Eyeballing expertise

Introduction

Within science and technology studies (STS), Harry Collins’ work, individually and in collaboration with Robert Evans, on ‘tacit’ and ‘explicit’ knowledge in relation to expertise has been a landmark for considering the nature of scientific thought and activities and their relationship to mundane and ‘societal’ considerations. ¹ There are several thrusts to Collins’ and Collins and Evans’ arguments that have spanned a number of decades: (1) a concern with the appearance of tacit knowledge in the activities of experts (e.g. Collins, 1974); (2) a concern with preserving the possibility of studying expertise as ‘real’ and substantive, which Collins and Evans (2002, 2007) find lacking in actor–network theory and constructivist STS; and (3) a commitment to promoting the recognition and fostering the management of expertise by providing analytical distinctions regarding expertise and its reliance on tacit knowledge (e.g. Collins, 2010; Collins and Evans, 2002, 2014).

As these thrusts have developed, the treatment of concepts like ‘tacit knowledge’ and ‘expert’ has increasingly turned into a theoretically grounded analysis geared toward helping readers distinguish between different forms of knowledge and different modalities of expertise. The schemas Collins and Evans provide deliberately maintain some distance from how people display, assert, attribute, negotiate, contest, and experience expertise in actual courses of action and interaction. Studying these matters, or treating expertise and tacit knowledge as ‘actor’s categories’, according to Collins and Evans, is insufficient if we want to understand what constitutes expert knowledge and draw normative conclusions about how it should be deployed in democratic societies.

In this article, we take issue with Collins and Evans’ (2002) assertion that expertise and its importance for society are best understood by treating expertise ‘as an analyst’s category as well as an actor’s category’ (p. 240, italics removed). ² We show that the concerns represented by the three thrusts in Collins’ and Collins and Evans’ work can, in fact, be addressed empirically by locating them in the occasions in which actual people (when they do) articulate and display tacit knowledge, explicit knowledge, and expertise as relevant for and in their undertakings. We are not denying that the terms ‘tacit’ and ‘explicit’ may be relevant in how persons organize their activities and interactions where they animate those terms in their practical ways. However, the usefulness of categorizing knowledge as ‘tacit’ or ‘explicit’ or ‘expertise’ needs to be made accountable in developing an appropriate understanding, and an understanding of what would be appropriate, of the things people do and the ways in which they do them, and the displays of what people know and how they know it, on the occasions that provide for their relevance.

This article attempts to follow through on these arguments by grounding tacit and explicit knowledge and expertise in the actual doings involved in detecting and judging the severity of problems in the retina that are associated with diabetes complications commonly referred to as ‘diabetic retinopathy’. People whose training and job scope qualifies them to make such assessments range from retinal specialists to family physicians to specially trained technicians known as ‘graders’. ³ They do grading by looking for telltale signs of, among other things, bleeding and the growth of new vessels on the interior surface of the eye opposite the lens (the fundus) in digital images of the fundus. In this article, we examine graders’ practices of assessing fundus photographs for signs of diabetic retinopathy and their categorization of these signs in the context of research on the (eye-related) complications of diabetes. ⁴ In their grading activities, graders seemingly display matters that have been of interest to Collins and Evans, and to others in STS and related domains: explicit knowledge of the formulations of retinopathy, tacit knowledge of matters involved in retinopathy but not explicitly articulated, and expertise.

In relation to the above-mentioned thrusts in Collins’ and Collins and Evans’ work, this article proceeds as follows. First, we note that the activities of graders, and the knowledge on which these depend, demonstrably involve more than a set of formal, explicit procedures; they also involve mundane and ordinarily encountered competencies. We bring this observation to bear on Collins’ (1974) arguments that ‘all knowledge, however pure, consists, in part, of tacit rules …’ (p. 167). Second, we ask what makes diabetic retinopathy grading distinctively what it is. We bring our observations of the constitutive practices and methods of diabetic retinopathy grading to bear on Collins and Evans’ (2002) claim that the study of expertise requires a ‘realist’ approach that ‘emphasize[s] the role of expertise as an analyst’s category as well as an actor’s category’ (p. 240, emphasis removed). Third, we note that it is apparent that graders’ everyday concern is the job of classifying eyes, and we ask what we get to know about grading by characterizing it as a particular type of expertise. We bring this question to bear on Collins and Evans’ commitment to offering normative guidance regarding the role of expertise in decision-making based on the ‘what’ rather than the ‘who’ of expertise.

In presenting our study of eye grading, we recognize that Collins and Evans’ research concerns and orientation differ from ours in several fundamental respects. However, we do share with them an interest in what expertise consists of. In this respect, we see value in discussing the work of graders in relation to some of the key assumptions and distinctions that underpin Collins and Evans’ policy-facing approach, and in relation to the conceptual tools they offer to facilitate thinking about expertise (including the distinction between ‘tacit’ and ‘explicit’ knowledge). By maintaining thematic continuity with the three thrusts identified in Collins and Evans’ work, we attempt to show that treating expertise and tacit knowledge as actor’s categories – by which we mean treating them, and the activities relating to them, in terms of how they are ordinarily understood by those involved – helps further our understanding of how science works, what expertise consists of, and the normative statements STS scholars could make about these matters. Our ambition is not to provide (or even attempt to provide) an alternative generalized account of expertise, but to remind those with an interest in these matters that examining how science is recognizably organized in the details of its doing allows us to broach the concomitant issue of what expertise consists of, and the occasioned relevancy of its use here.

