AJIS
Special Issue
December 1001
A THEORETICAL BASIS FOR IS? THE CONTRIBUTION OF ANT
Jim Underwood
University of Technology
Sydney, Australia
jim@it.uts.edu.au
ABSTRACT
Representation is a key issue of IS design and operation that is often ignored. Actor-network theory (ANT), a semiotic
theory of stakeholders, provides a way of dealing with representation. Combining aspects of ANT and Foucault's
discourse theory allows us to include concepts as actors and promises a flexible and durable foundation for IS practice,
but ANT itself indicates that the search for a purely theoretical foundation for IS is misguided.
INTRODUCTION
An information system is a system - an interaction of people, devices and procedures; a technology - for
providing information: news, facts which are useful and previously unknown, experiences that change our view
of the world (Avison and Fitzgerald, 1995; Bou1ding, 1961). Such systems usually operate by acquiring
information form other sources such as people, documents,sensing devices, traces of events, and representing it
as data (possibly highly structured) and making these data available to potential users (interpreters) by either
sending messages or storing the data for later retrieval. Many introductory texts (e.g. Oz, 2000) deal with the
storage and transmission of data, the uses to which they might be put and sometimes the way in which data may
be collected. Representation and interpretation are seldom treated comprehensively in information systems (IS)
texts, except technically as coding and decoding. In this paper I will argue that representation is a key issue in IS
design and use, and that semiotic theories provide an important foundation for the study of IS. I will further
argue that actor-network theory (ANT) is a semiotic approach that is particularly suited to understanding
information systems.
REPRESENTATION IN INFORMATION SYSTEMS DEVELOPMENT
In traditional life-cycle approaches systems development takes place through a number of fairly clearly
differentiated stages. Transitions between successive stages are marked by "sign-offs" where all stakeholders
agree that the work so far has been completed satisfactorily. What is being agreed upon is some sort of
document, usually a mixture of text and diagrams, which is supposed to represent in a certain level of detail
(depending on the particular development phase) the system that will eventually be installed. From a semiotic
point of view these text elements are signs that represent parts or features of the proposed system. There is
considerable anecdotal evidence that different stakeholders have widely varying understandings of what they
have agreed to in signing-off. We have also noticed, at least in student projects, that developers see the
documentation as representing existing business rules and development work already completed, and that their
idea, if any, of the final system, is imagined privately and separately from the agreed documents. Prototyping
has come to be used as a supplement, or in some cases a replacement, for life-cycle approaches. With
prototyping, representation of the future system is in the form of hardware and software artefacts as well as
documents. Agreement on the design is negotiated continuously rather than at a number of discrete sign-off
points. Concerns have been raised over the extent to which prototypes can represent issues of system
management, maintenance and performance under normaloperating conditions (Davis, 1984). From our point of
view prototypes are also signs, and raise similar issuesof understanding to documents.
Beyond prototyping there is evolutionary development. Here requirements construction, implementation and
operation occur simultaneously. An embryonic system is built and put into use. With use some features are
found to be helpful and are enhanced, others are ignoredand gradually disappear from the system (although they
often remain buried in unused code). As with prototyping, the speed of evolutionary change is facilitated by
rapid software development environments, particularly where changes can be easily made by end users. In a
sense, evolution of information systems is what always happens over time anyway. Adopting evolutionary
development as a deliberate policy is a matter of realising that we will never be able to faithfully represent a
future system, of removing controls that attempt to make the future system represent the original design, and
allowing the evolutionary process to speed up.
We see design activities as social activities, exercisesin communication. Where there is a designated "designer"
their role involves at least two communication relationships- assuring the users that the system-to-be will make
their lives better and guiding the system builders (often programmers) in their construction work. When
prototyping is used extensively the designer might take a less dominant role, mainly facilitating communication
between the users and the builders. Successful information systems design would seem to depend on
establishing mutual understanding either between users and designers and designers and builders or between
users and builders - and preferably among all three groups. An appreciation of how information systems are
constructed through representation and shared understanding is critical for anyone attempting to manage the
86
AJIS
Special Issue
December 2002
systems development process. This is so even with an evolutionary approach: design, building and use occur
simultaneously, but even within one user there are conceptual shifts as they move among these three modes of
system construction.
