As the process unfolded, prior beliefs were gradually

redefined: what had been known to be true started appearing to be only partially so (especially

through the discovery of new causal links between technical events). The causal links connecting

separate technical events, which used to be tacit and intuitive, were tested and systematically proven.

By this way, the articulation process led to more robust beliefs and unquestionable know-how within

the firm by changing the experts’ and operators’ local cognitive representations. The Sachem project,

also led to the creation of generic knowledge. However, this was not simply based on technical

parameters but depended vitally on the social co-operation of the blast furnace experts, the main

22

holders of this knowledge. In other words, the way in which this tool was used was crucial to its

survival, as we will see in the following section. Usinor’s management was aware of these

fundamental issues and let the experts play an active role in this process.

3) The co-evolution of tacit and codified information: a continuously updated system and a

permanent activation of individual skills

Sachem is able collect systematic data (with the aid of over 1000 sensors placed inside the blast

furnace), interpret and qualify them. The system continuously compares the collected information

against a reference situation and can, in this way, make certain predictions, detecting problems and

making recommendations to operators. It provides a global vision in real time, a selection of

interesting indicators and some solutions. Operators retain a certain amount of autonomy in decision-

making, as they are able to ask the system to provide explanations on a number of the phenomena it

detects. The system has been set up in such a way as to avoid automatic guidance, which would risk

reducing operators to a passive role and thereby undermine their ability to solve problems.

In order to stimulate the operators’ energy and attention, the company introduced an important

training policy (‘off the job training’). This was put in place in 1995, before Sachem had begun

operating effectively. In 1995, a team of operators and blast furnace experts was created in order to

anticipate potential problems with Sachem and to encourage interaction with the new technological

artefact. The training policy put an end to the old apprenticeship system and particularly affected the

transmission of know-how from experts to newcomers.20 Prior to the introduction of the expert

system, it took 10 years of ‘on the job training’ to become a confirmed operator. Now only 3 years of

work within the firm are required to reach operational status and 5 to become a confirmed operator.

The new training policy meant that experts and operators themselves had to gain an explicit

20

23

understanding of their own know-how. In fact, they were required to recognize why they were solving

particular problems and had to justify their repertoires. This transformed their knowledge and

expertise, that had been largely based on “knowing how” and activated new pieces of detached

knowledge, based more on “knowing that”21.

The training policy was implemented in order to avoid a “cognitive prosthesis syndrome”. In

other words, it was meant to ensure that operators and experts remained constantly vigilant and active

and did not become too confident in the new technological artefact. In fact, the system still requires

direct input from users and blast furnace experts for its updates. Moreover, the system’s

recommendations resulting from its interpretation of the data (especially the identification of causal

links between different events) and the operators’ ultimate decisions are still compared and

scrutinized. The gaps between the system’s recommendations and the operators’ decisions are

systematically deconstructed in order to detect divergences. This analysis allows the database to be

enriched and updated. Parts of the tacit know-how that were not articulated by the operators and blast

furnace experts and were considered insignificant during the second stage of articulation (in 1992),

are gradually incorporated into the system. Updates are formalised in an annual meeting with the

operators, foremen and experts. Experts play a crucial role in this context as they formalise and

systematise the results of the divergence analysis of users (operators). Moreover, they introduce new

articulated knowledge that will later be turned into codes by the knowledge managers. The system’s

knowledge base is therefore constantly enhanced by new articulated knowledge, a process that

prevents it from becoming rapidly obsolete. The constant activation of human skills is very important

because, as Dreyfus reminds us, the coupling of human skills and machine capacity is crucial in order

to make use the expert system’s entire potential (Dreyfus, 1972). In fact, the expert system is unable

to deal with new situations and solve novel problems. Its abilities are limited by its existing,

previously articulated, knowledge. In the absence of integration of new pieces of knowledge and of

their codification, the system would, in the long run, become outdated.

21

24

Let us now turn to the impact the creation of generic knowledge had not only on the Sachem

project but on the steel industry as a whole.

V) Knowledge appropriation in the steel industry : an open discussion

Usinor’s articulation and codification of knowledge is also symptomatic of the reinforcement of

private appropriation within the steel industry. This point has to be emphasized because it stands in

stark contrast to the ancient tradition of knowledge sharing and ‘collective invention’, which prevailed

at the end of the nineteenth century. Nevertheless, the very same period that saw knowledge become

internalised in Usinor also witnessed the establishment of new areas of co-operation. These displaced

the ancient locus of ‘collective invention’ (now tightly located within Usinor’s epistemic community)

and put the creation of generic knowledge in its place. We shall examine these different aspects more

depth.

1) Knowledge sharing and ‘collective invention' in the blast furnace: changes in firm behaviour

during the twentieth century

The steel industry has attracted the attention of many observers because its traditions of

knowledge dissemination inherited from the past are very important. Allen (1983) introduced the

notion of ‘collective invention’ to characterise a knowledge sharing attitude whereby inventors

disseminate their knowledge instead of protecting it, in order to benefit from knowledge disclosure. In

the United States as well as in England, the greatest improvements in the blast furnace process were

achieved primarily through classic ‘learning by doing’ and local experimentation rather than research

and development inside firms:

25

“During the nineteenth century, non profit institutions and firms undertaking R&D were of little

importance as sources of invention for the iron industry. The British and American governments

funded no research, and university faculty published scarcely any findings. Firms did not maintain

research departments and seem rarely to have expended any appreciable quantity of resources with

the main intent of advancing knowledge or improving products and processes. What is more

remarkable is that the independent inventor was also not the major supplier of new inventions to the

iron industry. Some major improvements can, of course, be attributed to famous inventors –to the hot

blast to Neilson, the Bessemer converter, and to the basic process to Thomas and Gilchrist. However,

if one examines a sector like the blast furnace industry and determines the inventions whose diffusion

were important for the growth in efficiency, it proves impossible to attribute their discovery to any

single inventor. Certainly, no one received a patent for many of these advances. Thus the increase in

furnace height and blast temperature that were so important for productivity growth in England’s

Cleveland district evolved through that actions of many individuals over a twenty years period.(…)

Such experiences are usually dealt with by labelling them as ‘the accumulation of minor

improvements’ ”(Allen, ibid, p. 1-2).

Allen showed how potential entrants could benefit from this knowledge during the building of

a new furnace: the information was available to both professional associations and the labour market,

where rotation contributed to the diffusion of knowledge. So, many advances in the steel industry

were the outcome of collective efforts. For example, the idea that capping the blast furnace could lead

to the use of waste gases as fuel or for the reduction of limestone was invented collectively (Allen,

ibid).22 In this context, knowledge sharing, far from being an exception, was the generalised rule:

“Each individual has some cherished bit of knowledge, some trade secret which he hoards carefully.

Perhaps by sharing it with others, he might impart useful information; but by an open discussion and

interchange he would almost for certain, learn a dozen things in exchange for one given away.

22

26

General increase of knowledge would give general improved practice, most likely a larger use of the

materials in which a manufacturer is interested” (Muntz, 1909 in Allen, ibid, p. 19).

This behaviour evolved when oligopoly became the rule in manufacturing in every country.

By this way the transformation of the industry in the twentieth century, aimed to take advantage of

economies of scale and scope, led firms to concentrate their efforts on private research and

development programmes. In this new context, collusion and knowledge exchange ceased to be the

rule and were gradually replaced by secret policies and closures.

In France, the jealous protection of secrets eventually predominated although it proved

difficult to implement. Leaks23 and the disclosure of generic information to technical and professional

associations did not disappear altogether (Jollivet, 1999).24 In practice, however, the concentration of

French firms and the change in their legal status during the 70s led to the gradual reinforcement of

private knowledge appropriation, not only between rival firms but also within companies. At the same

time, an important knowledge memorisation programme was implemented and coincided with the

creation of internal R&D centres. The IRSID R&D centre for example, now integrated in Usinor,

directed its resources towards the company’s exclusive needs, despite its historical mission, which

was oriented towards knowledge disclosure to the French steel industry as a whole.25

In Usinor, the change of attitude towards knowledge dissemination was in line with the

centralisation movement that took place inside the firm and was aimed at building a common identity

within the group (see above) and to increase blast furnace control.

