Read the article and then answer all of the questions (750 words): 1. What data do the authors draw on?? 2. Is there anything notable about how they gathered t

1

Abstract: In this paper we argue that the new availability of digital data sets allows one to revisit Gabriel Tarde’s (1843-1904) social theory that entirely dispensed with using notions such as individual or society. Our argument is that when it was impossible, cumbersome or simply slow to assemble and to navigate through the masses of information on particular items, it made sense to treat data about social connections by defining two levels: one for the element, the other for the aggregates. But once we have the experience of following individuals through their connections (which is often the case with profiles) it might be more rewarding to begin navigating datasets without making the distinction between the level of individual component and that of aggregated structure. It becomes possible to give some credibility to Tarde’s strange notion of ‘monads’. We claim that it is just this sort of navigational practice that is now made possible by digitally available databases and that such a practice could modify social theory if we could visualize this new type of exploration in a coherent way.

Bruno Latour*, Pablo Jensen¥, Tommaso Venturini*, Sébastian Grauwin¥ and Dominique Boullier* Sciences Po, Paris

¥ Institut Rhônalpin des Systèmes Complexes (IXXI) and Laboratoire de physique, UMR CNRS, ENS de Lyon

Correspondance should be adressed to!: [email protected]

Accepted by the British Journal of Sociology FINAL VERSION AFTER ENGLISH HAS BEEN CORRECTED

The Whole is Always Smaller Than Its Parts A Digital Test of Gabriel Tarde’s Monads1

1. We thank Terry N. Clark, Paul Girard, Grégoire Mallard, Dominique Pestre, Paul-André Rosen- tal, Livio Riboli-Sasco and the referees for their comments at various stages of this manuscript. We thank Michael Flower for checking the English.

2

It is generally accepted in the various sciences dealing with complex collective behaviour that there exist some fundamental di!erences between the individual and the aggregate levels (Knorr, 1981; Calhoun et al. 2007). This is why it seems common sense to state that there should exist two levels of analysis: the micro level that focuses on individuals; the macro level that focuses on the aggregates. The consequence of such a distinction is that almost all the questions raised by social theory have been framed as the search for the right pathway that leads from one level to the other: should the inquiry begin from the micro or from the macro level? Is the macro a mere aggregate or a sui generis? How do some macro features end up emerging out of the interactions going on at the micro level (Boudon, 1981)? Is it possible to ‘reconcile’ the two levels by another more encompassing theory (Bourdieu, 1972, Giddens, 1984)? Is it possible to imagine an intermediary level, a ‘meso’ one? And so on. This framing of questions is not limited to social theories dealing with humans, but has a bearing on all collections of non-humans living organisms (flocks of birds and swarms of social insects in particular) (Axelrod, 1984; Moussaid et al. 2009) as well as on the very notion of how an organism comes to be organized (for instance, how do individual cells relate to the whole body) (Dawkins, 1982)? Those same questions have been extended to a wide range of phenomena such as mental processes (Minsky, 1988) or artificial entities living in silico (for instance, multi-agents models) (Epstein & Axtell, 1996). Although this division in levels has had an enormous role in shaping many research programs in the natural and social sciences, it has also obfuscated the central phenomenon those sciences wished to account for: how to follow stronger, wider and longer lasting associations. By presupposing that there exist two levels, they might have solved too quickly the very questions they should have left open to inquiry: What is an element? What is an aggregate? Is there really a di!erence between the two? What is meant by a collective entity lasting in time? In this article, we wish to consider how digital traces left by actors inside newly available databases might modify the very position of those classical questions of social order. Our aim is to test an alternative social theory developed by Gabriel Tarde (1843-1904) in the early days of sociology but which never had any chance to be developed because of the lack of empirical tools adjusted to it (Tarde, 1903; Clark, 1969/2011; Milet, 1970; Candea, 2010). Instead of starting by saying that the really important question is ‘to find out how individual decisions relate to collective actions’, we want to do exactly what Tarde suggested and refrain from asking this question so as to lessen its import and to turn our attention to a di!erent topic: is there a way to define what is a longer lasting social order without making the assumption that there exist two levels (Latour, 2005)? To dramatize the contrast, we will claim that there is more complexity in the elements than in the aggregates, or stated a bit more provocatively that ‘The whole is always smaller than its parts’. We call this hypothesis ‘the one level standpoint’ (1-LS) in contrast with the ‘two level standpoint’ (2-LS). Such a hypothesis has a chance to fly only if it makes an empirical di!erence in the treatment of data. This is why we will attempt to demonstrate two points: a) some of the new digital techniques and in particular a few of the tools o!ered by network analysis may allow the tracing and visualization of the social phenomenon in a way that makes the 1-LS slightly more commonsensical than the 2-LS alternative; b) it might now be possible to account for longer lasting features of social order by learning to navigate through overlapping ‘monads’ instead of alternating between the two levels of individual and aggregate. (Note that in what follows, the adjective ‘social’ should not be limited to human agents but extended to all entities treated severally). To go some way toward proving our points, we will proceed in the following way: we will first make use of the notion of profile to give the general flavour of our argument (section 1); then, we will explain how our approach is di!erent from from the idea of structures emerging out of atomistic agents in interaction (section 2) and then how the notion of structure should be replaced by the circulation