Peirce's Semiotic Model
For these reasons, an important theoretical foundation for research and teaching in IS is semiotics (Noth, 1990),
the theory of signs and meaning. Fundamentally, semiotics is simply about representation, about how signs
stand for things in the world of objects. One popular semiotic theory is Peirce's triadic model.
A sign, or representamen, is something, which stands to somebody for something in some respect or
capacity. It addresses somebody, that is, creates in the mind of that person an equivalent sign, or perhaps a
more developed sign. That sign which it creates I call the interpretant of the first sign. The sign stands for
something, its object. It stands for that object, not in all respects, but in reference to a sort of idea, which I
have sometimes called the ground of representation. (peirce Collected Papers of C.S. Peirce (2.228) quoted
in Spinks, 1991)
In the case of IS development some of the things represented, systems-to-be, do not yet exist and we might
expect to find that the same sign could have different meanings for different stakeholders. Hence IS
development is about networks of meaning - the existing network of the organisation, the project development
network where the information system becomes an object of shared understanding, and the changed
organisation network which will include the completed information system. Documents, models and discussions
are the links, which carry, or fail to carry, meaning in these networks.
Traditional descriptions of information systems rely not only on natural language but also on diagrams
(structured or unstructured) and restricted or formal languages (such as pseudocode). Some authors attempt to
use semiotics as a basis for yet another specialised IS specification language (Stamper, 1992), or use semiotic
concepts to classify user communications in existing systems (Clarke, 1998). Our image of information systems
development is strongly constructivist. This means that the world, in this case the supposed reality underlying
the system design - the business rules and the technical possibilities, is created by the participation of the
participants (users, designers, programmers, managers) (Watzlawick, 1984). Although some methodologies give
the impression that system design follows logically from predetermined requirements, it is fairly obvious that
information systems are in fact designed and built by people, almost always working with others, and
sometimes with computers, in a social setting. But we also believe that the requirements themselves are socially
constructed during the design process. In this situation aspiring IS developers do not need a semi-formal
semiotics based methodology, nor a theoretical model of "best practice" organisational communication, but a
knowledge, sensitivity and attitude which will keep them continuously aware of representation issues during the
development process.
One representation issue, which (as we have mentioned) is of particular significance in design and development,
is that of signs referring to future objects. Peirce provides some light on this issue when he distinguishes objects
as immediate, dynamic and real. The immediate object is that which takes part directly in the sign occurrence,
and is in a sense created by the act of interpretation (semiosis). The dynamic object has an existence
independently of the particular sign occurrence and to some extent can be thought of as co-producing the sign. If
there is anything real that "sufficiently corresponds" to the immediate object then this can be called the real
object (Oehler, 1987, pIO). The introduction of an immediate object avoids or at least defers a difficult semantic
question: can a sign refer to an object that does not exist (such as a unicorn) or which is nonsensical (such as a
square circle)? For Peirce anything referred to in a sign is meaningful, because by definition it means in some
respect something to somebody (perhaps as an example of impossibility); the sign itself validates the immediate
object. But can these unreal objects also be dynamic objects? We can regard them as such if the same object
seems to take part in a variety of signs. A unicorn can represent good fortune, we can say "a unicorn is a
mammal" and many children can draw one. It is reasonable to believe that these cases all refer to the same
(dynamic) object.
