The Bread-making Machine: Tacit Knowledge and Two Types of Action Rodrigo Ribeiro and Harry Collins ABSTRACT We analyze Nonaka and Takeuchi’s (1995) claim that a master baker’s tacit knowledge was made explicit and incorporated into a home bread-making machine and its manual – the ‘knowledge capture’ thesis. In order to test the claim, bread was made without and with a breadmaker and we carried out an analysis of the bread-making actions before and after mechanization. Based on the theory of action morphicity (Collins and Kusch 1998) it is shown that the machine only mimics the mechanical counterpart of just a few of certain special kinds of human bread-making actions. The remaining success of the machine and its manual is due to what other human actors bring to the mechanical bread-making scene; this way the breadmaker can be an adequate social prosthesis. Action mimicking, action substitution, and the contributions of these other human actors, who are not needed in the case of the master baker, explain why the machine and its manual do work. It is not a matter of the explication or incorporation of tacit knowledge, but of fitting a social prosthesis into a rearranged world. Keywords: Tacit knowledge, explicit knowledge, knowledge conversion, knowledge management, technology transfer. Introduction ‘With the breadmaker, even the most inexperienced baker can achieve the satisfying experience of baking a loaf of bread. All the mystery and hard work is gone. Inside this talented machine with an electronic brain, the dough is mixed, kneaded, proofed and baked without you being present.’ (Morphy Richards 2004: 13) The above extract, taken from the manual of a bread-making machine – the Morphy Richards Compact Breadmaker built in Britain – reminds us Nonaka and Takeuchi’s (1995) description of the first Japanese home bread-making machine. Both contain claims about the machine’s intelligence and skills vis-à-vis the lack of experience required from users: ‘[the 1/26 machine] transforms raw ingredients into freshly baked bread, doing everything from kneading and fermenting the dough to actually baking bread … the machine is remarkable in that it embodies the skills of a master baker in a device that can be operated easily by people with no knowledge of bread making’ (1995: 95). The bread-making machine, taken as an example of ‘knowledge capture’, has gained an appeal that goes beyond the kitchen. Its development is the main empirical case used by Nonaka and Takeuchi (1995) to support their ‘theory of organizational knowledge creation’. The latter is based on four possible interactions between tacit and explicit types of knowledge. Among the alternatives of ‘knowledge conversion’, Nonaka and Takeuchi stress the ‘mobilization and conversion of tacit knowledge’ into explicit knowledge as ‘the key to knowledge creation’ (1995: 56; 66). At this point, the Japanese breadmaker becomes central. For Nonaka and Takeuchi allege that a master baker’s tacit knowledge was ‘converted’ into explicit knowledge and ‘embodied’ into the machine. Under this theory, tacit knowledge is reduced to ‘hidden’ (1995: 71) or ‘not-yet-articulated knowledge’ waiting to be uncovered and explicated (Tsoukas 2005: 154). Here we will not discuss the overall theory of Nonaka and Takeuchi but the main empirical case on which it was based. If their analysis of the case is wrong then, presumably, their theory is wrong, but we leave these implications to theorists of organization and management studies. Nonaka and Takeuchi’s description of the breadmaker has been endlessly discussed and, more recently, criticized (Essers and Schreinemakers 1997; Tsoukas 2005; Gourlay 2007; D’Eredita and Barreto 2007). One striking characteristic of the entire literature is that, starting with Nonaka and Takeuchi themselves, no author has tried to make any bread before putting words about bread-making into papers about bread-making. Here we make bread as we analyze. Consequently it is easy to see why the breadmaker does work as well as it does even as we uncover the problems with the earlier descriptions of the way it works. 2/26 We demonstrate that there is no ‘conversion’ from tacit to explicit knowledge in breadmaking machines. Tacit knowledge is still necessary for the bread to be produced, but it is supplied by members of the wider human group in which the machinery is embedded. Any ‘intelligent’ or automatic machine is a ‘social prosthesis’ (Collins 1990). The way a physical prosthesis such as an artificial heart works can only be understood by watching the way it interacts with the rest of the body. Likewise, the way a social prosthesis works cannot be understood by examining it in isolation but only by looking at the way it fits into the web of activities in which every other human activity is embedded. In making bread, the role of the whole network on the one hand, and the specific elements that are taken over by the breadmaking machine on the other, are more readily discovered. As we will see, as far as the direct replacement of human activity is concerned, what the machine does is to mimic the mechanical counterpart of just a few of a certain special kinds of human bread-making actions. To use the analysis and language of Collins and Kusch (1998), the whole ‘automatic’ bread-making ‘action tree’ involves substituting certain of the ‘polimorphic actions’ of normal human bread-making and mimicking certain of the ‘mimeomorphic actions’ that are already found in human bread-making. At the end, the master baker’s tacit knowledge has been neither explicated nor incorporated into the machine. Part of it was substituted by the tacit knowledge of the other actors brought to the automated bread-making scene, such as the users at home, the workers in the factory and repair specialists, while the other part has disappeared entirely at the cost of a standardized set of products and procedures. In what follows, Collins and Kusch’s 1998 theory of action morphicity is first described. This is followed with the description of bread making, first without, and then with a machine. The breadmaker itself, and the instruction manual that accompanies it, is then discussed. 