Entwined tacit and explicit knowledge

Early articulations of the concept of tacit knowledge, by Polanyi in 1958, and later by Collins in the 1970s, positioned it as saying something new and unexpected about science and the methods by which science proceeds. In particular, the concept provided a way to challenge the received (and considered inaccurate) understanding of science as based on fully explicated empirical procedures that guarantee its objectivity. In Collins’ study of the TEA laser, tacit knowledge became part of the broad constructivist agenda of demonstrating the social nature of scientific work by attending to scientific practice. If science embodies tacit knowledge, which unlike scientific knowledge is ‘impossible to formulate in principle’ (Collins, 1974: 167), then science is something else than it is claimed to be in its received understanding. In its own terms, its objective status is questionable if it embodies that which cannot be formulated. ⁵ For Collins (1974), tacit knowledge is that which is germane to a ‘paradigm group’, and it can vary from one group to another. Tacit knowledge in these groups is learnt in the process of conducting science; it is knowledge that has been built up regarding how to do the science, and it is not explicitly formulated in the processes and procedures of the scientific endeavour.

The work of diabetic retinopathy grading, the subject of our study, certainly involves more than a set of formal explicit procedures, though it also involves those procedures. This becomes apparent in the ways in which the graders follow the protocol for grading. The protocol defines which features on a fundus photograph are worthy of note (see Figure 1). ⁶ For example, a key indicator of early-stage diabetic retinopathy is the presence of dilations in the smallest blood vessels known as microaneurysms, which are visible on a photograph as small red dots. The protocol also specifies how lesions should be reported. The basic decision that graders have to make for each particular lesion is whether it is present or absent in a particular eye. In the protocol, this judgement is associated with a percentage:

If there is a suggestion that a lesion may be present, but the grader is less than 50% certain that the lesion is in fact present, the grader uses none, or absent, for that lesion. ‘Definite’ lesions are those for which a grader is ‘at least 90% certain’. (Diabetic retinopathy (DR) protocol, emphasis in original)

OPEN IN VIEWER

Figure 1.

Overview of the lesions associated with the milder, non-proliferative, form of diabetic retinopathy (DR). These are the lesions most commonly encountered in the graders’ everyday work.

If a grader rates his or her certainty level between 50% and 90%, the lesion should be marked as ‘questionable’ and a second, generally more experienced, grader will be consulted on the case.

For some lesions, once they are confirmed as present, graders need to indicate how severe they are. Once microaneurysms have been spotted, for example, they need to be grouped together with haemorrhages to rate the level of ‘HMA’, the acronym for Haemorrhages and MicroAneurysms, or what graders colloquially call ‘blood’ in the eye. This is where so-called Standard Photographs (SPs) are used as benchmarks (see Figure 2). ⁷ SPs 1 and 2A are of eyes that have been differently affected by microaneurysms (red dots) and haemorrhages (red patches). Graders need to compare the amount of blood seen on the photograph they are grading with these standards and determine whether it is ‘less or equal than SP1’, ‘greater than SP1 but smaller than SP2A’, or ‘greater than SP2A’. Other SPs, in total more than a dozen, are used to rate the severity of other types of lesions.

OPEN IN VIEWER

Figure 2.

Examples of Standard Photographs generated by the Early Treatment Diabetic Retinopathy Study (ETDRS): Standard 1 and Standard 2A.

After considering the presence and severity of 14 types of lesions in one eye, the grader assigns an overall severity rating for diabetic retinopathy. The severity scale in use in the Singaporean centre where we made our observations contained 11 levels, in ascending order of severity from 10: ‘DR absent’ to 81 and 85: ‘Advanced PDR’ (proliferative diabetic retinopathy). It also contained two additional options to indicate that the condition had already been treated (‘inactive PDR’) or that the fundus images were of insufficient quality to be graded (‘ungradable’). Figure 3 shows a copy of this classification scheme with the technical definitions associated with each level. A hardcopy of this scheme is kept ready to hand by every grader when grading. Thus, a grader might encounter an eye in which ‘microaneurysms’ are present, but other lesions absent. Looking at the classification scheme, the grader sees that this can be classified as ‘minimal NPDR’ (non-proliferative diabetic retinopathy), which in turn is known as level 20. This overall severity rating for the eye ends up in a spreadsheet together with a numerical classification indicating the absence, presence, and (where applicable) severity of each of the lesions represented in the scheme. These numbers in turn become data for research. Examples of such research at the centre where we conducted observations include epidemiological studies on prevalence and risk factors of diabetic retinopathy (e.g. Wong et al., 2008) and the study of associations between diabetic retinopathy and other diabetes-related complications such as stroke (Cheung et al., 2007).

OPEN IN VIEWER

Figure 3.

The Diabetic Retinopathy classification scheme in use at the Singapore Eye Research Institute. Note how the Standard Photographs – indicated with the acronym ‘SP’ – are invoked in about half of the levels.