The idea of a dynamic object is clarified when signs are used in a design process. Suppose we have a data-flow
diagram of part of an existing payroll system. The data-flow diagram and the phrase "add new employee" are
two sign vehicles referring to the same dynamic, and presumably real, object. In another context the data-flow
diagram may signify a quite different object: if someone is photocopying the system documentation this
diagram could indicate the middle page of the documentation. With an existing system we assume that the
objects are to some extent already established; when designing a new system the objects are gradually brought
into being. Suppose now that we begin our design by drawing boxes representing subsystems of the payroll
system. We label one of these "add new employee". By this action we create an immediate object, which we
imagine as a process, to which the sign refers. As we think about this object, mention it in reports, discuss it
with others and perhaps ask someone else to draw a data-flow diagram for it, the immediate object gradually
achieves the status of a dynamic object, existing independently of any particular sign occurrence. If we spend a
large amount of money on hardware, software and training and persuade several people to change their payroll
processing behaviour we might eventually feel confident in calling the object real.
87
AJIS
Special Issue
December 2002
The interpretant is what the sign means, or how the sign works. It is the "respect in which" the sign vehicle
stands for the object. But what exactly is it? For Peirce the interpretant is that which the sign produces in the
interpreter; it may be a feeling, an action or another sign. Since Peirce was an empiricist and pragmatist the
ultimate meaning of the sign must be its observable effect, so the feeling must be eventually expressed by
speech, action or symptom; in the end all interpretants are available as vehicles for further signs. As with
objects, there are various types of interpretants. The immediate interpretant is the actual understanding of the
sign (perhaps a process), the dynamic interpretant is the reaction provoked and the final interpretant is the
"proper" reaction to which all interpreters converge (Oehler, 1987, p6). The combination of sign vehicle, final
interpretant and dynamic object gives a general sign that is understood through objective consensus, not
individual thought. This is important for Peirce because it removes signs from the unverifiable mental space of
the interpreter. The disadvantage is that signs can only be understood in terms of other signs and in theory this
leads to an infinite regress; but in practice we always manage to stop.
Until the object becomes real the assumed existence of a dynamic sign shared by a number of sign occurrences
depends on the existence of a stable mental image. Agreement among sign users depends on two links: the
object in question must invoke the same mental image in all users, and they must use the same sign to indicate
that image. Because the common element, the mental image, is private and cannot be accessed except through
other signs, it is impossible to tell whether agreement has been reached or how it has broken down. Management
and workers may agree that workers should be "more productive" but management may have an image of more
output per dollar while the workers imagine more output per person (and more pay). If, referring to the same
events, managers say "we have down-sized by 10%" and workers say "500 people were sacked" management
may be using a euphemism (different sign, same object and image) or may have a quite different mental image
of the events.
In practice, although our mental images related to a sign differ and although any clarification must be based on
equally ambiguous signs, we eventually build a network of connected signs that we are fairly confident others
understand. That this confidence could be grossly misplaced is apparent when we discuss "red" with a colour-
blind person, or are told by a teenager that certain behaviour is "sick". Nevertheless, such confidence is exactly
what keeps communication going.
Different Ways of Using Signs (Morris)
The pragmatic semiotics of Peirce and James, which believes "that, to develop a thought's meaning, we need
only determine what conduct it is fitted to produce: that conduct is for us its sole significance" (James, 1991,
p23), is taken to its logical conclusion by Charles Morris. Morris has been credited with defining the terms
syntax, semantics and pragmatics with reference to signs (posner, 1987, p25). His own theory of signs is based
on behavioural psychology. For Morris all behaviour is goal-seeking: the organism selects a desirable object
(orientation), performs appropriate actions (manipulation) and fulfils its desire (consummation). Of course while
this is going on many other organisms are doing the same thing so it is easy to become distracted or completely
diverted; and organisms can have very obscure strategies where even they find it difficult to distinguish ends
from means. The goal-seeking model can, however, still provide the basis for a definition of sign.