3/26 The Theory of Action Morphicity Collins and Kusch (1998) distinguish between the intentional and behavioural parts of human actions. Action is defined by them as ‘the behaviour plus the intention’ (1998: 32), behaviour being ‘the physical movements humans use to execute the actions they intend’ (1998: 8). Collins and Kusch then divide human actions into two types: ‘polimorphic’ and ‘mimeomorphic’. The same polimorphic action is generally executed with many different behaviours depending on the social circumstances. For example, the action of ‘greeting,’ if always carried out with exactly the same movements or tone of voice, would cease to be a greeting and become something more like a ‘salute’ or even an insult. But there are no available instructions for how to vary the behaviour associated with the action in order to carry it out successfully. Executing a polimorphic action successfully, and most of what we do consists of polimorphic actions, depends on the tacit knowledge needed to live in society. So long as machines do not understand social life it will be impossible to mechanize polimorphic actions. Mimeomorphic actions, on the other hand, are generally carried out with the same behaviour on every occasion or, to be more exact, with behaviours which differ only randomly or in such a way as to be a matter of indifference. For example, the different behaviours often used in punching in a particular number on a telephone keyboard are a matter of indifference; one can say that the number is always punched in by executing ‘the same’ behaviour. Here the notion of ‘the same’ implies an area of tolerance (1998: 47) around the degree of variation (and implies tolerance to exact similarity if it could be managed – in contrast to the example of greeting.) Machines which mimic mimeomorphic actions can be constructed. The proper word is ‘mimic’, rather than ‘reproduce’, since to reproduce an action requires that the intention is 4/26 also present and machines do not have intentions. Where mimeomorphic actions are concerned, human areas of tolerance are then translated by designers into the limits within which machines are to behave. Machines do not need to understand the surrounding culture to mimic mimeomorphic actions. Another way to put this is that there is no discernible difference to an outside observer between mimicry of the behaviour associated with a mimeomorphic action on the one hand, and repetition of the action, complete with intention on the other, whereas this is not the case with polimorphic actions. Polimorphic and mimeomorphic actions can be combined in different ways to form ‘action trees’ (1998: 71-77). Action trees have high level actions that are executed by a series of lower level and more specified actions. Bread-making can be described as an action tree, with the actions of kneading and baking within it. Bread-making is itself part of the action tree of ‘cooking’. Nearly all skills consist of polimorphic actions trees with bits and pieces of mimeomorphic actions embedded within them. Skills Embodied: The Home Bread-Making Machine by Nonaka and Takeuchi (1995) Nonaka and Takeuchi’s (1995: 95) account is about the development of the first ‘fully automated bread-making machine for home use’ developed by Matsushita. It was introduced in the Japanese market in 1987 and was a sales success. According to the standard account, the design team faced three problems in developing the machine. The first was ‘how to mechanize the dough-kneading process, which is essentially tacit knowledge possessed by master bakers’ (1995: 63). The other two concerned temperature and ingredient variability: ‘The ideal [ambient temperature]… was 27 to 28 degrees centigrade, yet the variation in … Japan ranged between 5 and 35 degrees centigrade … different brands and kinds of flour and yeast further complicated the control system’ (1995: 102-103). 5/26 It is said that in order to solve the dough-kneading problem, Ikuko Tanaka was sent to learn how to make bread with a famous master baker. After a period, ‘[Tanaka] noticed that the baker was not only stretching but also “twisting” the dough, which turned to be the secret of making tasty bread’ (1995: 64). At this point, Nonaka and Takeuchi’s argument is that tacit knowledge can be explicated by ‘taking the shapes of metaphors, analogies, concepts, hypotheses or models’ (1995: 64) and then incorporating them into machines by their designers. Kneading dough is presented as the key example: ‘[Tanaka] was able to transfer her knowledge to the engineers by using the phrase “twisting stretch” to provide a rough image of kneading … Her request for a “twisting stretch” movement was interpreted by the engineers ... After a year of trial and error … The team came up with product specifications that successfully reproduced the head baker’s stretching technique … The team then materialized this concept, putting it together into a manual, and embodied it in the product’ (1995: 104-105; our emphasis). The temperature problem was solved by ‘adding the yeast at a later stage in the process … It was the way people had made bread in the past … This method … was the result of the socialization and externalization of the team members’ tacit knowledge’ (1995: 107-8). In this quotation, Nonaka and Takeuchi seem to use the term tacit knowledge to refer to knowledge which is easily verbalized, but nobody has thought to mention. Finally, the solution for different brands can be identified in a ‘marketing-feature’ of the machine: ‘For even further convenience, a pre-measured bread-mix package can be used to save the trouble of measuring out the required ingredients’ (1995: 95). We now examine this account in the light of the experience of bread making both without and with machines. It is certainly the case that, for novices, bread-making is more reliable and more efficient with a machine. But since we are going to argue that the skills of the master baker are not embedded in bread-making machines we must ask how it is that this success is achieved. 