In their grading activities, graders can be observed to follow the formulated procedures, and they are able to formulate what they do in terms of those procedures. These activities are recognizable as grading activities by other graders, who are thus able to substantiate a colleague’s actions as observably accountable to the formal procedures. In this respect, in their grading activities, graders display a mastery of the formal procedures, and thus a seeming expertise in grading, certainly in terms of Collins’ (1974) definition of expertise. Indeed, their expertise is institutionally attested to as they move from being novices who are learning to grade to fully accredited graders who are recognized as having displayed, to the satisfaction of their community’s standards, that they have a mastery of the grading protocols and procedures. However, while displaying what is recognizably (to other graders) an expertise in grading, graders also display knowledge of how to do grading that is not part of the formal procedures. Indeed, the notion that knowing how to do grading encompasses more than what is set out in the procedures is conveyed to novice graders when they first report for work. In the introductory training session, grading manager Haslina Hamzah ⁸ tells the new recruits,

So I’ll take you step by step [through] how to grade diabetic retinopathy. It’s more of an overview, but when you first start off with grading you will be mentored by somebody who is senior, like, say, like Aniya, or Wen Jie or myself or Tina who will be able to […] help you pinpoint whether this is a haemorrhage, this is a microaneurysm, how severe is the lesion, and things like that. So it takes a bit of mentoring; this is just an overview presentation.

Novices are told that an introduction to common signs of diabetic retinopathy, and what these look like on a fundus image, will not readily enable them to recognize and grade haemorrhages and microaneurysms across a corpus of images. To be able to act as graders, novices first need to be mentored. Through mentoring, graders acquire the kind of knowledge that Collins’ (1974) use of the term tacit is intended to capture, namely the knowledge upon which the following of the formal procedures relies. This knowledge is learnt in the very course of familiarizing oneself with the formal procedures, and learning how to apply them.

Let us single out a few aspects of grading to discuss what this entails. Graders look from the image at hand to their SPs and back again in order to assess whether to mark the level of ‘blood’ as ‘more or less than SP1’ or ‘more or less than SP2A’ (Figure 4). Yet how to compare the blood in the image at hand with the blood shown in the standards is something that is not written in the training slides or communicated as part of briefing sessions. Rather, in the course of working on particular cases, and doing grading with others, graders figure out how to do such comparisons confidently in their mind’s eye: how to map – as one grader put it – the ‘big bleeding’ on Standard 2A to two smaller patches on the image being graded, and how to group scattered microaneurysms and haemorrhages to assess whether they ‘roughly add up’ to the amount of blood shown on the standard image.

OPEN IN VIEWER

Figure 4.

A grader assesses the severity of haemorrhages and microaneurysms in a patient’s image in relation to Standard Photograph 2A.

In this respect, graders would seem to exemplify the ideas concerning explicit and tacit knowledge articulated by Collins. They display a mastery of the formal procedures of their undertaking, and thereby demonstrate explicit knowledge of these. However, they have also learnt, through being part of their community, and in the course of their grading activities, how to assess whether the ‘blood’ in an image they are viewing qualifies as being ‘more’ or ‘less’ than that shown in an SP. This ‘more or less’ judgement is an example of what Collins might call tacit knowledge at work in the activities of the graders, because how to make that judgement is something that is not formulated in their procedures but is learnt in the course of developing experience.

However, while there is no doubt that graders learn, through being part of their community and in the course of their grading activities, how to deploy ‘more or less’ judgements in their work, it is important to note that this particular way of judging or weighing things – namely, by determining whether something is ‘more’ or ‘less’ than something else – is not itself learnt in learning how to do grading. Making such quantitative judgements is an ordinary, everyday method of deciding something: it is not confined to professional grading and neither is it likely that graders will not have had to make such an order of judgement in other situations outside of their work. In making ‘more or less’ judgements about the ‘blood’ in the fundus images they are viewing, graders display what they know about making quantitative judgements, and this knowledge is grounded in common sense.

Another example of commonsense competencies that relate to the work of grading is ‘knowing how to determine whether two instances are similar to or different from each other’. Similarity judging comes into play when graders assess a red dot in one part of the image, which they may decide is a microaneurysm, in relation to a red dot in another part, which they may decide is an artefact. It also comes into play when graders look back and forth between the image and the SP to identify blood they can match. Like making ‘more or less’ judgements, similarity judging is an ordinary competency deployed to achieve many things. For example, printers judge the quality of a printed colour against a pantone standard, looking from the colour of the printed page to the standard and back again to judge similarity and dissimilarity in tone, hue, and saturation of colour. Outside of work activities, people may amuse themselves with spotting dissimilarities in seemingly similar cartoons in the newspaper.

Consequently, while the graders’ following of the grading protocols involves drawing on stocks of knowledge that are not articulated by the formal protocols but are built up through learning to grade, at the same time it involves drawing upon stocks of cultural knowledge – for example, how to make a ‘more or less’ judgement and how to determine similarity or difference – which are neither articulated in the formalized protocols and procedures for grading, nor acquired as particular to learning to follow the protocols and procedures. As such, the work of grading illustrates something common to all of what Garfinkel (2002) referred to as instructed action. Following formal procedures, instructions, directions, rules and the like is simply marbled through with everyday commonsense knowledge of how to do things. Wittgenstein’s (2009 [1953]) description of rule following drew attention to this when he described how a rule is mute with respect to its own application. For example, the signpost on the road tells us how far it is to a particular place, but it does not in itself tell us in which direction that place is. We have to draw on our stock of cultural knowledge to know that it is the one end of the sign rather than the other that indicates the direction. In a study of how coders follow instructions for analysing hospital records, Garfinkel (1967: 18–24) demonstrated the entwining of commonsense methods and formal procedures – noting that ‘ad hocing practices are used in order to recognize what the instructions are definitely talking about’ (p. 22, italics original).