Thus a sign is a preparatory stimulus that, in the absence of an impulse-satisfying object, causes in an
organism a sequence of responses of the same type that would be caused by the object itself. (posner,
1987,p28)
Here Morris was probably thinking of something like Pavlov's bell and he would most likely expect his
definition to expand to include more complex situations. If we imagine a specification as a sign for an
information system then this defmition could work in some cases. A customer might, under strict conditions,
pay for the specification as if it were the system and a court of law might (again only in some circumstances)
accept the specification as signifying the system. But an end user could not actually use the specification to do
their work and a software engineer faced with the specification might be tempted to build the system, something
they wouldn't do if the system already existed. In the same place, however, Posner compares the sign to the
behaviour sequence itself. The sign vehicle alerts the organism to the desired object, which will satisfy the
impulse, while the interpretant is the impulse itself. In this case we can say that the meaning of the specification
to the engineer (or the client) is the satisfaction they anticipate when using the completed system. Thus we
might respond to an advertisement (the sign vehicle) for a new operating system by imagining that obtaining it
will lead to an experience (the object) that will give us greater reliability or will make our friends think we are
top computer scientists (the interpretant or imagined consummation).
Corresponding to the behavioural sequence, Morris defined three dimensions of signification. In the orientation
phase it is important to assess the facts of the situation and designative signs are most helpful; in the
manipulative phase we need rules or guidelines about what to do through prescriptive signs; and when
considering consummation we need to evaluate the desired object using appraisive signs. We would normally,
for instance, use a prescriptive sign if we wished someone else to do something for us ("please back up the disks
regularly"); but in some circumstances it may be more effective to use designative ("did you hear how Fred lost
all his data last week?") or appraisive ("our database is our most valuable resource") signs for manipulative
purposes (posner, 1987, p4l).
88
AJIS
Special Issue
December 2002
REPRESENTATION IN ACTOR-NETWORK THEORY
From the above we can see that meaningful signs, particularly for yet to be constructed objects, depend on the
development of shared meaning, but that ambiguity of meaning can be used for a form of manipulation. Systems
development methodologies often assume that all parties to the development share goals or, in the case of Soft
Systems methodology, that a common direction is established at an early stage (Checkland, 1972). If various
participants continue to have different interests then the likelihood of shared meaning developing is greatly
reduced. Stakeholder theory is a widely accepted model for describing conflicting interests, but gives us little
insight into the expression, communication and manipulation of these interests. Actor-Network Theory (ANT)
provides a semiotic approach to stakeholder analysis.
Actors, Networks and Meaning
Actor-network theory (ANT) was developed as a method in science and technology studies. It attempts to
anchor meaning to the desires of people and the constraints of nature (Latour, 1993, p64) by granting status to
hybrids or quasi-objects, which are heterogeneous networks of phenomena. The hole in the ozone layer, for
instance, "links the most esoteric sciences and the most sordid politics, the most distant sky and some factory in
the Lyon suburbs, dangers on a global scale and the impending local elections or the next board meeting"
(Latour, 1993, pl ), Similarly, a management information system is "an integrated user-machine system for
providing information to support operations, management, and decision-making functions in an organization.
The system utilises computer hardware and software; manual procedures; models for analysis, planning, control
and decision-making; and a database" (Davis and Olsen, 1984, p6). By tracing the activity of these hybrid
networks we can see how the world is interpreted, reinterpreted and changed. These networks of actors are the
hybrids ("network", "hybrid" and "quasi-object" are interchangeable terms).
From the point of view of much traditional philosophy, confusing machines, people, ideas and organisations
would be regarded as a category error. ANT makes category error its major virtue. Latour (1993, pll) explains
this in the following diagram.