6/26 Experiences of Bread-Making The making of bread was entirely the responsibility of the first author and, for easy of exposition, the next passages of the paper are written in the first person with ‘I’ always referring to Ribeiro. I made bread using a Morphy Richards Compact Breadmaker borrowed from a friend. It is like the Japanese breadmaker, at least as it is described by Nonaka and Takeuchi, in that it mixes, kneads and bakes the bread. It differs from the Japanese machine in that the yeast is mixed at the beginning of the operation. Crucial in the case of both breadmakers are their printed manuals and their role must also be taken into account. I also made bread by hand, making use of a do-it-yourself booklet. I had never made bread before and my experience of cooking is almost none. Making Bread by Hand To make bread by hand I started by reading a self-help booklet ‘Bake Your Own Bread’ (Deutch, 1976), which presents bread-making as an ‘art’. For a novice, such as myself, problems presented themselves from the beginning. Thus I learned that ‘controversy still rages over fresh yeast versus dried yeast. … Because the plants in fresh yeast are active and alive it is highly perishable and can be kept only for four to five days in an air-tight container in the refrigerator. Fresh yeast should feel cool and like putty to touch. It should be grey in colour and practically odourless … Do not use yeast that is dry or sour-smelling or has dark streaks’ (1976: 1-2). Filled with fear about the recalcitrant nature of yeast I was uncertain of my judgement about what cool putty should feel like. Timings were given in ranges and the use of adjectives abounded: ‘After the end of 15 to 20 minutes the yeast will be puffed up and frothy’, ‘If the dough is too soft and wet, more flour may be worked in’, or ‘Continue kneading for about 10 minutes until the dough feels smooth and elastic’ (1976: 2, my emphasis). Once more, these terms seemed to call for more experience, learned in shared 7/26 kitchen spaces, than I had. I did not know the acceptable zone of tolerance around any of these descriptions, such as how wet something should be that was not to be counted as ‘too wet’ or too dry. On the other hand I felt I could adopt a tip from the booklet that could not be used in a breadmaker: ‘A shiny, glazed crust, characteristic of French and Vienna bread or rolls, can be obtained by placing a flat pan of boiling water in the bottom of the oven just before the bread is put in and leaving the tin in the oven throughout the baking’ (1976: 2). The breadmaking machine does not have an oven in which a pan of water can be placed. To make bread by hand was physically hard. I had not only to mix the ingredients, but to knead, wait for the rising, knead again, wait for a second rising, pre-heat the oven, shape the dough, and bake. I was either working on the bread-making or monitoring it for nearly every moment of the 4 hours and 23 minutes it took for the bread to be ready. When there were still 35 to 45 minutes to go, I smelled burning. The top of the loaf had become darker. I took the loaf from the oven and, following the instructions, rapped its bottom with my knuckles. It did not sound like a ‘drum’ (this part of the instruction I could understand) and it had to go into the oven again, upside down. After 10 minutes, the bottom ‘sounded’ right and the job was finished. The loaf was still dark on the top, but the crust was a little bit softer, probably due to the tip on the boiling water. Nevertheless, to my disappointment, the bread was undercooked. Ten minutes more in the oven and it was ready. Despite the problems it was edible. Making Bread with a Bread-making Machine In the case of the machine, the first thing I did was to read the manual from cover to cover. I found that, while hand made bread can come in any size and shape, the machine offers the possibility of baking only two sizes in only the one, cuboid, shape. Part of the skill of a master baker will be in choosing sizes and shapes even inventing new ones. The 8/26 machine also offers three options of ‘crust colour’; it can be set up to produce a ‘light’, ‘medium’ or ‘dark’ crust. The machine, one might say, presses the user into shape, size, and crust colour (Woolgar 1991). For each of the two sizes of loaf 12 recipes are given. Each recipe corresponds to a ‘program’ that has to be selected by the user. The programs define the timing of the different machine operations: Knead 1, Rise 1, Knead 2, Rise 2, Rise 3, and Bake (2004: 21). Choosing a recipe among the ones listed in the manual is a polimorphic action. The user must know the recipes and which of them is the most appropriate for the social circumstances in which the bread is to be eaten. The actual selection of the program once it has been chosen is, however, a mimeomorphic action. Selecting the chosen program is like dialling a telephone number – a matter of pressing buttons, the exact way it is done being irrelevant within a wide envelope of tolerance. The instruction manual places a lot of stress on ‘measuring’ and ‘ingredients’. A section on ‘Measuring ingredients’ includes explanations on how to measure accurately and the consequences of doing it wrong: ‘scooping or tapping a measuring cup will pack the ingredients and you will end up with more than required. This extra amount could affect the balance of the recipe’ (2004: 12). ‘Measuring the ingredients accurately’ is the solution for 15 out of 30 possible problems listed in the ‘Troubleshooting’ section (2004: 24-25). In addition its importance is mentioned, on average, every three pages. In contrast to the 12 pre-set programs, this part of the manual initially seems to be very much like the do-ityourself booklet in tone: the user would have to bring special skills to the task and make skilled judgements about the size of the areas of tolerance around what counts as correct measurement in that local area of society (i.e. bread-making). Nevertheless, a closer look at the manual shows that the nature of measuring was transformed. Standardized measuring tablespoons and teaspoons as well as a cup are provided with the breadmaker. A drawing showing how to ‘level’ – not to ‘heap’ – is presented in the manual and the use of ‘normal kitchen spoons’ is not recommended, as ‘these are inaccurate’ (2004: 12; 20). The 9/26 polimorphic action of measuring just by looking at quantities and adding a pinch of this or a handful of that, something that a master baker can surely do, has been substituted by mimeomorphic actions of filling and levelling measuring cups and spoons. This is something that could be easily automated as the automation of raw material measurement in more complex industrial plants demonstrates. The section devoted to ‘Know your ingredients’ (2004: 11) explains the different kinds of ingredients, how they affect the bread, and what users should use or not. Examples are: ‘Selfraising Flour contains unnecessary leavening ingredients that will interfere with bread … It is not recommended for use’ or ‘It is recommended that fast action yeast be used’ (2004: 1112). Finally, in the ‘Recipes’ section users are ‘informed’ that ‘These recipes have been developed using Allison flours and Easybake Allison yeast’ (2004: 14). Here it seems that the range of ingredients is being narrowed down to a restricted number of pre-set options. I took a recipe for French bread from the machine manual for the experiment. I went to the market, searched for the ingredients and brands listed in the manual, and checked the valid date. It then took me 28 minutes to prepare the ingredients, put them into the machine, and enter the options of program and loaf size. I pressed the ‘start button’ and 3.5 hours later the bread was ready. The owner of the bread-making machine said that the bread had the same taste as when she makes it and this leads me to assume that I had used it successfully. Analysis of the Fieldwork For a novice, bread-making by machine was more reliable, more efficient, less stressful and physically easier: the machine ‘worked’. It did not, however, embody the skills of a master baker or even the skills I learned (or failed to learn), from the do-it-yourself booklet. (Although there were differences in the outcomes – the hand-made bread being considered better – I treated the final products as ‘the same’ for the purpose of analysis). ‘Breadmaking’ with the machine was simplified and standardized. The pre-set possibilities of loaf 10/26 size, recipes, brands and types of ingredients offered countable and unambiguous – digitized – options for me to choose. As foreseen in the manual, my skills of reading and measuring were enough: ‘This machine requires only that you carefully follow the recipe instructions. In basic cooking, normally “a pinch of this and a dash of that” is fine, but not for breadmakers. Using an automatic breadmaker requires [that] you accurately measure each ingredient for best results’ (Morphy Richards 2004: 13). We can summarise what was learned from the fieldwork in tables. Table 1 shows the way the actions of the master baker baking a specific and fixed size, shape, and type of bread, were incorporated or transformed by the breadmaker and the user of the machine. BREAD-MAKING MACHINE MASTER BAKER MIMICRY BY MACHINE SUBSTITUTION BY USER Picking up the ingredients Mimeomorphic Mimeomorphic Measuring Polimorphic Mimeomorphic Setting program -- Mimeomorphic Setting dough size -- Mimeomorphic Mixing and kneading Mimeomorphic Mimeomorphic Shaping Mimeomorphic Mimeomorphic Baking Mimeomorphic Mimeomorphic TABLE 1 - Cooking a fixed size, shape and type of bread in two ways Once the size, shape, and type or recipe of bread has been decided, the machine and user do not have a difficult task of automation. Only one action – that of measuring – has to be transformed from polimorphic to mimeomorphic and this is not done by the machine but by the supply of measuring cups and spoons and a very careful and oft-repeated set of instructions and warnings. (Note that, so long as we do not concern ourselves with the tacit knowledge the reader needs to understand the manual, we could say that the manual is written as though to address a machine. In principle, novice users do not have to understand 11/26 the intentions behind the mimeomorphic actions they need only to mimic. For instance in picking up the ingredients that follow from the ‘French recipe’ Ribeiro was mimicking an action without understanding the intention behind it. This is why, in principle, all the mimeomorphic actions performed by users in the upper part of Table 1 could be mechanized.) Everything else that is done by the master baker is a mimeomorphic action and these can be mimicked by the machine without running into any of the normal problems of trying to mechanize socially embedded human action. Not all mimeomorphic actions are explicable to humans. Kneading, though it is a mechanisable mimeomorphic behaviour is something that is only mastered as a piece of tacit knowledge by humans (like balancing on a bicycle). These heavily tacit-knowledge laden mimeomorphic actions are learned by humans in social groups just as polimorphic actions are learned and that is why the literature on tacit knowledge often does not make the distinction between what can be automated and what cannot be automated in the right place. That humans can learn a certain kind of behaviour only the way they learn polimorphic actions does not mean that the behaviour is polimorphic – after all, it is easy to automate balancing a bike. What cannot be so easily automated is bike-riding in traffic, which requires social understanding (for these distinctions see Collins and Kusch, 1998). The way the breadmaker used by Ribeiro mixes and kneads differs from the way it is done by the Japanese bread-making machine and probably differs from the way humans do it, but the mimeomorphicity is demonstrated by the tolerance to the exact behavioural instantiation of the kneading act. (One way to make a bike balance is by using gyroscopes!) The fact that learning mimeomorphic actions calls for humans to learn how to judge if the behaviour is within the areas of tolerance is also a confounding factor. For instance, Gourlay (2007) does not distinguish between somatic-limit tacit knowledge and collective tacit knowledge (Collins 2007, in preparation) when he argues that the socialization between the master baker and Tanaka can be seen just as individual learning-by-doing, and not as a 12/26 case of tacit knowledge transfer. Tanaka had to go through some individual trial-and-error in order to knead (which is a mimeomorphic action based on somatic-limit tacit knowledge). But in order to knead properly, she needed someone to tell or show her what an error was or not, she needed the master baker to teach her how to make ‘correct judgments’ (Wittgenstein, 1976 [1953]: 227e) on kneading. This means the transfer of collective tacit knowledge. Going back to Table 1, the user is also made responsible for a couple of mimeomorphic actions that are unique to the use of the machine – namely pressing the program-setting buttons. The table makes it easy to see what is done by the user and what is done by the machine, and how it is that the machine can and the user can do their jobs without too many problems. This, however, is only a small part of the story of mechanising bread-making. Pulling back from the narrow focus on a fixed size, shape, and style or recipe of loaf, reveals more of the action tree of bread-making and enables us to see that in the upper regions there are considerable differences between the master-baker’s action tree and that of the breadmaker. To make bread by hand and by machine consist of different ‘actions trees’. Although the higher intention is the same – to make bread – this is accomplished by distinct sets of actions and action mimicking under it. Gourlay too (2007) draws on the distinction between polimorphic and mimeomorphic actions in order to highlight these two ways of making bread: ‘In competitive economies, the set of tasks of any practice are often reorganized by transforming polimorphic tasks into more mimeomorphic ones, otherwise known as deskilling.’ ‘Deskilling’ is an unfortunate choice of term because many mimeomorphic actions carried out by humans require a great deal of skill. For example, the golf swing is mimeomorphic and can (at least in principle), be better accomplished by machine (the psychology of match play is polimorphic). Synchronised swimming, marching, ski-jumping, 13/26 high-board diving and such like sports are mimeomorphic – humans measure their performance against machine-like perfection. Thus a better term for what Gourlay wants to describe would be ‘de-socialising’. Most actions are a complex weave of mimeomorphic and polimorphic which can be untangled only with the most enormous care. Printing books, for example, seems purely mimeomorphic but consider the way that meaning is affected by the chosen font and the difference between a justified and a ragged-right hand margin. Laying out a page is a context sensitive and therefore polimorphic action. That aside, Gourlay is right, at least, partially right. What we will add to Gourlay’s analysis is to show that mechanization of bread-making also involves the substitution of polimorphic actions by other polimorphic actions executed by actors both near and far from the kitchen. We also add a great deal more about the way these processes work.) Table 2 displays actions prior to the choice of a specific size, shape, and style of loaf, once more as executed by a master baker in one column and the machine and human actors in other columns. A difference in the rightmost heading of this Table 2 and Table 1 should be noted. In Table 2 ‘USERS’ become ‘HUMANS’ because here, some of the humans responsible for the actions are not the users of the machine. 14/26 BREAD-MAKING MACHINE MASTER BAKER MIMICRY BY MACHINE SUBSTITION BY HUMANS Setting up the production scene Polimorphic Polimorphic Choosing recipe, size and crust colour Polimorphic Polimorphic Dealing with the variability of ingredients and brands Polimorphic Polimorphic Choosing level of tolerance of final product Polimorphic Polimorphic BELOW THIS LINE A SINGLE STYLE OF BREAD HAS BEEN CHOSEN. THE FIRST TABLE FITS BELOW HERE IN THE ACTION TREE TABLE 2 - Actions prior to the choice of a fixed size, shape, and style of bread In row one we see the action of setting up the production scene. In one hand, these are not so different for master baker and machine user. Both must gather ingredients and utensils – albeit the choices are different – and be responsible for hygiene. Both must make sure that the ingredients are not contaminated or spoiled. These actions call for social judgments, such as what counts as a satisfactory level of cleanliness and hence both are polimorphic actions. On the other hand, users must be members of a culture where domestic appliances and their manuals are commonplace. This enables them to understand the machine ‘microworld’ and what should be taken into account when preparing the scene for it to mimic properly and safely – all of which call for social judgments as well. The concept of ‘microworld’ as first used by Minsky and Papert: ‘Each model – or “micro-world” … talks about a fairyland in which things are so simplified that almost every statement about them would be literally false if asserted about the real world’ (1970: 39, quoted by Dreyfus, 1979 [1972]: 9). Dreyfus (1979 [1972]: 13) responds that: ‘a set of interrelated facts may constitute a universe, a domain, a group, etc., but it does not constitute a world, for a world is an organized body of objects purposes, skills, and practices in terms 15/26 of which human activity have meaning or make sense’. Bread-making by hand is a ‘world’; that part of bread-making done by machine is a ‘microworld.’ Choice of recipe, size, and crust colour are also polimorphic actions for both master baker and machine user. These are certainly polimorphic actions. One ‘rule-of-thumb’ for determining the ‘morphicity’ of an action is what we might call the ‘pigeon test.’ Can one imagine a pigeon (or monkey in the case of kneading), being trained by behavioural conditioning techniques to carry out the action. Choosing the right kind of bread for the circumstances involves thinking about the season and whether or not the bread is to celebrate some social event or just be eaten in the normal way of things, what kind of food it might go with, the nationality of the consumers, the time available to cook it and the ingredients to hand. Even if a device were constructed such that the choice were made by pecking at one of a number of buttons, a pigeon could not be trained to make the choice. On the other hand, once the choice has been made, a pigeon could be trained to peck at the appropriate program keys so as to execute the choice. The choice, it should be noted, is made, on the one hand, by the master baker working alone, on the other hand, by the user of the machine along with the designers of the machine. The designers give the machine a much more restricted range of options than that