We can draw a conclusion that bears upon the relationship between tacit and explicit knowledge as developed in Collins’ early work. Collins deconstructed scientific knowledge into two component parts, ‘tacit’ and ‘explicit’ knowledge, and showed the latter to be dependent on the former. This served a purpose in reorienting understandings about science and its methods. However, the conceptual separation of tacit and explicit knowledge, even if this was not the intention of Collins’ work at the time, also set a precedent for treating ‘tacit’ and ‘explicit’ as abstract categories of knowledge in their own right. The point that we can take from Wittgenstein and Garfinkel, and which is illustrated in the work of the graders, is that, in following the formulated rule, or in using formulated knowledge, the rule and the explicit knowledge cannot be separated from what might be considered tacit knowledge; in following the rule, or in articulating the explicit knowledge, what is said to be tacit knowledge has to come into play. This is not a failing in formal method: it is just the ‘logic’ of the action of following a procedure that it will involve matters not specified by the procedure.

Indeed, in Wittgenstein’s description of rules and Garfinkel’s description of coders, the issue is not that rules and instructions, even the clearest and most detailed ones, are incomplete, because how could it be possible to include what anyone must do to follow them? The issue is how, in and through their following of rules or instructions – which necessarily entwines knowledge of the procedure or rule and commonsense knowledge of matters relevant to following it – people organize the details of their undertaking so that this undertaking is recognizable, and is recognizably what it is, to those in the know. This is something we have yet to articulate for the work of grading, and to which we will turn our attention in the next section.

Expertise and its constituents

The second thrust we noted in Collins’ more recent work with Evans was a concern with preserving the possibility of studying expertise as ‘real’ and substantive. This has sometimes taken the form of a defence of science against those accused of seeking to deconstruct it. Collins and Evans have at times singled out proponents of actor–network theory, an approach that they view as dissolving expertise – though they also have expressed dissatisfaction with the direction that constructivist STS in general has taken with regard to the status of science. In their ‘waves’ article (Collins and Evans, 2002) and in the book Rethinking Expertise (Collins and Evans, 2007), Collins and Evans present a stark choice between ‘relational’ approaches to expertise, which they say help pinpoint how expert status is negotiated but not what substantive expertise consists of, and their own ‘realist/substantive’ sociology of expertise built around a theoretically derived classification scheme that defines, at different levels, the various knowledges and abilities people draw on when they make judgements. The notion of ‘tacit knowledge’ at the heart of the scheme is intended to reinforce this demarcation between expertise as attributed by others and expertise as acquired over time and through the efforts of an individual in gaining and maintaining membership in an expert community. ⁹

Our concern here is not whether Collins and Evans’ conclusions regarding actor–network theory or the ‘Second Wave’ are justified, but whether a theoretically derived ‘Periodic Table of Expertises’ is the most appropriate means to elaborate and understand the substantial nature of expertise. ¹⁰ The way in which Collins and Evans carve out a space for ‘realist/substantive’ studies of expertise, namely by opposing these to work that treats expertise as an actor’s category, seems to imply that analyst categorization is necessary to understand what expertise consists of and what it ‘is’, substantively. But why would this be the case? If we are attempting to substantiate the notion that scientists, for example, have a special kind of expertise, it makes sense to consider how what is actually done in the course of doing scientific work somehow constitutes it as special. In this respect, it is important to note that people (scientists and others) do not move through the world as experts; whenever this is relevant as a description, they move through the world as experts in ‘something’. In these terms, expertise may be substantiated in understanding the particular details of its practical constitution, its relevance as a description, how it is carried in people’s actions and its recognizability. Garfinkel’s (1967, 2002) thinking on the ‘the interactional what’ of the matter being done is instructive in this respect. By ‘the interactional what’, or the ‘thises and thats’, ‘the haecceities’ of the matter, Garfinkel meant to draw attention to how, in what they do, members achieve the recognizability of what they do as, for example, ‘making music together’. Similarly, it is in the ‘thises and thats’ that expertise is claimed and demonstrated, and made relevant as a description of some human activities, and consequently, it is at this level of consideration that the substantiation or the refutation of expertise resides. ¹¹

We are, thus, not so much at odds with Collins and Evans’ intention to substantiate expertise as a domain of human activity; we are at odds with their way of substantiating it. While their work does not lack evocative examples, its particular theoretical nature does not provide for how cases of expertise are ordinarily encountered in the world. This then begs the question how the analysis of particular instances of judgement or decision-making as involving a certain ‘type’ of knowledge and constituting a certain ‘kind’ of expertise is warranted. In order to demonstrate how expertise can be relocated in the ordinary world of its relevant doing, we now return to the work of those involved in grading images of the fundus. We shall need to examine the practical activities of the graders in some detail, to understand how, in the details of what they do, they display and recognize an expertise in grading. In so doing we can make warrantable the description of their activities as expertise (in grading).