Introduction

3

The gist of our argument may be o!ered by considering how profiles now available on so many digital platforms are quickly modifying the very definition of what individuals are —and correlatively how we should handle aggregates. Although this reduction of the social connections to html pages linked to other html pages will may sound too drastic, it is this experience of clicking our way through platforms such as Flickr™, Academia.edu™ or MySpace™, of surfing from document to document, encountering people and exploring communities without ever changing level that we wish to use as an occasion to rethink social theory. Of course, there exist many other platforms, but in this article we will draw heavily on Web 2.0 to exemplify our arguments because it has turned 1-LS navigation into a mainstream experience which might be captured in a sentence: the more you wish to pinpoint an actor, the more you have to deploy its network. Let’s take a simple example. We all have had the experience of preparing a meeting by searching on the web the name of the person we are soon to meet. If for instance we look on the web for the curriculum vitae of a scholar we have never heard of before, we will stumble on a list of items that are at first vague. Let’s say that we have been just told that ‘Hervé C.’ is now ‘professor of economics at Paris School of Management’. At the start of the search it is nothing more than a proper name. Then, we learn that he has a ‘PhD from Penn University’, ‘has written on voting patterns among corporate stake holders’, ‘has demonstrated a theorem on the impossibility of aggregation’, etc. If we go on through the list of attributes, the definition will expand until paradoxically it will narrow down to a more and more particular instance. Very quickly, just as in the kid game of Q and A, we will zero in on one name and one name only, for the unique solution: ‘Hervé C.’. Who is this actor? Answer: this network. What was at first a meaningless string of words with no content, a mere dot, now possesses a content, an interior, that is, a network summarized by one now fully specified proper name. The set of attributes —the network— may now be grasped as an envelope —the actor— that encapsulates its content in one shorthand notation. In such a now common exploration, an entity is entirely defined by the open-ended lists in the databases. Using the terminology of actor-network-theory (ANT), an actor is defined by its network (Law, 1999). This network is not a second level added to that of the individual, but exactly the same level di!erently deployed. In going from the actor to its network, we remain safely inside the 1-LS (Law, 2004). The main point is that this definition is entirely reversible: a network is fully defined by its actors. If we now wished to go from this particular professor to some of his attributes, we might not be forced to change levels: the paradigm of ‘stakeholders voting’ will be defined by another list, this time the list of ‘all’ those scholars who write in it, and of ‘all’ the articles published that used those key words —something that bibliometry and scientometrics allow doing with a few more clicks (see figure 1 and section 4 for examples). The same would be true if we wished to know what is this strange university called ‘Paris School of Management’: its profile will be given by the list of its academics. So there is no real di!erence in searching the identity of a person, a place, an institution, an event and so on. In all cases, the empirical and cognitive operation is the same. By circulating in such a way from the actor to the network and back, we are not changing levels but simply stopping momentarily at a point, the actor, before moving on to the attributes that define them. It is because there is no jump to another level that ANT defines as ‘flat’ the connections thus designed by its method of circulation through data sets (Callon & Latour, 1981; Latour, 2005). This new experience of moving easily through profiles already makes clear that what is meant by 2-LS and 1-LS social theories does not refer to di!erent domains