Let us now discuss the different states of knowledge present in the company in order to assess the

changes that followed the introduction of generic knowledge into the different communities of

practice.

23

24

25

27

2) Generic, local and universal knowledge in Usinor: a discussion

Sachem generalised local know-how and selected some of the daily-activated routines. The firm

benefited from this transfer of knowledge and has, in the last few years, improved its productivity,

increased operator reactivity, seen a global improvement in casting quality and a decrease in steel

making related incidents. In 1996, these gains were estimated to amount to 0.18 % of global

turnover.26 From a long-term perspective, the creation of strategic assets involving the memorisation

of most of the company’s know-how is difficult to evaluate in simple monetary terms. The creation of

‘generic knowledge’ is also a strategy of ‘knowledge disclosure’, which prevents the company from

losing crucial technical know-how and helps improve its capabilities. Generic knowledge is key to

this process as it broadens the original know-how created inside the various communities of practice.

As Nelson emphasises:

“In many cases specific practice is buried deep in the structure and custom of the host institution and

is difficult to disentangle. In this cases technology, or at least a particular technique, is more like a

private good that a latent public good (....) Generic knowledge not only tends to be germane to a wide

variety of users and user, but technological communities and professions tend to grow around such a

knowledge, which then become the stock in trade of the insiders of a field” (Nelson, 1993 p. 12).

Generic knowledge can also be applied to a broad range of domains that lie outside its original

locus of creation (i.e. the original community of practice). Our case studies show, however, that

generic knowledge is not a public good.27 According to the literature, for knowledge to qualify as a

public good, any agent should be able to use it as universal information without incurring any costs

and without having to make any investment in its duplication. When knowledge is considered as a

public good it is seen as a set of ideas or universal information that can be spread indefinitely to a

large number of agents. This simple step (consisting in the reduction of knowledge to information and

26

27

28

to a quasi-public good) proves very problematic with Sachem. What we have observed is that the

creation of generic knowledge does not generate a free access to technology, but on the contrary,

enhances collective appropriation within a particular group. This means knowledge held by the

‘epistemic community’ cannot be simply reduced to “knowing that” and conscious information.

Moreover, “knowing how” is still activated in the update of the system, as an ingredient aimed to

avert the system’s rapid obsolescence, as we saw above.

In effect, Usinor must preserve its specificity and technological advantage vis-à-vis its

competitors and generic knowledge is a means to achieve it. Consequently, the articulation and

codification processes are also compatible with the reinforcement of control required for firms to

benefit from Schumpeterian rent (Teece 1998).28 This is one of the reasons why generic knowledge in

the steel industry may be articulated, codified and tightly controlled, notably inside the ‘epistemic

community’. Compared to the old dialects and practices prevailing in individual plants before the

implementation of Sachem, generic knowledge is, of course, more likely to be redeployed and

articulated into more widely shared categories. It is therefore more likely to be shared and

homogenously distributed and also more likely to be centralized (Lazaric and Marengo, 2000). This

kind of knowledge usually stands in contrast to the local knowledge involved in daily practices and

dispersed inside the firm, as Fleck recognizes:

“The local practical knowledge is highly contingent to the particular firm concerned. It deals with the

specific ‘knowledge base’ built up over many years of experience in carrying out the firm’s business.

Moreover, this local component is usually distributed among many operatives involved in day-to day

activities. In many cases, this knowledge is tacitly embodied in skills and practices which have been

gradually formed over a period and are not available in any other centralized or formalized form.”

( Fleck, 1994, pp. 641-642).

28

29

The differences between the different states of knowledge (local, generic and universal) are

discussed and summarized in the following table (table 4).

As the above table shows, the implementation of the Sachem project introduced a certain

amount of generic knowledge and therefore reduced the dissemination of collective know-how inside

the industry. The centralisation of best practices that followed Sachem is a good example of a new

compromise between tacit and universal knowledge and a change in the locus of collective invention,

previously dispersed in many different organizational entities and spread out in the sector.