Working with and within a taxonomy

Part of the graders’ expertise is displayed in their making their work accountable to a grading taxonomy of retinopathy. It is the taxonomy that provides a structure for what the graders should look out for and treat as notable, or disregard and treat as unremarkable. In doing their work, graders display knowledge of the taxonomy as well as how to work within that taxonomy. This knowledge and work are thus on display, and can consequently be recognized by others working within the taxonomy who can judge the expertise of a grader by referencing his or her activities to the terms of the taxonomic structure. Grading judgements are then both questionable and verifiable within the taxonomy and are preserve-able in documentable terms drawn from the taxonomy. The fact that expertise in grading is displayed in the manner in which grading is done so that it is accountable to a taxonomy of retinopathy is manifest in a number of ways in grading activities:

Naming

Part of the expertise that graders display is the ability to name occurrences in the images of the fundus they are viewing, using technical terms drawn from the nomenclature of retinopathy. Non-experts may view these images and be able to see and point to occurrences in the images that stand out against the more uniform background colour of the fundus, using vernacular terms such as ‘blotches’, ‘squiggles’, ‘marks’, ‘smears’, and the like. Graders, however, assign such occurrences particular technical terms such as ‘venous beading’, ‘preretinal haemorrhage’ or ‘vitreous haemorrhage’.

To be sure, a grader may as well be able to point to an occurrence and describe it as a ‘squiggle’, or ‘a bead’ as may anyone, and may appeal to an occurrence’s obvious characteristics that are see-able by anyone. When training a neophyte, for example, a grader may point to the resemblance of venous beading to a necklace –‘this looks like a beaded necklace’. However, in the course of their grading activities, graders are able to name that necklace-like appearance of a vessel as ‘venous beading’, something that not anyone can do. Furthermore, occurrences that might be undifferentiated in describing them as ‘blotches’ may be differentiated by a grader; thus one blotch may be a preretinal haemorrhage, while another blotch may be a vitreous haemorrhage.

The expertise of grading then displays knowledge of the technical elements within a nomenclature and of the characteristics of those elements. For example, one of these characteristics is that within a grading nomenclature, a technical term does not have a diversity of meaning: it does not reside within different contexts of use that might otherwise provide it with different meanings. When technically used, a term embodies a universally agreed definition within the nomenclature; its meaning resides within the nomenclature as finite, and thus requires the nomenclature to be understood. Consequently, in naming an occurrence in terms drawn from the nomenclature, a grader is displaying knowledge that is held in common within the community of people working with the taxonomy of retinopathy and which is not available outside of that community.

As one might expect, there still can be disagreements as to whether an occurrence does or does not accord with a particular definition. There still can be appeals to contextual details of the image: for example, ‘given insufficient dilation of the eye, something looks to be one case when it may not be’. A grader may still consult another colleague to double check that what the image shows is a preretinal haemorrhage. Moreover, it may still turn out that an occurrence in the image was incorrectly named, or that an occurrence was missed. But if an occurrence is named as, for example, a ‘preretinal haemorrhage’, then everyone within the grading community knows what it is that a grader is saying he or she has observed. In this respect, the term may lack a meaning without reference to that community, and, consequently, outside of that community an explanation of the term’s use would have to make reference to its use within that community. An explanation could not gain a purchase in other uses, for either there are no other uses, or its technical use would have to be differentiated from other contexts of use. Naming occurrences in images of the fundus by using terms drawn from the nomenclature supplied and defined by the taxonomy of retinopathy then displays an expertise in grading as something that cannot be done by anyone.

The naming of an occurrence in the assignment of a technical term within the nomenclature of grading diabetic retinopathy also supports a number of ‘professional practices’ in grading, practices which are themselves, in part, constitutive of grading expertise.

First, it supports standardization. In naming a haemorrhage within the nomenclature of retinopathy, the observed haemorrhage is being described as ‘another one of these’, and consequently makes available to other graders the significance of what a grader has viewed. Thus, everyone using the taxonomy will know that the presence of intraretinal microvascular abnormalities (IRMAs), together with microaneurysms, pushes a case from mild to moderate diabetic retinopathy; they will also know that it can be difficult to spot IRMAs and that the grader may, therefore, have requested support from a colleague to establish their presence or absence. An occurrence is then standardized in its naming in the terms of the taxonomy of retinopathy – it is a ‘one-of-these’ and as such implicates routine matters that pertain to any such occurrence.

A second professional practice supported by naming within a nomenclature of retinopathy is efficient and economic expression. This in turn reduces the potential for ambiguity. Naming an occurrence a ‘blotch’ opens up questions as to the nature, size and so on of what was seen. Furthermore, what looks like a blotch to one grader may look like a smudge or spatter to another. When trading in the currency of technical terminology, the graders provide a clear, concise, definition of an occurrence for others to then act on. Because technical terms are used that have agreed-on meanings within the community, an occurrence does not have to be explained, its description does not have to be elaborated or disambiguated.

A third professional practice that is supported is objectivity. The terms within the nomenclature of retinopathy are ‘literal’ ones, used to literally describe, among other things, various ‘kinds’ of bleeding. Thus ‘vitreous haemorrhage’ literally means bleeding in the vitreous (the jelly-like substance in the eyeball), and ‘preretinal bleeding’ literally means bleeding before the retina (between the retina and the internal membrane of the eye). Contrast this with descriptions like ‘bad bleeding’ or ‘nasty bleeding’, both of which are terms that provide an assessment of the bleeding in their invocation. Graders do not use such evaluative descriptions. The judgements they make about what they see reside not within the nomenclature of retinopathy (whose terms are literal ones and hence support objectivity), but within the knowledge of different kinds of bleeding embodied within the taxonomy.