1- How digitally available profiles modify the element/ aggregate relations

of di!erently conceived wholes (section 3). The remaining sections o!er visual descriptions of ‘wholes’ that are much smaller than their parts (section 4) and suggest another type of navigation through data sets than the one associated with the idea of modelling (section 5).

4

of reality but to di!erent ways of navigating through data sets (Franzosi, 2004; Michel et al. 2011). ‘Specific’ and ‘general’, ‘individual’ and ‘collective’, ‘actor’ and ‘system’ are not essential realities but provisional terms that depend rather on the ease with which it is possible to navigate through profiles and to envelop them inside their names. The more cumbersome the navigation is, the stronger will be the temptation to handle them through the 2-LS. As long as it is di"cult to reach the list of all the articles of a subfield such as ‘super majority voting’, one will be tempted to define it generally as ‘a whole’ —the very notion of ‘paradigm’ does just that (see below)— of which the individual professor named ‘Hervé C.’ is just a ‘participant’. It is the same thing if there is no good web site listing all the academics in this university called ‘Paris School of Management’. Then, one will be tempted to say that there is a generally defined entity —for instance a ‘corporate body’— whose proper name is ‘Paris School of Management’, which exists in relative independence from all the actors that define its envelope. This is where the two-level argument begins to take hold: one for the parts, another for the whole. It will seem irresistible to argue that to define general features, one should look at the level of structures; if one wishes to look at specificity, go to the level of individuals. But in e!ect, this distribution of roles between levels is a consequence of the type of technology used for navigating inside datasets.

Fig. 1 Detail of the ‘profile’ of the keyword ‘self- organization’. The network is built using as nodes all keywords, authors, references and addresses of the articles which use the keyword ‘self-organization’ in the Web of Science© between 2006 and 2010. The size of the nodes and labels is proportional to the number of articles in which an author, institution, reference or keyword appears. Links between two nodes are created whenever these two entities appear in the same article. Weights are attributed to these links depending on the frequency of these co-appearance. Node spatialization is performed using Gephi’s ForceAtlas 2 algorithm (Jacomy, M., Heymann, S., Venturini, T., & Bastian, M. (forthcoming). ForceAtlas2 , a graph layout algorithm for handy network visualization). In this approach, links are interpreted as springs, and nodes which are strongly linked tend to appear close to each other. The node corresponding to self-organiza- tion has been deleted to improve readability as it was connected to all nodes in the graph.

5

Fig. 2 A typical screen experience with the aggregates on top, the statistics on the right hand side and the individual blogs on the bottom left with highlighted words. It is this superposition that renders synoptically coherent the two end points of so many social theories that, we claim, is the experience that should provide the occasion to rethink Tarde’s ancient argument that the two end points are an artifact of the ways data are handled.

The best proof that those two levels do not correspond to any real ontological domains is that they begin to disappear, to be literally redistributed, every time one modifies or enhances the quality of access to the datasets, thereby allowing the observer to define any actor by its network and vice versa. This is exactly what the striking extension of digital tools is doing to the very notions of ‘individual’ and ‘wholes’. The experience (more and more common nowadays) of navigating on a screen from elements to aggregates may lead researchers to grant less importance to those two provisional end points. Instead of having to choose and thus to jump from individuals to wholes, from micro to macro, you occupy all sorts of other positions, constantly rearranging the way profiles are interconnected and overlapping. This is what has been well recognized not only by ANT, but also by scholars working with network analysis (White, 2008). Of course, we do not claim that digitally available profiles are so complete and so quickly accessible that they have dissolved the two levels, but that they have already redistributed them enough to o!er an excellent occasion to see that those levels are not the only obvious and natural way to handle the navigation through datasets about entities taken severally. To sum up this first section, we will claim that one is tempted to treat an entity di!erently from its context only because of a lack of access to the list of attributes that make up that entity. At the very least, the digitally available profiles open new questions for social theory that don’t have to be framed through the individual/ collective standpoint.