Conclusion

In our view, the articulation and codification of knowledge in the French steel industry is a

manifestation of the new behaviour that has come to dominate the sector. This is driven by the

willingness to capture value from knowledge assets, to improve understanding of the knowledge

associated with the workings of the blast furnace and to increase the furnace’s productivity and

reliability. One could conclude briefly that this is an example of the end of ‘collective invention’ in

the sector: collective know-how is now more likely to be held by private firms intentioned to protect

their knowledge instead of sharing it with their competitors. In effect “collective invention” has been

gradually displaced around the ‘epistemic community', preventing knowledge dissemination among

steel competitors. In Usinor’s case, it was a new technical tool that changed the role of the blast

furnace experts and created non-exclusive knowledge. Experts have now ceased to be the only

disseminators of knowledge and have a new part to play in the update and maintenance of existing

organizational knowledge. The organisational consequences of this process are not neutral, of course,

as they create centralised know-how and generic knowledge, parts of which could be tradable outside

the firm. In other words, capabilities have changed: they are now less dependent on human holders

and are anchored to the organisation and its shareholders.

30

Social context and managerial input play an active role in changing the content of knowledge

(from personal and private to shared and generic), by creating the climate of trust and mutual regard

required to achieve true co-operation between individuals and active participation. Participation,

which is essential to the creation of a new representation of the blast furnace, involves renouncing old

beliefs (concerning technical events such as fluidisation), once prevalent amongst different

‘communities of practice’. The system relies substantially on the permanent combination of local and

generic knowledge for its improvement, as parts of it are revised by adding new causal links between

technical events. The implementation of a training policy has helped this combination by allowing the

system’s performance to improve through permanent activation and update. In order to prevent a rapid

fossilisation of the company’s know-how, live expertise, human skills and tacit knowledge are all

required: they are fundamental to moving the system beyond the stage of codification and are crucial

to its constant enrichment. New pieces of knowledge, which originally were deemed to defy

articulation, were in fact articulated over time as new causal links between technical events were

discovered. Nevertheless human beings and machines are not substitutes but complements and human

ability is always required for solving new problems: generic and local knowledge are two faces of the

same coin and are both necessary in order to improve the technological tool and organizational

repertoires inside the company.

________________________________

Notes

A “speedy guidance” produces a molten metal that is “too hot” (i.e. including too much silicon), whereas a

“slow guidance” can produce a “cold” molten metal (with insufficient silicon and too much sulphur), which is

unusable.

31

2 Blast furnaces are extremely expensive: building one costs around EUR 300m and repair costs can reach

EUR 150m. The casting process alone accounts for 56% of Usinor’s average steel making costs. This is one of

the reasons why improvements in casting and blast furnace guidance are crucial.

3 Natural language has a more precise definition because its grammar produces a certain degree of consistency

between the various signs. On this important debate, see Choamsky (1988), Piaget (1970) and Polanyi (1958).

4 Traditional know-how may be difficult to articulate as it is affected by the way in which its “master”

transmits his know-how during training and apprenticeships and because of the presence of important tacit

know-how. For example, Verry (1955) reviews the sensorial know-how involved in the steel industry’s casting

process. Before 1914, rollers in the Ardennes region were directly involved in the temperature control of metal

before it was put into the rolling mill. They directly observed the colour of the metal, which can take a variety

of colours ranging from pale straw (225°) to brownish purple (260°), dark purple (280°) and sea-green (350°).

Six different shades of red were known. Rollers would strike a piece of dry wood on the hot metal and

determine the metal’s temperature by observing the reaction (slippery wood indicated 350°, smoking wood

indicated 400° and a spark indicated 425°). Even though some of this know-how was written down and

articulated in a book, it did not entirely capture the knowledge held by the rollers. The refined perception of

colours remained unformulated and deeply tacit. Acquiring this skill required a long apprenticeship with a

master.

5For a debate on codification, see Cowan and Foray (1997), Cowan, David and Foray (2000) and Nightingale’s

(2001) reply. Nightingale argues that tacit and codified knowledge are not substitutes and presents some

doubts on the way in which codification may precede articulation, most notably via a “codebook”.

6 For a debate on the notion of “community of practice”, see Brown and Duguit (1991), Wenger (1998) and,

more recently, Cohendet and Llerena ( 2001).

7 See L. Boy et al (1983), who describe this crisis and the actions taken by the French state to help various

firms in France.

32

8 For a description of the Italian case during this period, see Balconi (1993).

9 As Godelier (ibid) insists, Usinor drew mainly on an internal market and on the basis of length of service

rather than seniority. The elders in fact were the repositories of the company’s memory and were the only

people capable of introducing younger employees into the company and providing them with ‘on the job

training’.

10 The acronym stands for “Institut de Recherche de la SIDérurgie Française” (the French steel industry

research institute).

11 The work of Thierry (1940) is a good example of an attempt to describe empirical knowledge (i.e. a

description of the different situations and incidents that may occur during the process) in order to disseminate

it, by producing general guidelines based on the study of local experience. During the 1980s, the need to

overcome this limitation lay behind Corbion’s construction of a kind of dictionary of the steel industry’s

ancient language in an attempt to make it more explicit (Corbion, 1989). Corbion started off with 730 words

but many additions were gradually made to the original so that by the mid-80s the dictionary contained 1,760

words. By the end of the 90s the figure had reached 12,000. The aim of this glossary was of course to

articulate the industry’s language and to record different dialects and expressions in order to preserve the

social and linguistic identity of its holders.

12 Hatchuel and Weil deal with the same problem in the case of an expert system. Handbooks are usually the

first stage in a longer process and are intended to transfer, translate and articulate current practices through a

support instrument known to all contributors. For a more general discussion, see also Cowan and Foray (1998).

13 According to Mansell and Steimuller ( 2001) the notion of epistemic community has been firstly developed

by Haas (1992). Recently this notion has been used by Steinmuller 2000, Cohendet and Llerena, (2001) and

Cowan, David, Foray (2000). For these latter the notion concerns small groups of “ agents working on a

commonly acknowledged subset of knowledge issues and who at the very least accept a commonly understood

procedural authority as essential to the success of their knowledge activities” (Cowan et al. p. ).

33

14 The acronym that gives its name to the project stands for “Système d’Aide à la Conduite des Hauts

fourneaux En Marche” (aid system for the control of a working blast furnace).

15This was called AEG (the acronym stands for “general analysis of expertise”).

16 The acronym stands for ‘detailed analysis of expertise’.

17For a description of a similar way of knowledge acknowledgment and knowledge validation during the

process of a implementation of a knowledge based system, see de Jongh et al.( 1994).

18 More specific and technical details about KADS (Knowledge Acquisition and Documentation Systems) are

reviewed in a recent article by N. Shabolt (1998).

19 The way experts co-operated in the construction of Sachem and their involvement in the process of

extraction and articulation of their own expertise may appear curious to external observers. Several arguments

can explain their co-operation. First, experts and operators were sensitised to the fact that part of their own

memory had to be preserved and communicated to others. Most of the experts were also encouraged to transfer

their know-how before reaching retirement and were proud to participate in the passing of knowledge to

younger employees. Secondly, operators, technicians and blast furnace experts in the steel industry belong to

the same ‘community of practice’. The latter are very powerful because of their long-standing experience

(their expertise is essentially based on ‘on the job training’ and not necessarily validated by diplomas). Their

own legitimacy is entirely based on their personal knowledge and not on hierarchy. To the experts,

participation in the Sachem project meant that the company acknowledged their empirical knowledge and

validated their extensive work.

20 It is now impossible to hire unskilled operators, as all newcomers must already be endowed with a technical

diploma called BTS (brevet de technician supérieur), which can only be obtained after the completion of

secondary school.

34

21 Know-how usually refers to the total of unconscious and conscious knowledge. Know-that is described as the

conscious level and as knowledge that can be articulated. For a more in depth discussion on this distinction,

originally introduced by Ryle (1949), see Loasby (1999).

22 See Rosenberg (1982) on the use of different forms of fuel for the blast furnace and on ways to reduce fuel

consumption, and Jollivet (1999) on the introduction of Pulverised Coal Injection as a way of reducing coke

input into the French blast furnaces.