Graders thus display an expertise in grading through, among other ways, their naming of occurrences in the image of a fundus by utilizing technical terms developed within a nomenclature of retinopathy. It is through these technical terms that they view and understand the occurrences they are witnessing and make those occurrences visible and understandable to others within their community. It is through these technical terms that they, in part, distinguish their activities from the activities of others who are not experts in diabetic retinopathy grading.

Classifying eyes

Naming the occurrences of microaneurysms, haemorrhages and other features manifest in images of the retina is followed by graders then having to classify the eye by assigning it to one of several categories of severity. Graders display expertise in the classification of eyes in their use of a ranking scale drawn from the taxonomy of retinopathy (see Figure 3). This way of classifying eyes can be contrasted with other ways. For example, in Figure 5, ‘blotches’ can be seen in the images of the eyes. The extent of the blotch in Figure 5(c) might indicate, seemingly to any viewer, that something is badly wrong with the eye, but the blotch in Figure 5(b), although also seemingly bad, is much smaller. One conceivable way of classifying eyes could thus be in terms of ‘bad’, and ‘not so bad’. While such descriptions assign an eye to one of two categories of severity, they do not display an expertise in grading, for they do not classify the eye within the taxonomy of retinopathy. On the other hand, classifying an eye as a ‘35’ does display knowledge of the taxonomy of retinopathy and makes that classification a candidate for the status of an expert classification. We say ‘candidate’ because in itself, saying of an eye that it is a ‘35’ does not mean that it is expertly classified. It may merely have been classified by someone who, though knowing the grades within a classification system, knows nothing else and is just plucking a grade out of the air. But when such a classification is done as part of the practices of grading eyes, it displays an expertise in grading eyes. ¹²

OPEN IN VIEWER

Figure 5.

(a) Venous beading, and (b) Preretinal haemorrhage (both Early Treatment Diabetic Retinopathy Study Standard Photographs). (c) Vitreous haemorrhage (example retrieved from http://medlibes.com/entry/vitreous-haemorrhage).

As with the naming of haemorrhages in a way that supports an objective practice, the taxonomy also provides a system for classifying eyes that renders the classification descriptively neutral. As opposed to an eye being graded as ‘bad’ and ‘not so bad’, whereby both terms provide an understanding of the severity of the problem, the classification of an eye as part of formal grading activities is done with reference to a quantitative scale. In itself, ‘35’ does not reveal an understanding of a problem with an eye, nor does it pass a judgement regarding the severity of a problem. If the levels ran from 1 to 1000, and assuming that a higher number is worse than a lower one, then 35 might not be so bad, but if the levels ran from 1 to 40 then 35 might be considered bad. In the system the graders use, the quantitative scale is not continuous (it skips from 10, to 14 and 15, to 20, 35, 43…), and, while a higher number is generally worse than a lower one, the highest numbers are reserved for classifications that stand outside the order of severity judgement (90 = ‘inactive PDR’ and 99 = ‘ungradable’). However, it remains the case that a quantitatively classified eye can only be understood through the invocation of the grading levels. Thus, in assigning to an eye a level within the taxonomy of retinopathy, graders display their expertise in grading. Grading becomes a matter of quantification on the basis of the taxonomy, a knowledge of which is necessary for both the assignment of a level and for understanding what that level might mean. In this sense, it is the quantification of the eye according to the given criteria within a fixed classification system that assigns the severity of a problem. The graders’ judgement that an eye is of a certain level is grounded in that system and accountable to it.

Displays of commonsense knowledge within expertise

We mentioned above that graders draw upon their knowledge of how to make a ‘more or less’ judgement and how to determine similarity or difference when comparing an image of the fundus with photographic standards in order to be able to grade the occurrences they observe. As noted, this knowledge of how to compare is everyday knowledge used to accomplish a wide range of activities. Within grading, this knowledge, however, becomes part of the expertise of the graders in that the comparison is done within the taxonomic systems of retinopathy. This can be understood by realizing that the instruction ‘compare the image of the eye to the standards’ is not going to enable those who do not have an expertise in grading to do grading. Is it the background colour of the eye, its size, its shape that needs to be compared? Graders know to compare selected features of an image with selected features of the standard. These are features that are given within the taxonomic system, features of the eye-to-be-graded that are made relevant for the grader through that system. For example, as we have seen, SPs 1 and 2A are used to classify the amount of ‘blood’ in the eye. This requires graders to recognize certain red dots and blotches in both the standard and the image at hand as microaneurysms and haemorrhages, and to roughly quantify the cumulative size of the ‘blood’ on either side for the purpose of comparison. Other features present in the standard are irrelevant for the purposes of comparison done as part of grading.

In this respect, although grading draws on a commonsense method of comparison, doing that comparison within the taxonomic system of retinopathy (and making the terms of that comparison accountable to that taxonomy) constitutes the commonsense method of comparison, here, as part of the displayed expertise of the graders.