6

After having provided a flavour of our overall argument, we may now move to its more substantial and technical aspects. In 2-LS social theory, the most current approach to handling the distinction between macro-structures and micro- interactions consists in establishing a first level of individual entities, then adding to them a few rules of interaction, in order to observe whether the dynamics of interaction lead to a second level, that of aggregation, which has generated enough new properties to deserve to be called a ‘structure’, that is, another entity for which it is possible to say that ‘it is more than the sum of its parts’. Such is the way in which most models of collective behaviour are framed, no matter if they deal with atoms, gas, molecules, insects, swarms, markets, crowds, States, artificial lives, etc. (for examples, see Moussaid et al. 2009). The explanatory power and the sheer beauty of those models are tied to such a mini-max: the longer enduring structure with the lighter sets of rules. It is important to underline here that since the 17th century this paradigm has been set in opposition to its apparent alternative that starts with a sui generis entity —for instance a body, an organ, a superorganism, an anthill, a beehive, a society, a State, etc.— in order, then, to define its individual ‘parts’ as endowed with ‘roles’ and ‘functions’. Such an alternative is often called ‘holistic’ or ‘organicist’ (Weick, 1995). Although the two views usually di!er in the political consequences one can draw from them (Hirshmann, 1977), for us they are just two di!erent ways of handling the social phenomenon by using the same 2-LS standpoint since both rely, as we shall see, on much the same data collection techniques. Their main di!erence is in the time order in which they list the three concepts: from the micro to the macro for the first, from the macro to the micro for the second. What the latter takes at its starting point, the former takes as its future horizon. Let us take the former as our starting point since it is nowadays the most frequently used. To define the first level, the model builder has to imagine individual atoms limited to as few traits as possible; then to devise rules of interactions between those atomistic entities —again as simple as possible—; then to observe how those interactions, after many fluctuations, stabilize enough to deserve the name of a structure; and then to check if this structure is su"ciently robust to be used as substitute for the ‘wholes’ that their adversaries —the holistic or organicist theorists— claim to exist before or above the ‘parts’ (Wilson, 1975). These are the research strategies that are followed, for example, when, against the arguments of the anthill as a super-organism, ethologists succeed in obtaining the highly complex geometry of the ant nest with only a few rules of interaction between blind ants considered as interchangeable actors (Pasteels & Deneubourg, 1987; Moussaid et al. 2009; Kuong et al. 2011). But it is also the fascinating beauty of market models when, without the push of any ‘invisible hand’, the sheer interaction of selfish but calculating individuals succeeds in settling on an allocation of resources more optimal than those any State would generate. Or when ‘selfish genes’ are said to provide a coordination of body parts that no notion of an organ superior to the cells could ever dictate (Kupiec & Sonigo, 2000). Or again, what happens when sociologists manage to map out the segregation patterns of city dwellings with only two rules of attraction and repulsion among individual neighbors (Schelling 1971, Grauwin et al. 2009), and so on and so forth. This approach can succeed in reproducing and predicting the dynamics of some collective phenomena when the individuals’ behaviour can be satisfactorily described with a few parameters and fixed rules. For example, the ‘ola’ can be explained by characterizing the reactions of humans in a football stadium by only three states (excitable, active and passive) (Farkas, 2002). By calculating the transition probabilities between these states, scientists might be able to predict the size, form, velocity and stability of the emergent «#ola#», and even how the probability of occurrence of a wave depends on the number of initiators (triggering an «#ola#» requires a critical mass of initiators). When only a handful of parameters su"ce to simulate the system’s dynamics, it makes sense to treat individuals as atoms (Barabasi, 2003; Cho, 2009). This has proved useful to understanding some features of queues, tra"c jams, panics, etc.

2- How to trace overlapping ‘monads’

7

However, humans do not spend most of their time in queues, in tra"c jams or in stampedes… To limit the grasp of quantitative social theory to just those few behaviours would be a pity. The problem with the ‘atomistic’ approach is that it has proved incapable of understanding more complex collective dynamics. Many reasons have been put forward to explain this: for example, human behaviour cannot generally be captured with context-independent rules —which are needed to write an algorithm (Flyvjberg 2001). But the real reason, for us, is that the very project starts from a restricted vision of the social: why assume that there first exist simple individual agents, then interactions, then complex structure —or the opposite? Why distinguish successive moments —in whatever order? Such apportioning is especially strange when it is not only possible but also easy to gather a lot of information on each of individual entity taken severally so as to draw its extended profiles. If the complexity of individual agents can be observed and handled, why would it be necessary, first, to strip individual entities of all their attributes? Why should models proceed according to the usual way by adding simple rules of interactions between atoms now deprived of the network of attributes they possessed before? And why should complexity be obtained, in a next step, as a provisional whole since it was there at the beginning? What might have appeared common sense within a di!erent technology of data collection might cease to be so now that profiles are so conveniently available. In 1-LS, by contrast, agents cannot be said, strictly speaking, to ‘interact’ with one another: they are one another, or, better, they own one another to begin with, since every item listed to define one entity might also be an item in the list defining another agent (Tarde, 1903; 1895/1999). In other words, association is not what happens after individuals have been defined with few properties, but what characterize entities in the first place (Dewey, 1927). It is even possible to argue that the very notion of ‘interaction’ as an occasional encounter among separated agents is a consequence of limited information on the attributes defining the individuals (Latour, 2010). But is there an alternative to the common sense version that distinguishes atoms, interactions and wholes as successive sequences (whatever the order and the timing)? An alternative that should not oblige the inquirer to change gears from the micro to the macro levels as is required by the 2-LS, but remains fully continuous or, as is claimed by ANT, fully ‘flat’. It appears to us that one alternative to the atom-interaction-structure is what has been called by Gabriel Tarde, in reference to Leibniz, a ‘monad’ (Tarde, 1895/1999). A monad is not a part of a whole, but a point of view on all the other entities taken severally and not as a totality. Although historians of philosophy still dispute what a monad was for Leibniz and although there exist many confusing definitions of what it was for Tarde (Milet, 1970; Candea, 2010), our claim is that the definition of this admittedly exotic notion may be rendered fully operational provided one uses the illustration o!ered by just the type of navigation through digital profiles we have sketched above. This argument relies on the practice of slowly learning about what an entity ‘is’ by adding more and more items to its profile At first the entity is just a dot (in our example it is nothing but a proper name ‘Hervé C.’ a clickable entry on a computer screen) but then it ‘fills in’ with more and more elements that specify it more and more until the observer considers that he or she knows enough and begins to take the name of the entity for the entire list. What has happened? In e!ect, we have drawn a monad, that is, a highly specific point of view —this or that entity— on all the other entities present in the dataset. The point of this navigation is that it does not start with substitutable individuals —as in the 2-LS— but individualizes an entity by deploying its attributes. The farther the list of items extends, the more precise becomes the viewpoint of this individual monad. It begins as a dot, a spot, and it ends (provisionally) as a monad with an interior encapsulated into an envelope. Were the inquiry to continue, the ‘whole world’, as Leibniz said, would be ‘grasped’ or ‘reflected’ through this idiosyncratic point of view. As we saw, the crucial interest of the notion of monad —even if its fancy