23 For an amusing story on the French steel industry, see Fuchs (1980), who shows how a Japanese company

got hold of most of the Dunkerque plant’s plans and strategies during the 1970s.

24 For more details on the active role played by technical and professional associations, see Jollivet (1999).

25 Even if some of IRSID’s projects constitute fundamental research (such as physical and mechanical

properties, different types of corrosion, secondary processing and surface treatments, etc.), most of them are

directed towards Usinor’s private needs and relate to the application of artificial intelligence.

26 This is Usinor’s own estimate.

27 Since Arrow made his contribution, many parts of the production process have in practice been assimilated

into information. More recently, Romer gave new life to the debate by considering information as a quasi-

public good. Here private individuals who respond to market incentives can control many ideas (Romer,

1993). By contrast with private competencies, the principal advantages of quasi-public goods derive primarily

from their unlimited extension beyond the holder: “when the person dies, the skills are lost, but not a non rival

good that this person produces- a scientific law; a principle of mechanical, electrical or chemical engineering;

a mathematical result; software; a patent; a mechanical drawing; or a blueprint-lives on after the person is

gone”(Romer, 1990). Two main attributes are necessary for a good to qualify as quasi-public: it must be ‘non

rival’ and ‘non exclusive’. In order to be non exclusive, quasi-public goods, must be available for use

regardless of the availability of human support. The perfect example of a non rival good, according Romer, is

35

a design, because ‘once a design is created, it can be used as desired in as many productive activities as

desired’.

30 “Intellectual property systems have been strengthened since the 1980s, both in the US and abroad. Moreover,

intellectual property is not just important in the new industries- such as microelectronics and biotechnology- it

remains important in pharmaceuticals and chemicals and its receiving renewed interest in more mature

industries such as petroleum and steel” (Teece, 1998, p. 57).

References

Allen R.C. , 1983, Collective Invention, Journal of Economic Behavior and Organization, 4, 1-24.

Arrow K., 1962 , Economic Welfare and the allocation of resources for inventions, in : Nelson editor, The rate and the direction of inventive activity : economic and social factors ( Princeton University Press, Princeton), pp.

Balconi M., 1993, The notion of industry and knowledge bases : the evidence of steel and mini-mills, Industrial and Corporate Change, ( 2) 3, 471-507.

Balconi M., 1999,  Codification of technological knowledge, firm boundaries, a cognitive barriers to entry , mimeo project DYNACOM, TSER Programme.

Brown S. and Duguit P., 1991, Organizational learning and communities of practice: towards a unified view of working, learning and innovation, Organization Science, 2, 40-57.

Boy L. et al., 1983, Sidérurgie et droit de crise, (Presses Universitaires de France, Paris).

Choamsky N., 1988, Language and problems of Knowledge : The Managua Lectures, ( MIT press).

Corbion J., 1989, Le savoir-fer, glossaire du haut fourneau (Sérémange-Erzange).

Cohendet P. and Steinmuller W.E., 2000, The codification of knowledge : a conceptual and empirical exploration”, Industrial and Corporate change, 9, Number 2, 195-209.

36

Coombs R. and Hull R., 1998, Knowledge management practices' and path-dependency in innovation, Research Policy (27) 3, 239-255

Cowan R., Foray D., 1997, The economics of codification and the diffusion of knowledge , Industrial and Corporate Change, (6,) 3, 595-622.

Cowan R., David P.A. and Foray , 2000, The explicit economics of codification and tacitness Industrial and Corporate change, ( 10) Number 2, 211-253

Divry C. and Lazaric N., 1998, Mémoire organisationnelle et codification des connaissances, Revue Internationale de Systémique, ( 12), 3-11.

Dosi G., Nelson R. and Winter S., 2001, The nature and Dynamics Capabilities of the Firm, (Oxford University Press, Oxford) .

Dreyfus H and S., 1986, Why computers may never think like people, Technology Review, ( 89), 20-42.

Dreyfus H., 1972, Myths and limits of artificial intelligence,

Fleck J. 1994, Learning by trying : the implementation of configurational technology, Research Policy, (23) 6, 637-652.