Managing expertise

The third thrust that can be discerned in the work of Collins and Collins and Evans is a commitment to promoting the recognition and fostering the management of expertise. The role and value of scientific expertise in the public sphere has been key to Collins and Evans’ (2002) arguments concerning ‘the waves’ they detect in, and project for, STS – and to the ensuing debate. ¹⁷ Collins and Evans argue that the constructivist underpinning of STS undermines the exclusivity of scientific knowledge and has ‘levelled the playing field’, by providing for others than those from the scientific community to be involved in science and technology-related policy formation and decision-making. For instance, they discuss Wynne’s (1992) argument concerning the inclusion of Cumbrian sheep farmers in the formation of a policy response to radioactive fallout from the explosion at Chernobyl in 1986, as confusing ‘stakeholders’ – whose claim to inclusion rests on their social identity – and ‘experts’ – whose claim to inclusion rests on what they know (Collins, 2011; Collins and Evans, 2002). While consulting stakeholders is, in many democracies, an important part of the process of achieving political legitimacy for a public decision, this should not, according to Collins and Evans, detract from the role and value of substantive expertise in conferring technical legitimacy. The sheep farmers’ social identity does not qualify them for participation in technical decision-making, which is not to say that they should not be included, only to clarify the grounds on which such inclusion should happen. Collins and Evans argue that such clarification requires the focus to shift from, as they put it, the ‘who’ to the ‘what’: from the social identities of stakeholders to the substantive issues at hand and the knowledge people have about these.

For Collins and Evans, then, returning to the ‘what’ of expertise is crucial for understanding the role of experts and the management of expertise in policy-making. We do not disagree with them on this point; however, in terms of our study of grading, we find problematic how Collins and Evans would arrive at this ‘what’. The root of the problem lies in the conceptual analysis of the nature of knowledge that underpins the distinctions presented in the ‘Periodic Table of Expertises’ (Collins and Evans, 2007) and operationalized in case studies and experiments conducted under the banner of SEE. The character of this analysis is well-illustrated in Collins’ (2010) book Tacit and Explicit Knowledge. In this theoretical treatise, Collins explores different kinds of tacit knowledge in relation to their ability to become explicable. A key aspect of the foundational work undertaken in Tacit and Explicit Knowledge involves the idea of ‘strings’. For Collins (2010), strings are physical entities that ‘contain information’ but that ‘in themselves are not meaningful’ (p. 16). On this view, Figure 3 can be considered a string. It requires an act of interpretation to turn it into ‘The Diabetic Retinopathy Classification Scheme’. In this respect the idea of strings separates and dislocates a thing or activity from the meaning that can be given to it, the interpretation that can be laid upon it. This separation, by way of a ‘philosophical ground-clearing exercise’ (Collins, 2010: 3), paves the way for refining the conceptual distinction between explicit and tacit knowledge, a distinction which in turn underlies the Periodic Table of Expertises. The effort to classify tacit and explicit knowledge through a conceptual scheme thus anchors the classification of expertise to the world it classifies: this is Collins’ way of proceeding to specify the ‘what’ of expertise. However, it exacerbates the problem we noted in the previous section, because this understanding of social action ignores that the very recognizability of social action is part of the action itself. ¹⁸ As Garfinkel (1967) has noted in his description of accountable action, action is done so that it is recognizable. If it were otherwise, it is difficult to understand how courses of action and interaction could ever proceed, for how would a co-participant ever be able to do appropriate next actions such as returning a greeting or answering a question if a first greeting or a question were not done so as to be recognizably those things? What an action is, what its sense is, is displayed in the course of its doing, rather than it being some bodily movement and sound awaiting interpretation. The sense of what an action is, is part and parcel of its doing.

If we proceed, as Collins proposes, on the basis of having a physical world that is interpreted, then what the matter at hand is is down to that interpretation (or the outcome of disputed interpretations). However, there still remains the action as done for those doing it. Indeed, it is the action as done that remains consequential for those doing it and those witnessing it, at least in as much as it is consequential for what they, and their interlocutors, might do next. Understanding that, of course, is not the object of Collins’ endeavours; the object for Collins is the doing of the interpretation, the definition and classification of knowledge and expertise, and this can proceed outside of the experience of those who exhibit knowledge or expertise. As a result, however, what is actually done, and how it is done, in its own terms, remains unaddressed in these endeavours.

In order to examine what remains when Collins and Evans’ classification scheme is applied to the world in order to interpret it, we have attempted to unpack the activities of the graders. In other words, we have located ‘the what’ of grading expertise as manifest in the doing of grading. However, if we change the question from ‘how is expertise manifested in the activities of the graders?’ to ‘what are the graders actually doing when classifying eyes in relation to diabetic retinopathy?’ we arrive at a seemingly strange answer. The answer is strange, on Collins and Evans’ rendition of expertise, because it is not that they are ‘doing expertise’. What they display and make recognizable when they do grading are the practices and methods of grading, not the practices and methods of expertise, whatever those could be. Calling the graders ‘experts’ is to call them experts in grading images of the fundus for signs of retinopathy; their job title is not ‘expert’, it is ‘grader’. Graders may be describable as experts, but the very issue of expertise only becomes relevant on occasions, such as, for example, when they are learning to grade and at some point are institutionally deemed to be expert enough – meaning to have displayed sufficient mastery of the practices and methods of grading – to do it for real as opposed to doing it on practice cases.