available to the master baker. The owner of the particular breadmaker used in this test explained that the option of ‘light crust’ actually produces too dark bread for her taste, but that she could do nothing about it. Dealing with the variability of ingredients and brands is a polimorphic action as it involves judgments regarding differences in taste, flavor or texture and their impact the final product. In making bread by machine, this action is assumed by those who write the manual with its standard recipes, ingredients and brands. They try to sort this problem out beforehand as it is clear that novice users, such as myself, cannot make such evaluations. Another possibility is the use of the ‘pre-measured packages’ mentioned by Nonaka and 16/26 Takeuchi. In this case, the designers and workers who produce them are the ones who substitute the master baker’s skills in regard to this task. The fourth row of the table is choice of level of tolerance. We may imagine that a master baker making bread for the Queen of England would be far more careful about quality than one making bread for sale off a cart. In the very flow of the cooking the master baker will be executing a choice about how tolerant to be and this is a polimorphic choice because it turns on social circumstances. This choice is made for the user by the designer and makers of the machine. Again it is a polimorphic action, this time the designers have to think about what degree of variation (i.e. quality of bread) users of the breadmaker will accept. This is another difference between the two bread-making action trees. While the master baker can modify the levels of tolerance all the way through making bread, machine users cannot. The areas of tolerance within which the breadmaker works are made fixed at the time of its design – and so is the quality of its products. We could now construct a ‘Table 3’ if we wished. Table 3 would fit below Table 1 rather than above it. It would show, for instance, the actions required after the loaf had been completed. After the cycle was completed the user has to decide if the bread is within the range of acceptability. It could be that the user would notice an unacceptable smell, colour, or texture, or taste – all matters of social convention. This in turn might imply an openended investigation of the breadmaker and or the ingredients. Was the level of hygiene right, was the yeast pure, was the fuse blown? If the problem could not be found and fixed by the user then repair specialists would have to be brought in to do their polimorphic action laden tasks. There are still more polimorphic elements to the bread-making action tree carried out by those more remote from the scene. These include inventing new cooking techniques – such as is represented by the pan of water in the oven – inventing better breadmakers. It would be tedious to analyze all this in detail and actually draw up Table 3, but the wider focus is worth mentioning because, once more, it stresses the fact that the bread-making 17/26 machine itself is embedded in a society that adjusts to make it fit, and that the notion that the skills of a master baker are simply transferred to this small mechanism is misplaced. The Continuum of Human Contribution to Mechanical Bread-Making The fieldwork described above represents just one user who embarked on bread-making from a low level of expertise in the domain of cooking. A breadmaker could be used more flexibly. A more experienced user could have used it just for the hard physical work of kneading, the prepared dough then being shaped and baked by hand. Perhaps experiments could have been done with novel ingredients, not listed in the instruction manual but anticipated therein: ‘Flours, while visibly similar, can be very different by virtue of how they were grown, milled, stored, etc. You may find that you will have to experiment with different brands of flour to help you make the perfect loaf’ (2004: 11); ‘When creating your own yeast bread recipes or baking an old favourite, use the recipes in this cookbook as a guide for converting portions from your recipe to your breadmaker’ (2004: 13). It might also be possible to use the machine still more creatively, perhaps switching the power off to make intermediate modifications to the bread before restarting the cycle. Actually, the manual demanded that some flexibility be drawn on: ‘Humidity can cause problems, therefore humidity and high altitudes require adjustments. For high humidity, add an extra table spoon of flour if consistency is not right. For high altitudes, decrease yeast amount by approximately ¼ teaspoon, and decrease sugar and/or water or milk slightly’ (Morphy Richards, 2004: 13). All of this, of course, requires more of the master-baker’s skills to be added to the social network surrounding the social prosthesis, ‘repairing’ its deficiencies to a greater and greater extent and making Nonaka and Takeuchi’s account still more misleading. 18/26 The Instruction Manual The whole of the above analysis could be repeated for the instruction manual, which is just as mysterious a piece of ‘machinery’ as the breadmaker. One can see that an instruction manual itself depends on a huge input of skills from the user – the skills of language interpretation as well as some previous practice – and from those who write and modify manuals in response to users’ failure to understand. Nonaka and Takeuchi’s (1995) argument that analogies, metaphors, concepts, and the like can explicate and transfer tacit knowledge fails for the same reason. They say ‘[Tanaka] translated the kneading skill into explicit knowledge. The knowledge was externalized by creating the concept of “twisting stretch”’ (1995: 105) but, contrary to what is claimed, the concept of ‘twisting stretch’ was not enough for Tanaka to transfer her knowledge. The Japanese engineers failed when they tried to understand what Tanaka was ‘talking about’. As put by Nonaka and Takeuchi themselves, ‘neither the head baker nor Tanaka was able to articulate knowledge in any systematic fashion. Because their tacit knowledge never became explicit [although here Nonaka and Takeuchi hold a position closer to the one in this paper, their whole theory is based in the argument that tacit knowledge associated with kneading dough was made explicit] … engineers were also brought to the hotel … For those who had never touched dough before, understanding the kneading skill was so difficult that engineers had to share experiences by spending hours at the baker to experience the touch of the dough’ (1995: 104). ‘Twisting stretch’ only started to convey its practical meaning when it was learned through social contact with the master baker. Ribeiro also faced this problem when reading about ‘soft’, ‘wet’, ‘smooth’ or ‘elastic’ dough in the booklet. Although the adjectives were written, they were not explicit. To look them up in the dictionary would do no good. Again, the only way for Ribeiro to understand what they mean in practice is if he starts socializing with bakers. 