8

metaphysics is put aside— is that it is fully reversible, a feature that was impossible to render operational before the access to digital media. Each of the attributes used in order to define the entity is itself modified by becoming the attribute of this entity. In our example, whereas being ‘professor in Paris School of Management’ specifies who is ‘Hervé C.’, when we shift, with a few clicks, to ‘Paris School of Management’ we realize that it has become a slightly di!erent academic body now that it is able to attract a ‘mathematician’ and a ‘well known economist from abroad’ to be its ‘dean of academic a!airs’, which was not the case before. ‘Paris School of Management’, too, is individualized and in no way can it be taken for an element of the ‘context’ inside which ‘Hervé C.’ should be ‘framed’. In other words, ‘Paris School of Management’ too is a monad depending on how one navigates through its profile. What is so refreshing with the new habit of circulation is that they never end up tracing an entity as “part of a whole” since there is never any whole. The reason is that with 1-LS there are, strictly speaking, no individual atoms (profiles are fully deployed through their attributes), nor aggregates (each attribute is nothing but the list of actors making it up). The experience of navigating through profiles available on digital platforms is such that when you move from one entity —the substance— to its network —the attributes— you don’t go from the particular to the general, but from particular to more particulars. In other words, the notion of a ‘context’ might be as much an artifact of navigational tools as is the notion of an ‘individual’ (Hagerstrand, 1953; Garfinkel, 2002; Latour, 2005). Extend the list of items, smooth the navigation, visualize correctly the ‘interior’ of each monad, and you might not need the atom- interaction-structure or the actor-system apportionment at all. You will move from monads to monads without ever leaving the solid ground of particulars and yet you will never encounter atomistic individuals, except at the first click, when you begin to inquire about one item and get only an empty dot. By now, our working hypothesis should be clear: it might be feasible to move from particular to particular and yet to obtain along the way partial totalities without ever relying on any of the three sets of concepts that make up the 2-LS: there is no individual agent; they don’t interact; there is no whole superior to the parts. Such a radical conclusion is made at least plausible by the new datasets that allow entities to be individualized by the never-ending list of particulars that make them up. Such is what is meant by a monad, a point of view, or, more exactly, a type of navigation that composes an entity through other entities and, by doing so, particularizes all of them successively —‘all of them’ being an open ended list the size and precision of which always depend on more inquiries and never from the sudden irruption of a superior level accessible through a sudden shift in methods. In other words, datasets may be handled through two opposite navigational procedures, one that is based on a series of leaps of aggregation (from atoms to interaction to structure —and back), and the other one, the monadological principle. Introduced in social theory by Tarde through literary means and then abandoned because of the lack of empirical handles, this principle can be given a new career through the newly available techniques of digital navigation and visualization (Candea, 2010). In summing up this second section, it is important to stress that we are well aware that such an alternative definition remains highly sensitive to the quality and quantity of information available as well as to the visualization techniques at our disposal. Remember that our argument is strictly limited to the search process through data sets and that we don’t consider how those attributes are gathered from ‘real life’. We recognize that tracing monads will not be always feasible. For most entities, the profiling will be impossible for a number of reasons: a) our observation techniques are too rough to follow each entity individually —this will be the case with ants in an anthill, cells in an organ, human actors in a large survey; b) the entities are really interchangeable since there is no way, even with the most sophisticated tracking device, to detect di!erences among them —this will be the case with atoms in a gas (Jensen, 2001); c) even though it is possible to track their di!erences, most of the information has to be deleted or kept secret for ethical

9

reasons —this is most often the case with telephone calls, social networks, health files, etc.; d) in spite of their claim to transparency and equality, most present day databases are rife with inequalities of status and most entrench rather crude definitions of the social world. What we claim is simply that every time it is possible to use profiles, then the monadological principle will obtain. The reason why we insist so much on this feature is to follow another of Tarde’s insights that a 1-LS social theory should in no way be limited to human actors. Every time inquirers have succeeded, through clever research strategy, to trace individualizing profiles of agents — baboons (Strum & Fedigan, 20

Collepals.com Plagiarism Free Papers

Are you looking for custom essay writing service or even dissertation writing services? Just request for our write my paper service, and we'll match you with the best essay writer in your subject! With an exceptional team of professional academic experts in a wide range of subjects, we can guarantee you an unrivaled quality of custom-written papers.

Get ZERO PLAGIARISM, HUMAN WRITTEN ESSAYS

Why Hire Collepals.com writers to do your paper?

Quality- We are experienced and have access to ample research materials.

We write plagiarism Free Content

Confidential- We never share or sell your personal information to third parties.

Support-Chat with us today! We are always waiting to answer all your questions.