Indeed, taking out any references to ‘expert’ and ‘expertise’ in the previous section of this article does not change our description of what the graders do and how the graders do it, and does not change the sense that it has as they do it. What we are left with when these terms are taken out is what we have really been describing all along, which are the constitutive practices and methods of grading images of the fundus as a practical professional matter. In doing grading, graders display their understanding of, their knowledge of and their dexterity in the use of, these practices and methods, and in working these they make recognizable that they are grading. Rather than moving from the ‘who’ of expertise to the ‘what’ by means of an abstract schema, it is possible to relocate the latter in what is made redundant by this schema: the interactional thises and thats, the ‘whats’ of those concerned, the particular practices and methods that constitute a job of work. ¹⁹

Providing for ‘the what’ through a classification scheme that stands outside of actual courses of action and interaction leads Collins and Evans to make particular recommendations regarding expertise, and STS in policy matters. However, considering ‘the what’, as displayed in persons’ actions and interactions, provides an alternative way in which expertise and STS can be involved in policy-making. Collins and Evans’ typology provides for the selection and inclusion (or exclusion) of defined types of expertise in discussion of policy matters. Alternatively, however, studying the particular practices and methods that constitute the relevance of expertise may provide for the inclusion of relevant bodies of expert practice.

For example, there have been spectacular, high-profile, expensive errors in recent years with respect to policy decisions to develop new computer systems in the United Kingdom. The London Ambulance system was designed to automate the allocation of ambulance crews to emergencies, a new air traffic-control suite for Heathrow was intended to modernize air-traffic control and the National Health System database was intended to facilitate the easy movement of patient information. However, these, and other high-profile developments in the private sector such as the London Stock Exchange system, and lesser developments in both the private and public sectors, have not only failed, but failed at a high cost to the public and investors and brought embarrassment to government and corporate board members. There was a common element to all these failures: they ignored the practices and methods through which the work of those whom new systems were intended to support was constituted, and either impeded or prevented the operation of these practices and methods. ²⁰ This problem for systems development has been extensively documented within various studies of situated work. ²¹ In this respect, the technical expertise of systems developers may not be sufficient to support policy decisions regarding new developments that cost UK taxpayers and consumers many millions of pounds; money has been wasted on systems the technologists advised on, then promised but failed to deliver. By attending to the constitutive practices of the work, an understanding of the ‘whats’ of the knowledge and expertise that systems are intended to automate or support can then inform policy concerning the commitment to and the development of, for instance, publicly funded systems.

Therefore, if Singapore’s Agency for Integrated Care decided to automate diabetic retinopathy grading, it would do well to include grading expertise as well as software development and business expertise in efforts to this effect and to bring such expertise to bear on the question of whether new software should aim to automate or support grading (or perhaps a combination of both). With respect to the question of how one builds grading expertise into decision-making of this kind, Collins and Evans’ classification scheme might seem to provide an answer by, for example, offering the category of ‘interactional expertise’. However, this begs the question: the interactional expert must have expertise in some relevant undertaking, but how is the question of what is relevant to be judged independently from the practices and methods of that undertaking? Being able to identify a type of expertise does not give us the ‘what’; in order to understand what the relevant body of expert practice is, we need to examine how it is that practitioners recognize others as being able to know what they are talking about. Such an examination can then provide the basis for explicating, as we have attempted to do for the graders, what that body of knowledge consists of. ²²

Exactly how the expertise of the graders might be built into the decision-making process is an interesting question: do the software designers use traditional requirements capture techniques, ²³ do they involve the graders directly in the design process, as exemplified by the Scandinavian School of systems development, or do they incorporate the expertise of social scientists, such as ourselves, who study the work of the ‘experts’ under consideration? That any of the above is a candidate possibility and that any of the above could be corralled into a particular category type of expertise is predicated on first understanding what constitutes the expertise. Such an understanding is derived from an investigation of the practical actions and interactions involved. The relevance of a category of expertise in policy-making thus needs to be anchored in the practices and methods through which people organize their actions and interactions as expertise ‘in something’.

Conclusion

Tacit and explicit knowledge, their relationship and the associated idea of expertise make up an important theme in STS, and Collins and Evans have been very consequential in its development. We have not disputed Collins’ claim that there is more involved in doing science than is provided for by the formal procedures and specifications of science. We have not disputed Collins and Evans’ argument that social scientists should be able to provide descriptions and understandings of expertise as, in their words, substantive and ‘real’. We have also not disputed their claim that ‘real’ expertise has a particular place in how policy-making and decisions should proceed. However, our examination of the work of grading has made us uncomfortable with the way in which these themes have been substantiated and elaborated and how Collins and Evans have progressively considered tacit and explicit knowledge and expertise as abstract categories, divorced from vernacular descriptions and understandings that, however, are not, and cannot be, overturned by those categories. Our study of diabetic retinopathy grading has shown that studies of knowledge and expertise can and should attend to the very real ways in which such matters are displayed and witnessed as and in ordinary courses of action. Consequently, this article has not been concerned with dissolving an important theme in STS, or criticizing the interest that it holds, but with attempting to relocate that interest to the doings of those who display expertise, and to understanding how the relevancy of describing what they do as expertise in ‘this’ or ‘that’ has been occasioned. By doing so, we point to a way in which the themes of tacit and explicit knowledge and expertise can be considered that offers an alternative to the way propounded by Collins and Evans.