19/26 The point is also exemplified by the difficulty of the owner of the machine used in the fieldwork. She complained that the manual had a mistake. There was a recipe in it that called for the user to measure 5/8 of a cup of raisins (Morphy Richards, 2004: 14) with a measuring cup. The problem, she said, was that the standard cup did not provide such a mark. To translate the manual Ribeiro had to calculate 5/8 by counting 20 out of the 32 subdivisions in ‘ounces’ that were provided. The cup had an opportunity of use for Ribeiro – to measure 5/8 of a cup – that was not clear to the machine owner. (Though this in itself provides another illustration since Ribeiro, not used to the English measuring system, had no idea what the symbol ‘Oz’, which was inscribed on the cup, meant!) The examples above show that explicit pieces of knowledge can only be understood by others, or serve as a way of transferring knowledge, if the individuals concerned already share some similar experiences or backgrounds. Polanyi (2002 [1958]: 31) makes the point in discussing maxims: ‘Maxims are rules, the correct application of which is part of the art which they govern. The true maxims of golfing or of poetry increase our insight into golfing or poetry and may even give valuable guidance to golfers and poets; but these maxims would instantly condemn themselves to absurdity if they tried to replace the golfer’s skill or the poet’s art. Maxims cannot be understood, still less applied by anyone not already possessing a good practical knowledge of the art.’ (See also Wittgenstein (1953 [1976]), Quine (1966 [1959]), Collins (1974), Tsoukas (2005), and Polanyi (1969) quoted by Gourlay (2007) for the general argument that ‘explicit’ forms of knowledge presuppose shared tacit understandings.) That is, analogies, metaphors, maxims, ‘stories’ (Orr, 1990; Stiles, 1995), ‘MBA case-studies’ (Adler, 1995), manuals, books, and so on are means of enculturating someone into a social group. But it is the whole process of socialization that allows one to acquire the tacit rules regarding a task or a community which, in turn, enable them to perform polimorphic actions. It is all this tacit knowledge that is part of a culture where 20/26 natural language, cooking, and domestic appliances are commonplace which was taken for granted in the earlier analyses of the bread-making machine and its instruction manual. Final Remarks To make bread without and with a breadmaker demonstrated that when humans perform a task mimeomorphic and polimorphic actions are mostly carried out in concert. When mechanization takes place, what is done in an intertwined way by humans is separated into distinct actions and the behavioral counterparts of actions, which are distributed among humans and machines, some of them distant from the kitchen. To overlook and mix up these transfers and how machines and humans interact can lead one to claim that machines can do more than mimic mimeomorphic actions. This problem is also present on a much grander scale in many other settings. The most spectacular is the failure of the attempt to replace human skills by expert system programs in computers, most notably the Japanese ‘Fifth Generation’ project. Launched in 1981 by Japan’s Ministry of International Trade and Industry, the fifth-generation computer initiative was a ‘bold attempt to leapfrog the advancement of computing technology to produce a computer that processed knowledge rather than numbers’ (Cross, 1992: 16) and ‘solve problems with human-style reasoning’ (Reid, 1992: c1). The problem with expert systems is that, without the contribution of the social knowledge of users, they simply fail whenever they are confronted with a problem marginally outside their data base (Collins, 1990). What is particularly surprising is that Nonaka and Takeuchi’s (1995) description of the breadmaker, coming from Japan as it does, echoes the assumptions of the Fifth Generation enthusiasts just after it was considered a failure (Cross, 1992; Pollack, 1992; Reid, 1992). The breadmaker case study, then, far from being an example of the incorporation of human skills into a machine, is an example of how a machine can be made to work without incorporating human skills. The lessons of this small case study can be transferred to 21/26 automation and anything else which involves tacit knowledge. For example, [first author’s surname] has found that the analysis of mimeomorphic and polimorphic actions which fits the breadmaker can be used, almost without modification, to explain the difficulties of transferring the tacit knowledge of steel and mining technologies from Japan and Australia to Brazil. The analysis also applies to attempts to automate specific tasks, whole factories and even science. And, of course, it should be thus since the overall critique of automation and artificial intelligence goes back at least as far as 1972 (Dreyfus, 1979 [1972]; Winograd and Flores, 1986; Suchman, 1994 [1987]; Collins, 1990; Collins and Kusch, 1998) and for the specific case of science, see the argument about Herbert Simon’s BACON program, which was claimed to be able to discover scientific laws (Collins, 1989, 1991; Simon, 1991) or, more recently but yet not discussed, the ‘Robot Scientist’ (Roach, 2004). Machines and pieces of ‘explicit’ knowledge, such as instruction manuals and books, are deceptive. Their meaning seems to be carried within them but actually it is provided by us. Their potential lies in the tacit knowledge and social understanding brought to their use by both their producers and their users. This is acquired through common enculturation and socialization within similar groups or forms of life. The point is best understood by the realization that there are two kinds of human action which relate in different ways to mechanization: machines can mimic mimeomorphic actions without loss but they fail when asked to reproduce polimorphic actions. A much better understanding of the process and possibilities of mechanization results from first decomposing human actions into their mimeomorphic and polimorphic components. With such a decomposition in hand, far better automated machines can be built and there will be far fewer false dawns and disappointed users. 