Fleck J. 1998, Expertise : Knowledge Power and tradeability in : Williams R. Faulkner and Fleck J. (editors), Exploring Expertise, (Macmillan Press Ltd, London), pp 143- 172.

Fuchs G., 1995, Dunkerque, l’odyssée d’un sidérurgiste (Foyer Culturel de l’Houtland, Steenvoorde).

Godelier E., 1997, De la stratégie des sites à la stratégie de groupe: contingence et changement chez Usinor, Gérer et comprendre, Annales des mines, Juin, 79-92.

Hatchuel A., Weil B. , 1992, L’expert et le système ( Economica, Paris).

Hayek F., 1945, The use of knowledge in society, American Economic Review, (35), 519-530.

Jollivet E., 1999, L’innovation technologique comme processus d’apprentissage industriel : analyse de la formation des connaissances dans le cas des hauts fourneaux à injection en France et au Japon, Aix en Provence, LEST, Phd dissertation.

de Jongh. P. J. et al., 1994, Future: A knowledge based System for threat Assessment, Interfaces, 24, 261-272.

Knudsen T., 2000, Why tacit knowledge protects the firm’s evolutionary potential and why codification does not, Working Paper presented for the workshop « Cognition and Evolution in the theory of the firm », Mack Planck Institute Jena.

Lave J. and Wenger E., 1990, Situated learning : legitimate peripheral participation, (Cambridge University Press, New York ).

Lazaric N., 2000, The role of routines, Rules and Habits in collective learning : some epistemological and ontological considerations, European Journal of Economic and Social Sciences, 14(2), 157-172.

37

Lazaric N. and Marengo, 2000, Towards a characterization of assets and knowledge created in technological agreements: some evidence from the automobile robotics sector, Industrial and Corporate Change, 9, 53-86.

Leibenstein H., 1987, Inside the firm : the efficiencies of hierarchy (Cambridge, Harvard University Press).

Loasby B., 1998, The organisation of capabilities, Journal of Economic Behavior &Organization, (35), 139-160.

Mangolte P.-A., 1997, La dynamique des connaissances tacites et articulées : une approche socio-cognitive , Economie Appliquée, L ( 2), 105-134.

Mansell R. and Steinmuller E., 2000, Mobilizing the information society, (Oxford University press, Oxford).

Nelson R. and Winter S., 1982, An evolutionary theory of economic change (Harvard University Press).

Nelson R., 1993, Technical change as Cultural Evolution, in : Thomson R. (editor), Learning and technological Change, (St Martin ‘s Press, New York)., pp. 9-23.

Nightingale P., 2001,  If Nelson and Winter are only half right about tacit Knowledge , Which Half ? a reply to David Foray and Cowan, Working paper presented to the Nelson and Winter conference, Aalborg, June

Nezosi G., 1999, La fin de l’homme de fer : syndicalisme et crise de la sidérurgie, ( L’harmattan, Paris).

Ngo-Mai S. and Rochhia S., 1999, Auto organisation et connaissance dans la firme, Revue d’Economie Industrielle, 88, 237-254.

Pavitt K., 1984, Sectoral patterns of technical Change : towards a taxinomy and a theory, Research Policy, 13, 343-373.

Romer P., 1990, Endogenous Technological Change Journal of political Economy, 98, 71-102.

Romer P., 1993, Two strategies for economic development : using ideas and producing ideas in : Proceedings of The World Bank Annual Conference on Development Economics, World Bank ed , pp 63-91.

Rosenberg N., 1982, Inside the black Box, Technology and Economics, Oxford University Press, Oxford.

Rosenberg N., 1986, The impact of technological innovation : an historical view, in The positive Sum Strategy, Landau R. and Rosenberg eds, pp. 17-32.

Piaget J. , 1970, L’epistémologie génétique, (Presses Universitaires de France, Paris).

Polanyi M., 1958, Personal Knowledge. Towards a post-critical philosophy, Routledge and Kegan Paul, London.