NATURE
o
CULTURE
o
PURIFICATION
TRANSLATION
Figure 1 - Purification and Translation
He says that the project of modernism attempted to divide the world of thought into the natural and the social, a
world of objects, forces and velocities, and a world of feelings, beauty and community. In one world statements
may be true or false, in the other right or wrong. Before we can discuss or even think of some phenomenon we
must decide on which side of the vertical divide it lies. This decision may vary over time, and may even divide
different aspects of what appears to be one being - at a particular time in history, human anatomy may have
been considered a natural phenomenon, while human thought may have been cultural. Purification ensures that
we talk about the right things with the right word, that we avoid such "errors" as anthropomorphism.
But below is the hidden world of translation. Practical people are building factories, musical instruments,
information systems and cloned sheep; artefacts in which the natural and the cultural are hopelessly mixed. The
modem project requires that we don't think of the two levels, the worlds of purification and translation, at the
same time. Otherwise the belief that thought can be clear, that ontology is possible, that our life can be
controlled by the right ideas, would dissolve. And because we cannot use our much refined tools from the world
of thought, the world of hybrids does run out of control and is filled with all sorts of monsters on which we
depend but which we don't understand. How can we bring meaning to these networks of translation?
89
AJIS Special Issue
December 2002
Scripts and Translation
ANT is a type of stakeholder analysis where the stakeholders may be other than the usual people and
organisations and the stakes may be reinterpreted as negotiations progress. Typical stakeholder phenomena such
as the formation of alliances are dealt with in detail by Callen (1986) and will not be covered here. We will
concentrate on the key concepts of ANT: actor, script and translation.
In IS development the obvious candidates for actors are the stakeholders such as owners, clients, operators and
regulators; and the potential components of the future system are hardware, software, people and processes. As
Callon comments, "... first ... the list of actors is not restricted to social entities; but second, and most important,
their identities and their wishes are all constantly negotiated during the process of translation" (Callon, 1986,
p227, p21). Scripts represent the interests of the actors. Notice that we have just uncovered another level of
ambiguity in "representation". A script may be as simple as "close the door" or a million line computer program.
The design of technical artefacts (quasi-objects) involves delegation of actions to various actors by inscribing.
This process is complicated with a collection of script-based activities such as conscription, prescription,
proscription and subscription, all of which may be ascribed to a particular situation by researchers (or project
managers) (Akrich and Latour, 1992, p259).
Technical artefacts such as database servers or application programs are judged by their actions; they make
performative statements. In systems development they will be conscripted to the purposes of other actors by
being inscribed with scripts in a language, which they (pragmatically) understand. This conversion of scripts
(perhaps into machine code) as they are inscribed from one actor to another is one of the several meanings of
translation in ANT. Translation can model the operation of IS components as well as their design, for example
when client requests are translated into keystrokes. It is widely accepted that failures in translation, both during
design and operation, are a common source of system "malfunction". ANT allows us to model this
mistranslation as possibly intentional (betrayal) and as an essential part of system design and operation, rather
than as an exogenous disturbance, which can be removed. In the pure world of modem thought, where humanity
and nature occupy separate domains, such a removal might be imaginable. In the messy world of modem
practice, the world of technology in general and information systems in particular, such a removal would render
translation impossible (Latour, 1993, p11).
Representing Concepts in ANT
Latour gives us a vivid picture of the way scripts move and change among the network of actors, but does not
address how the meaning of the scripts might change with the interests of various classes of actors. We have
proposed (Underwood, 2001) that, using Foucault's (1972) concept of discursive formation, we can see how
particular actors may translate scripts with reference to a "home discourse" such as accounting or computer
science. This translation takes place in the "world of thought" and is facilitated by the use ofleamed concepts
such as "efficiency" or "attribute". But in the "world of practice", where actors from different discursive
backgrounds must engage in at least minimal collaboration if any progress is to be made at all, these concepts
must be reinterpreted or betrayed to develop some level of shared meaning. As this process continues, some
type of local language or "hybrid discourse" will develop and actors may gradually detach from their home
discourse ("go native"). The ability to straddle these two worlds, shown in Figure 2, with relative comfort is a
critical skill for any system developer.
90