22/26 Note The authors acknowledge Sara Delamont for previous readings of this manuscript, and Eliana Keen and Ana Tonani for their support for the bread-making experiences. The first author is also grateful to CAPES Foundation, which is funding his research studies at Cardiff University. Any person or institution that publishes anything that mentions, discusses, broadens or contradicts the content of this paper, is kindly requested to send to the authors the references of the published material in order to make the process of discussion and generation of knowledge dynamic, thus offering contributions and increasing the opportunities for the growth of the parties involved and/or related areas of knowledge. References Adler, Paul S. 1995 ‘Comment on I. Nonaka; Managing innovation as an organizational knowledge creation process’ in Technology Management and Corporate Strategies: A Tricontinental Perspective. José Allouche and Gérard Pogorel (eds.), 110-124. Amsterdam: Elsevier. Collins, Harry M. 1974 ‘The TEA set: tacit knowledge and scientific networks’. Science Studies 4: 165-186. Collins, Harry M. 1990 Artificial experts: social knowledge and intelligent machines. Cambridge, MA: The MIT Press. Collins, Harry M. (in preparation) ‘Bicycling on the moon: collective tacit knowledge and somatic-limit tacit knowledge’. Collins, Harry M., and Kusch, Martin 1998 The shape of actions: what humans and machines can do. Cambridge, MA: MIT Press. Cross, M. 1992 ‘Japan’s Fifth Generation fails the artificial intelligence test’ in The Independent, June 8, Science and Technology: 16. 23/26 D’Eredita, Michael A., and Charmine Barreto 2007 ‘How does tacit knowledge proliferate? An episodic-base perspective’. Organization Studies (forthcoming). Deutch, Yvonne (ed.) 1976 Bake your own bread. London: Sir Joseph Causton and Sons Limited. Dreyfus, Hubert L. 1979 [1972] What computers can’t do; the limits of artificial intelligence. New York: Harper and Row. Essers, J., and Schreinemakers, J. 1997 ‘Nonaka’s Subjectivist Conception of Knowledge in Corporate Knowledge Management’. Knowledge Organization, 24: 24-32 Gourlay, Stephen 2007 ‘Conceptualizing knowledge creation: a critique of Nonaka’s theory’. Knowledge Management Research and Practice (forthcoming). MacKenzie, Donald, and Graham Spinard 1995 ‘Tacit knowledge, weapons design, and the uninvention of nuclear weapons. American Journal of Sociology, 110 (1): 44-99. Morphy Richards 2004 Compact breadmaker. South Yorkshire: The After Sales Division, Morphy Richards. Nonaka, Ikujiro and Hirotaka Takeuchi 1995 The Knowledge-creating Company. Oxford: Oxford University Press. Pollack, A. 1992 ‘“Fifth-generation” became Japan’s lost generation’ in New York Times, June 5, Financial Desk Late Edition, Section D: 1. Polanyi, Michael 1969 Knowing and being in Knowing and Being. Marjorie Grene (ed.), 123-137. London: Routledge. Quine, Willard V. 24/26 1966 [1959] ‘Meaning and translation’ in On Translation. Reuben A. Brower (ed.), 148-172. New York: Oxford University Press. Reid, T. R. 1992 ‘Japanese government ends development of computer – “Fifth Generation” falls short of goals’ in Whashington Post, June 2, Financial: c1. Roach, J. 2004 ‘“Robot Scientist” said to equal humans at some tasks’. [WWW] National Geographic News. <URL: http://news.nationalgeographic.com/news/2004/01/0114_040114_robot.html > [Accessed 26th June 2006] Suchman, Lucy 1994 [1987] Plans and situated actions. Cambridge: Cambridge University Press. Tsoukas, Haridimos 2005 Complex knowledge. Oxford: Oxford University Press. Winograd, Terry, and Fernando Flores 1986 Understanding Computers and Cognition: A New Foundation for Design. New Jersey: Ablex. Wittgenstein, Ludwick 1976 [1953] Philosophical Investigations. Oxford: Blackwell. Woolgar, Steve 1991 ‘Configuring The User: the case of usability trials’ in A sociology of monsters: essays on power, technology and domination. John Law (ed.),58-99. London: Routledge. Rodrigo Ribeiro Rodrigo Ribeiro is a doctoral student at the School of Social Sciences, Cardiff University, and a lecturer at the Department of Production Engineering of the Federal University of Minas Gerais, Brazil. His main interests and research are on technical, social, linguistic and cultural aspects of knowledge and technology transfer. His paper, ‘The Language Barrier as an Aid to Communication’, which deals with transfer of steelmaking technology between Japan and Brazil, will be published in Social Studies of Science in 2007. Address 1 (up to Sep/07): Cardiff University School of Social Sciences, Glamorgan Building, King Edward VII Avenue, Cardiff, CF10 3WT, U.K. 25/26 Harry Collins Email: RibeiroR@cardiff.ac.uk Address 2 (Oct/07 onwards): Rua Engenheiro Senna Freire 612, Belo Horizonte, CEP: 30360-660, Minas Gerais, Brazil. Email: RibeiroR@dep.ufmg.br Harry Collins is Distinguished Research Professor of Sociology and Director of the Centre for the Study of Knowledge, Expertise and Science (KES) at Cardiff University. His thirteen books include three research monographs on the sociology of scientific knowledge, for example Gravity’s Shadow: The search for gravitational waves (2004, Chicago), and two analyzing artificial intelligence. The introductory The Golem: What you should know about science has been followed by volumes on technology and medicine. 2007 should see the publication of Collins and Evans: Expertise: A new analysis (Chicago). Address: Cardiff University School of Social Sciences Glamorgan Building King Edward VII Avenue Cardiff - CF10 3WT - U.K. Email: CollinsH@cardiff.ac.uk 26/26