Shabolt N., 1998, Building models of expertise, in Williams R. Faulkner and Fleck J. (eds), Exploring Expertise, (Macmillan Press Ltd, London) , pp. 101-120.

38

Saviotti P., 1998, On the dynamics of appropriability, of tacit and of codified knowledge, Research policy, 26, 843-856.

Simon H., 1999, The many shapes of knowledge, Revue d’Economie Industrielle, (88), 23-39.

Steiler J. M. and Schneider M., 1994, L’élaboration de la fonte, in : Steiler J. M. and Schneider M. (editors), Le livre de l’acier, (Technique & Documentation – Lavoisier, Paris), pp. 1187-1229.

Steinmueller W. E. , 2000, Will information and communication technologies improve the “ codification ” of knowledge ?, Industrial and Corporate change, (10)2, 361-376.

Teece D.J. , 1998, Capturing Value From Knowledge Assets, California Management Review,.40 , Spring.

Thierry P., 1940, La pratique du haut-fourneau, Librairie Polytechnique, C. Béranger, Paris, Liège.

Verry M., 1955, Les laminoirs ardennais, déclin d’une aristocratie professionnelle (PUF, Paris).

Wengor E., 1998, Communities of practice : learning , Meaning and Identity, Cambridge University press, Cambridge.

Winter S., 1987, Knowledge and competence as strategic assets , in D. J. Teece (editor.), The competitive challenge : strategies for industrial innovation and renewal (Ballinger, Cambridge), pp. 159-183.

Zollo M. and Winter S.G., 2001, From organizational routines to dynamic capabilities, Forthcoming in Organization Science.

Table 1 : The creation of the Usinor group

Years USINOR SACILOR

Forges and Aciéries du

Nord et de L'Est

Forges and Aciéries de Denain-Anzin

1948 Usinor Sollac is created with a concentration in Lorraine

39

1951 Creation of Sidelor

1953-1966

Merger of Usinor and Lorraine-Escaut Creation of Lorraine- Escaut

1968 Creation of Wendel Sidelor

1974 Creation of SACILOR (merger of all existing plants)

1970-1980 French Steel Concentration: Usinor and Sacilor(many plants are shut down, including Valenciennes Longwy and Denain)

New manager, Mr Etchegaray, brought in from a different industry

1981 French state intervention in the two firms(90 % French state participation)

New manager: Mr RH. Levy

1986 Merger and birth of the public group Usinor-Sacilor

1995 Usinor-Sacilor becomes a private group

1997 Group renamed Usinor

40

Table 2 : The Sachem methodology

41

Table 3 : A summary of the different stages of the Sachem project

Years Main objectives Technical name of the sub-project and process of knowledge’s transformation

1989 Shared vocabulary among blast furnace experts Xperdoc(First stage of knowledge articulation)

1990 .Exploration of various opportunities provided by artificial intelligence

.Choice of the Sachem methodology

1991-1992 . Delimitation of the blast furnace expertise with representative case studies

AEG( Second stage of knowledge articulation )

1992-1995 . Selection of core know-how

. Encoding process and validation by blast furnace experts

. Official acknowledgments of the ‘epistemic community’ in Usinor

AED( Codification)

1996 . Sachem implementation in Usinor Sachem

Table 4: The different states of knowledge in the steel industry

Types of knowledge

Charac-teristics ofknowledge

Local Generic Universal 

Localisation . Knowledge is anchored to a blast furnace and some experts

. Knowledge is partly extracted from local context, diffused to each plant, and usable by various experts.

. Knowledge is considered as information, a public good

Diffusion . Circulation and communication of this knowledge is rather limited because decentralized ( transfer of knowledge is mainly under the control of the apprenticeship’s system)

. Knowledge is spread inside the ‘epistemic community’ of the blast furnace and centralized inside Usinor

. Circulation and appropriation of knowledge may be limited by property rights and institutional protection or by agent calculation.

Properties

42

. Tacit . Exclusive but partly shared inside a small ‘community of practice’

. Articulated and codified

. Non exclusive and shared inside the ‘epistemic community’ . Partially rival (Sachem is used in Usinor and controlled in order to avoid its expansion to competitors).

. Non rival . Non exclusive

43