Thursday 4 September 2014

Computational Social Science and Social Naturalism

In recent years it has become possible to computationally calculate relations between people, documents and words in terms of the analysis of communications and the modelling of social dynamics. The implications of this “computational social science” on naturalistic social inquiry are disputed. The way that utterances between individuals relate can be investigated: this might entail analysing the correspondences between the use of words, phrases and topics in different contexts, determining the relative probabilities of those topics or words occurring across contexts and making inferences about the related semantic content between different contexts as a result. This approach belongs to a tradition of studying information in terms of the probabilities of the occurrences of symbols which was instigated by Shannon, and which a number of authors have related to the processes of human communication (although Shannon himself excluded this possibility). As more communication practices are bound by technology, probabilistic analysis of communication presents remarkable results in affording the clustering of cognate areas through automatic analysis, making possible automatic topic identification, the clustering of social groups through communication patterns, automatic identification of learning needs, preferences, and so on. Consequently, economic opportunities present themselves through the emergence of recommendation services, direct marketing, learning support, social prediction, and so on.

In contrast to the statistical techniques which view communications as ‘bags of words’, purposeful and meaningful utterances constitute lived experience where communications implicate and manipulate social positions between agents: expectations, intentions, meanings, ambitions, experiences and power struggles contribute to the constitution of both social institutions and individual experience. Unlike relations, positions are inhabited by agents. Despite the success of the algorithmic analysis of communication relations, most communicating occurs away from the internet. What is the possibility of a computational analysis of the inter-human positioning of meanings, expectations and power struggles? What is its connection to the computational analysis of relation between people, words and documents? What import does this bear for a naturalistic approach to the social sciences?
Objects of Analysis and Analytical Objects

A computational analysis examines data produced through social interaction (usually from the internet). Internet-based interactions are declarations of facts, feelings, opinions, states of affairs made by agents for the purpose of positioning between agents. An analysis of the words that constitute declarations is also a declaration – also for the purpose of positioning. The present analysis here is also a declarative contribution to discourse and the purpose here is positional. How might an understanding of the social dynamics of what might be considered ‘codes of communication’ arise through what can only be a participatory process in the reproduction and transformation of those codes of communication?

We might begin by making a distinction between experience, utterance and analysis. Searle has recently suggested one way of doing this, making a distinction presenting “ontological subjectivity” and “epistemological subjectivity”. For Searle, consciousness belongs to the category of the “epistemological subjective”: these are the as-yet uncodified transcendental constructs relating to Husserls’s “horizon of co-givenness”, Freud’s primary process or Luhmann’s ‘psychic system’. By constrast Searle presents the “ontologically subjective” as a way of characterising social institutions, documents, artefacts, practices, technologies and social roles. Searle argues that ontologically subjective objects are brought into being by a particular kind of speech act called a ‘status function’. In arguing for this, he argues for a ‘collective intentionality’ which upholds status functions and thus serves to constitute social reality. Searle’s collective intentionality can be compared to a coordination of expectations within a system of communication as is described by Luhmann.  Status functions are themselves utterances – among their basic forms is the statement “X counts as Y in C”, but this is one of many forms of utterance.  In this way, Searle’s categories serve as a way of distinguishing between objects of analysis – spreadsheets,  social network graphs, economic forecasts and so on – about which status functions as expressions of collective intentionality and positioning will be made (“X counts as Y in C”), and the epistemically subjective realm of consciousness within which coordinations of utterances emerge.

Considering the difference between relations and positions in this light, it is possible to make the status function “this relation is a position” – indeed, this is a common status function among those who would wish to assert the legitimacy of simple social network analysis as isomorphic to social ordering. Such confusion between relation and position presents an opportunity to explore related status functions concerning the legitimacy of data analytics in the first place and the particular algorithms concerned. Since positions are inhabited by agents, we might expect the false assertion of positions to lead to difficulties in establishing the appropriate collective intentionality required to uphold the status function. However, assertions of relations as positions sometimes do become accepted for periods of time, supported by acceptance of other status functions (of the deontic power of those making the declarations, for example), only for them to eventually be critiqued and overturned.  Indeed, radical changes in positioning involve the dramatic overturning of status functions relating to social role or position: revolution is the name we give to the overturning 
of the status function “I am the king”; Western powers are currently engaged in an attempt to overturn the status function "This is a caliphate".

The Analysis of collective intentionality

Since status functions are upheld by collective intentionality, status functions are themselves positional (they implicate power, ambition, expectation and meaning). It is reasonable to ask whether algorithmic analysis which exposes position rather than relation is possible. Such an analysis depends on postulating dynamics of multi-dimensional communications of everyday life. Techniques for analysing relation rely on Shannon’s theories about information, entropy and redundancy which can be applied to the distribution of symbols of communication in a large system like the internet. These techniques applied the analysis of relation appear to produce striking and revealing results. However, the experience of encountering such an analysis is effectively one of being subject to a status function relating to the analysis by various means of other status functions contained in internet communications in which many agents (possibly including the viewers of the analysis) participate. In short, to be ‘convinced’ by a relational analysis is to be positioned by those making the status declaration about the analysis. This then raises the question as whether it is possible to produce a computational analysis where the positioning itself might be revealed, thus revealing the deeper dynamics of collective intentionality.

Before addressing this question, there is an obvious concern: is a status function concerning a positional analysis any different from a status function concerning a relational analysis? The problem here becomes one of assessing the impact of a status function of any analysis. Status functions of all kinds are positional – they implicate what Searle calls the ‘deontic power’ of the person making the declaration – however sophisticated, power, ambition, meaning and expectation all play a role. They further implicate status functions concerning the method (algorithms) and underlying epistemology of any analysis that is pursued. Since methods and epistemologies will entail transcendental  components which are metaphysical (and therefore unprovable) methods, and analytical objects will always be vulnerable to critique. So why bother attempting to develop an analysis of positioning between agents?


Moving beyond a relational algorithmic analysis to one that attempts to capture the dynamics of collective intentionality presents an opportunity to explore different empirical metrics whose values may be compared to logical structures which express different social theories. If a connection between logical structure and empirical measure can be established, then a naturalistic inquiry emerges as a possibility in the social sciences whereby theorized social ordering can be compared to measured data. This would address a problem in the social sciences whereby research practice frequently exhibits a kind of ‘naturalistic gap’ as theories are used to design interventions, but where theory-practice gaps result in changes to practice at the expense of theoretical critique  (resulting in pathological positioning).  In other words, richer analysis – particularly the analysis of collective intentionality should be seen in the context of an empirical activity which stimulates disputation and analysis between statements of logic (theorized social structures) and status of empirical fact (empirically measured structures). Might an analysis of positional data present an opportunity to exploit data analytics with regard to social naturalism?

The Logical characterisation of social structure and the possibility of naturalism

Social analysis – particularly as it is undertaken in economics – tends towards quantification of social structure through econometric modelling (Hodgson, 1988; Lawson, 1999). Social network analysis has provided new means of representing social structure as relational entities. Both econometric and social network analysis presents logical and empirical aspects. The logic of econometrics has been heavily critiqued in recent years as not only divorced from reality, but responsible for economic pathology whereby econometric models which have little predictive or explanatory power are nevertheless forced on the population (Lawson, 1999). A similar story of forcing social network idealisations on a population have emerged in the manipulation of user responses to social software and gamification. In such situations, the theories that sit beneath the models are not critiqued, whilst economic policy attempts to force practice so that its measurement fits the econometric models used for analysis. In other words, there is a ‘naturalistic gap’ between prescribing policy interventions without theoretical development or refinement.

Social network analyses, however, do provide a glimpse of the logic of a social structure: indeed, the graphs of positional social network analysis are essentially logical, not empirical. In representing relations between nodes and arcs, the logical structure of the diagrams bears comparison with what Kauffman (1995) calls ‘arrow epistemology’ alluding to Category Theory (Mac Lane, 1972; Goldblatt, 1982). In arguing for such logical presentation of structures as a network of arrows, Kauffman argues that “it enables us to draw connections with the kind of diagrammatics that occurs in artistic, linguistic, physical and philosophical contexts.” (Kauffman, 1995, p 38). Extending this to the category-theoretical constructs of objects, limits, exponents, push-outs, pull-backs and sub-object classifiers, Badiou (2006), Meillassoux (2008)  and others have argued that logical mathematical representations of social structures may present new directions for naturalistic investigation. In the present context, the connection between such logical structures and empirically-derived networks pertaining not only to relation but also to position are of interest. This is to address the naturalistic gap between theory and practice through the comparison of two status function: one which asserts the status of the empirical object (the social network analysis of relation or position); and the other which asserts the status of the logical description. A naturalistic approach may then explore the differences in ordering between the two, which in turn represents the gap between theory (as represented by the logical) and practice (as represented by the empirical).

This may be described at a simple level. Any status function, if it is to be upheld within its context (in other words, if it is maintained within a particular code of communication), will exist by virtue of the dynamics of other communications. Kauffman has suggested how certain ‘knot’ topologies may be mutually reinforcing. Using the example of the Trefoil knot, it is possible to see where a particular topology of communication may serve to maintain those communications owing to the mutual constraint that each communication bears on the others. Noting this, a logical representation can find fuller description in Category theory by articulating the limits bearing upon each aspect of the communication dynamics, and the way in which each limit relates to each other limit.
Within the empirical technique for identifying positions described above, the logical representation of constraint identified in category theoretical limits becomes translated into the empirical measurement of mutual redundancy. The empirical measurement of positions rests on the identification of redundant expectations which can be analysed through communications. A theory may express the view that certain status functions are upheld within certain dynamics of communication. Measurement of redundancies of communication can confirm or deny this, thus requiring changes to the representation of logical structures and deeper analysis of empirical data.

In normal science, reproducible experiment and regular successions of events (as described by Hume) may itself be reinterpreted as a probabilistic exercise in the identification of redundant expectations (Meillassoux). By this description, theory-building emerges from the identification of mutual redundancy produced through empirical practice. A positional data analysis coupled with the articulation of theory as logical structure similarly is a process of identifying mutual redundancies of expectation between the theory and empirical results. In particular, category theoretical articulations of social order articulate limits at different points in a structure which empirically correspond to the constraints of redundancies. The correlation between limits and redundancies only works in a positional analysis, since a relational analysis simply considers connections between nodes rather than constraints bearing upon them.

References
Badiou, A (2006) Logics of Worlds
Bhaskar, R (1979) The Possibility of Naturalism
Goldblatt, R (1982) Topoi: The categorial analysis of logic
Hodgson, G (1988) Economics and Institutions
Kauffmann, L.H. (1995) Knots and Applications
Lawson, T (1999) Economics and Reality
Luhmann, N (1980) Social Systems
Mac Lane, S (1972) Categories for the working Mathematician
Meillasoux, Q (2008) After Finitude
Searle, J (2007) Making the Social World
Shannon, C; Weaver, W (1952) The Mathematical Theory of Communication

No comments: