Download as a PDF

7 downloads 0 Views 140KB Size Report
In this section, we provide a high-level overview of what multi-agent systems are and some of the research issues in this field. 1 The Matrix, Warner Bros., 1999 ...
Wider perspectives on service-dominant logic Otago Forum 2 (2008) – Academic Papers  

Paper no: 11      

Martin Purvis University of Otago, New Zealand [email protected]

Andrew Long University of Otago, New Zealand [email protected]

       

Otago Forum 2: Academic Papers

Wider perspectives on service-dominant logic Abstract The contrasting marketing principles of service-dominant (S-D) logic and goodsdominant (G-D) logic are examined in this paper from the perspective of current research themes in information science. In particular, we believe that much of the distributed multi-agent work relates to the issues discussed in the service-dominant logic marketing literature, and we think that some multi-agent practices and results may be of interest to the service-dominant logic community. The paper begins with coverage of a fundamental divide that exists between two general modelling approaches. One of these approaches, “Interactionism”, is the foundation of multiagent research. After a top-level overview of multi-agent system technology and how it compares to S-D logic, some suggestions are offered from the information science perspective that may be helpful in terms of bridging the discourse between marketing and information science.

Keywords: S-D Logic; Service; Multi-agent systems; Interactionism; Agents

Introduction The current discussions in the field of marketing studies concerning service-dominant (S-D) logic, as outlined by the work of [Vargo and Lusch, 2004a, 2004b, 2006, 2008a, 2008b] and the responses their work has generated have attracted the attention of some of us from the field of information science. This interest stems from our belief that the underlying issues in the S-D logic discussions have affinities with a similar debate that has taken place within information science, and indeed more broadly over a number of years across a wide spectrum of Western culture. Since what is at stake in this debate is essentially how one perceives the very nature of information, this issue lies at the very heart of all scientific understanding. Consequently, we believe that the same issues about the nature of information that underlie some discussions in information science are also crucial to the understanding of markets and economies, and they are being uncovered in the current S-D logic discussions. In this paper, we offer our general viewpoint concerning the nature of information and describe how we believe our perspective correlates with that of the S-D logic community. Our aim in this paper is two-fold: (a) to associate and potentially coordinate some of the terminology across the two disciplines, and (b) to offer some suggestions from our information science perspective concerning potential exploratory directions which the S-D logic community may fruitfully pursue. The remaining portions of this paper are organised as follows. Section 2 offers a very brief overview of the Service-Dominant logic perspective and how it is distinguished in the marketing discipline from the Goods-Dominant (G-D) logic perspective. Section 3 presents a general discussion concerning the nature of information and how it can be understood from two contrasting standpoints, which we call “Objectivism” and “Interactionism”. This discussion provides the background context in which the specific divides in both information science and marketing can be situated. Section 4 describes how the Objectivist perspective 166

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

manifests itself and prevails in computer/information science. In Section 5 a high-level view of multi-agent systems (MAS) technology, an area of information science that derives from the Interactionist perspective, is presented. This technology in information science corresponds to the S-D logic view in marketing. Section 6 then presents some comparisons between multi-agent systems technology and S-D logic notions and offers some suggestions concerning how ideas from MAS technology may be applicable in the S-D logic context.

Service-dominant logic and goods-dominant logic In some recent papers, [Vargo and Lusch, 2008a, 2008b] have outlined and refined their view of “service dominant (S-D) logic” in the field of marketing and how it contrasts with “goods dominant (G-D) logic”. The use of the term ‘logic’ is here used broadly and evidently refers essentially to a foundational discourse pattern by means of which ideas and models are presented, rather than to a rigorous formulation in logic. The goods-dominant (G-D) logic is their term for the traditional way of viewing economic transactions, and it is founded on the idea that economic activity fundamentally concerns the production and distribution of “goods”, which are tangible units of output. Goods are objects that have an inherent value, and the goal of an efficient economic organisation or entity is to maximise profit (value) from the production and distribution of the goods. From the traditional G-D logic perspective, services are understood to be a special, intangible type of goods. Services are merely helpful in the provision of goods, but they are somewhat slippery and difficult to commodify when compared to “real” goods. From the G-D logic perspective, services can help make goods more accessible, but they cannot produce the most important entities, the goods. The contrasting service-dominant (S-D) logic, on the other hand, places services at the centre of the stage: it is services, rather than goods, that are understood to be the basis of economic exchange. In information science terminology, we would say that a service in S-D logic is defined as a “first-class” entity, and it is thus not defined to be derivative from the notion of goods. In the world of S-D logic, then, economic activity is entirely the provision of services, and goods are secondary. That is, goods may participate as accessories in some of the exchanges of service, but they are not essential. Two additional, special terms that are critical to the S-D logic discourse are “operand resources” and “operant resources”. Operand resources are essentially passive objects on which actions are performed, while operant resources are employed to act on the operand resources [Constantin and Lusch, 1994]. Vargo and Lusch have developed a set of foundational premises for S-D logic which provide a schema for its main principles (see Table 1), and we will take these to be canonical. We remark at this point that although the S-D/G-D logic discussion has emerged in the academic literature of marketing, the issues raised extend beyond the boundaries of that discipline and represent a re-examination of the fundamental basis of economic activity. As presented by Vargo and Lusch, the S-D/G-D logic dichotomy identifies two basic schemes for constructing general economic models. But in the next section we extend this generalisation even further and argue that an essentially similar division separates two fundamental approaches to the general problem of modelling the physical world. 167

Otago Forum 2: Academic Papers

Table 1. Foundational premises of service-dominant logic (adapted from [Vargo and Lusch, 2008a]).

FP

Foundational Premise

Explanation

1

Service is the fundamental basis of exchange

The application of operant resources (knowledge and skills, i.e. “service”, is the basis for all exchange.

2

Indirect exchange masks the fundamental basis of exchange

Because service is provided through complex combinations of goods, money, and institutions, the service basis of exchange is not always apparent.

3

Goods are a distribution mechanism for service provision

Goods derive their value through use – their service.

4

Operant resources are the fundamental source of competitive advantage.

The comparative ability to deliver services drives competition.

5

All economies are service economies

Service is becoming more prevalent as a result of increased specialisation and outsourcing.

6

The customer is always a cocreator of value

Value creation is always the result of an interaction.

7

The enterprise cannot deliver value, but only offer value propositions

Since value is created interactively, service providers cannot unilaterally create or deliver it.

8

A service-centred view is inherently customer oriented and relational

Service is understood in terms of customer-oriented benefit and therefore co-created.

9

All social and economic actors are resource integrators

The context of value creation is networks of networks.

10

Value is always uniquely and phenomenologically determined by the beneficiary

Value is experiential, contextual, and meaning-laden.

168

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

Modelling the world Beginning some three millennia ago the Greeks formulated general physical principles or metaphors about the structure and processes of the world. These included several analogies and mythological principles which culminated in the various theories of the Ionian philosophers Hesse [1962]. Aristotle subsequently collected the existing physical principles and rationalized them into four basic categorical causes: material, formal, efficient, and final. These four causes are understandable today, but we do not place them on an equal footing. We would invariably attribute the most significant causal efficacy to the efficient cause – for example, the causal agent that perpetrates a crime. In the courtroom, as in many other situations of our everyday lives, we attribute the principal element of causality to the efficient cause, the agent of change. Perhaps it’s not surprising that the Greek word for cause, aitia, also means guilt. The other causes would be merely seen as contributing factors or circumstances that were necessary for the event to occur, but were not the principal operating element. And so it was in the case of natural science, as it developed, too. There was a primary focus in locating and understanding the nature of the force that brought about a change. The ‘force’ could arise from some sort of push or pull exerted by an intelligent agent, or it could be associated with some natural phenomena. But associated with every force, even those arising from natural phenomena, was implicitly a causal agent. Thus when a rock pushes against another rock, the first rock is modelled momentarily as an agent doing the pushing. By the time of the 17th century, Kepler, Pascal, Descartes, Fermat, Galileo, and others were demonstrating that mathematics was a powerful tool with which to build precise models of natural phenomena. This led to a streamlining of Aristotle’s four causes into two ways of reckoning: ƒ On the one hand, there was a focus on identifying fundamental forces and characterising accurately how they operated. This entailed an assumption that there were (sometimes implicit) causal agents exerting these forces. This focus emphasised the importance of the efficient cause relative to the others and fit well with the common-sense notion of causality that is used in everyday life and in such practical domains as the law courts. ƒ

On the other hand, there was an increasing focus on the use of mathematics. Who could not be impressed by the degree of precision that could be given to planetary movements by using such mathematical tools? As success piled onto success, these developments led to increasing emphasis and reliance on mathematics, i.e. on Aristotle’s formal cause.

So the four principal causes had now been reduced to two main ones. The formal cause approach, however, did not always sit comfortably with the tendency to evoke the efficient cause. This was because the signal metaphor of the efficient cause, the human agent, was so difficult to understand and characterise precisely. If one could not describe the causal agent’s characteristics with precision, then how could one derive a precise mathematical model that could be validated by empirical observation? For this reason there was a growing tendency to eliminate any aspects of anthropomorphism from science. Descartes, for example, emphatically separated physical and mental phenomena into separate ‘worlds’, and he did 169

Otago Forum 2: Academic Papers

this in order to base his physics on clear and distinct phenomena, which would be the basis for precise and objective observation. By proceeding in this way, he sought to eliminate all the fuzzy anthropomorphism that existed in the physics of his day. Over the succeeding centuries, the predictive success of mathematically-based formalisms, led to a growing preference for formal-cause-based theories over causal-agent-based theories. In fact, Bertrand Russell argued that even the notion of causality was essentially anthropomorphic and should be dropped from scientific discourse: All philosophers, of every school, imagine that causation is one of the fundamental axioms or postulates of science, yet, oddly enough, in advanced sciences such as gravitational astronomy, the word `cause' never occurs ... The law of causality, I believe, like much that passes muster among philosophers, is a relic of a bygone age, surviving, like the monarchy, only because it is erroneously supposed to do no harm. Russell [1913]

So the notion of the causal agent, the efficient cause, receded from formal scientific descriptions, and “objective” science came to be dominated by just the formal cause. By the 20th century the scientific community century had gradually oriented themselves around a set of beliefs, which we will call, “Objectivism”, although there are many other terms, such as “Modernism” in use. Of course, most practicing scientists do not carry matters so far as to eliminate the notion of the efficient cause from their ordinary discourse, and they continue to use the metaphorical notions of causality and force in their everyday activities in the laboratory. Nevertheless, they generally do subscribe to the basic tenets of Objectivism as given in Figure 1: ƒ Sensory information from objectively independent entities is the starting point, “the given”. ƒ Basic phenomena/entities are the elementary particles of physics. ƒ Scientific results are independent of the investigator and independent of a human cultural context. ƒ All science understanding is fundamentally a set of linguistic statements. ƒ Science is cumulative, and the individual sciences can be understood to be part of a single, integrated logical scheme (the “unity of science”). ƒ A fundamental goal of science is to establish a complete axiomatisation, i.e. organise the statements into a coherent logical structure. Figure 1. Tenets of Objectivism Fundamental to Objectivism is the conviction that there is an objective, external world “out there”, which is independent of any observer. The observer is a passive onlooker who is subject to perceptual errors, and individual observers may frequently get it wrong. The spectacular success of the natural sciences (physics and chemistry) has reinforced belief in this point of view. Furthermore, the idea of the “unity of science” asserts that all of science is consistent, and the various individual sciences are simply specific representations of this single science. The 170

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

field of physics is the most fundamental: it investigates the most basic entities in the universe and discovers how these entities, the elementary particles, interact. So physics is the bedrock of science; and all other scientific fields must be consistent with it. Wherever there is an interface between two individual scientific fields, then the two fields must provide consistent and true predictions, with the more “fundamental” science having priority. The unity of science implies, of course, that all phenomena in the world, including consciousness, the notion of free-will, and the possible existence of a human soul, must be consistent with the fundamental laws of physics, chemistry, and biology. Despite the success and widespread acceptance of Objectivism, though, there has been a growing critical awareness that this approach is subject to limitations. This has given rise to an alternative modelling framework that has taken shape over the last century, which we will call “Interactionism”, although it could also be called “Phenomenological” or “Experientialist”. While Objectivism places primary emphasis on the pre-existence of the external objects of an objective world, Interactionism regards the interactions engaged in by the body-subject as primordial and taking precedence over any theories we might have of the external world. Merleau-Ponty stressed this difference in the following: The whole universe of science is built on the world as directly experienced, and if we want to subject science itself to rigorous scrutiny and arrive at a precise assessment of its meaning and scope, we must begin by reawakening the basic experience of the world of which science is the second-order expression. Science has not, and never will have, by its nature, the same significance qua form of being as the world we perceive, for the simple reason that it is a rationale or explanation of that world.

Merleau-Ponty [1962] The fundamental tenets of Interactionism are presented in Figure 2. ƒ Everyday world of embodied human experience is the starting point, “the given”. ƒ Scientific results are to be understood within the context of the interaction environment, of which the observer is a part. ƒ Understanding is achieved by iterating through the hermeneutic circle. ƒ Categories used in model-building are a result of the interactions of an embodied observer. ƒ Scientific models are always context-dependent. Figure 2. Tenets of Interactionism With Interactionism, the perceiver’s basic interaction with the world is primary. The observer is not some posited subjectivity that sits outside the world and oversees it. Rather, the experiential interactions that take place represent the starting point and the basis on which models are to be built. From the Interactionist perspective, an Objectivist model is not necessarily “wrong”, but it is 171

Otago Forum 2: Academic Papers

likely to have limitations in scope that need to be recognised. Just as a “flat Earth” geographic model can have practical use in local contexts, so, too, an Objectivist model can be convenient for certain contextual situations. This means that the Interactionist perspective is more comprehensive and more carefully specified than the Objectivist perspective; that is, it can employ Objectivist models where appropriate. The divide between the Objectivist and Interactionist viewpoints has appeared in many other fields besides philosophy. For example in the field of linguistics, George Lakoff [1987] has criticised Objectivism by coming to the conclusion that the field of linguistics is an Interactionist-inspired framework. It is our view that in the domain of marketing, the Goodsdominant (G-D) logic is fundamentally Objectivist, while the Service-dominant (S-D) logic is fundamentally Interactionist. In fact these two marketing perspectives represent the poles of the Objectivist-Interactionist divide in the field of marketing. In the following two sections we discuss how that same Objectivist-Interactionist divide manifests in our own field of information science.

Objectivism in Information Science As with other disciplines, the traditionally dominant view in information science has been that of Objectivism. From this standpoint it is possible to construct “objective” information about the world that is independent of any observer. In this connection there are efforts to construct objective “ontologies”, which are explicit, formal specifications of terms and concepts that describe a domain [Gruber, 1994; Calero, Ruiz, and Piattini, 2006]. Objectivists implicitly believe that it would be theoretically possible to construct a formal ontology of the entire world. In any case, it has been assumed possible to build up large sets of “objective” knowledge, and then construct reasoning engines to trawl through this “knowledge” and exhibit intelligent decision making. The traditional symbolic artificial intelligence community (sometimes referred to as GOFAI – “good, old-fashioned AI”), has for many years been working on techniques based on this proposition. Their conviction has been that provided that enough knowledge is collected, enough computational power is assembled, and sufficiently clever logical inferencing mechanisms devised, then it will be possible to achieve “intelligence” that is superior to the human brain. As part of this enterprise, and probably the most clear-cut exhibition of the Objectivist predilection, has been in an area of information science known as “Computationalism”. Computationalism is a form of Objectivism that asserts that human cognition, itself, is a form of computation, and this notion lies at the basis of most work in Artificial Intelligence and Cognitive Science. In other words, the Computationalists go beyond merely asserting that GOFAI work can match human intelligence and claim that the human brain is simply another Turing machine. Generally speaking, Computationalism usually entails a few further premises: ƒ The human mind is a computational machine, and therefore cognition is mechanical. ƒ Objective information is perceived by sense organs and passed to the brain for processing. ƒ Information is data that has meaning and can be encoded by mechanical, i.e. logical, formulations. ƒ The external world can be represented as a mechanical model with a finite number of 172

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

states. This implies that one needs computational mechanisms to search among these possible states for the desired one. There are, of course, disagreements as to the precise specification of Computationalism, but Chalmers’ description Chalmer [1994] points out that there are two different commonly understood aspects to Computationalism: ƒ computational sufficiency. A suitable computational structure is sufficient for the possession of a mind. Loosely speaking, this suggests that computers have the potential to possess a mind. ƒ computational explanation. This states that the theory of computation provides a framework for the explanation of human cognitive processes. And both of these aspects of Computationalism are widely held by most, but not all, cognitive scientists. Certainly we are familiar with the many ways in which modern computers appear to mimic, or exceed, the capabilities of the human mind. For example, they can play the game of chess and beat all but the most advanced masters of the game. But the proponents of Computationalism base their beliefs on more solid evidence namely the Church-Turing thesis [Lewis and Papadimitriou, 1981; Copeland, 2002]. Turing’s expression of this thesis was based on his “Logical Computing Machines” (LCMs), which are now known as Turing Machines and have been shown to have the same computational capabilities as digital computers: Turing’s thesis: Every function which would naturally be regarded as computable can be computed by a Turing machine. This thesis cannot be proven, because of the vagueness of the statement “function which would naturally be regarded as computable”. The intended meaning, however, is to assert that any algorithm can be computed by a Turing machine. In fact since any non-interactive computer program can be translated into a Turing machine and any Turing machine can be translated into a Turing-complete programming language (such as all the widely-used general-purpose programming languages, like Java, FORTRAN, and C), then we can say that any general-purpose programming language can express any possible algorithm. To see how Computationalism rests on top of Objectivism and the “unity of science”, consider this statement from the Encyclopedia of Artificial Intelligence: Suppose that the highest-level brain processes to which human conscious and unconscious thoughts and/or symbol manipulations are reducible, are algorithmic, and that those brain processes are really produced by precise, finite, deterministic recipes somehow “wet-wired” into the human brain. Then human cognition is simulatable by Turing machines. Hence, any limitative results about Turing-machine computations apply to humans too, and perfect computer modeling of human cognition is, in principle, possible. This form of mechanism is a principle assumption of modern cognitive science and implies that artificial intelligence can, in principle, do anything natural intelligence can.

Shapiro [1987] 173

Otago Forum 2: Academic Papers

Moreover, many Computationalists make assertions that go beyond what Turing originally asserted, for example, Any process that can be given a mathematical description (or that is scientifically describable or scientifically7 explicable) can be simulated by a Turing machine.

Copeland [2002] An extreme form of Computationalism holds that the entire physical universe is some sort of computational machine, such as interconnected cellular automata, and that all physical processes, including life forms, are simply higher-level programs running on this basic machine. In this view, all world processes are essentially simulations running on this machine, somewhat along the lines posited in the film, “The Matrix1”. Some information scientists dispute the claims of Computationalism, but they are presently in the minority. In addition some critics of Computationalism point out that there exists physical computer architectures more computationally powerful than Turing machines and thereby go beyond the Church-Turing thesis. But opinion is divided as to whether these “hypercomputation” architectures contradict the assertions of Computationalism or merely extend them [Copeland, 2002; Syropoulos, 2008]. But we set aside those arguments and move on to the Interactionist camp in information science.

Interactionism in Information Science (multi-agent systems) The Interactionist perspective in information science adopts a more cautious view about what is the nature of the informational world. The key point from the Interactionist perspective is that each person constructs models of his (or her) environment based on the interactions in which he (or she) participates. In order to interact effectively with others, it is, of course, desirable that the individual participant world models be correlated as much as possible. But under the Interactionist scheme such a shared ground is a goal, rather than the assumed starting point. Intelligent behaviour is still possible under the Interactionist scheme, but it arises as an emergent phenomenon from the interactions of many participants, rather than as a logical deduction from a single, GOFAI-constructed giant brain. The idea is that just as a community of individual people, each member of which is a specialist at something, can collectively cooperate to carry out a complex task, so, too, a community of software computational agents, each of which is a specialist in one area, can work together to carry out a complex software task. This has led to the active research community in the area of Multi-Agent Systems (MAS), although most MAS researchers do not explicitly refer to ‘Interactionism’ or its synonyms. In this section, we provide a high-level overview of what multi-agent systems are and some of the research issues in this field.

1

The Matrix, Warner Bros., 1999 174

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

Definition of Agent To begin with it is important to distinguish between two uses of the term ‘agent’ in the computing community: 1. a software architecture agent is an entity in a computing architecture that embodies certain properties said to be characteristic of “agents”. This entity operates like an agent. 2. a user-interface agent is a software utility that interacts with a human user by giving the appearance of human-like behaviour. This program looks like an agent to the user, but it may not have the architectural characteristics of an agent. An example of this type was “Microsoft Bob” Tynan [2006]. In the following discussion, our use of the term ‘agent’ is restricted to software architecture agents, i.e. option 1 above. Even with this restriction in scope, though, arguments persist in the MAS community concerning the precise definition of a (software architecture) agent. Nevertheless most academics and professionals do agree that an agent possesses at least the following characteristics: ƒ Interactive: the agent interacts with the environment and other agents. It maintains social relationships. ƒ Autonomous: the agent acts without direct external intervention. It has some control over its internal state. ƒ Reactive: the agent obtains information about its environment and responds by taking some action in a timely fashion. ƒ Reflective: the agent maintains some state information, which can be consulted and updated. This state can be said to represent the agent’s “beliefs” about its environment. ƒ Proactive: the agent is goal-oriented, not simply reactive. It can purposefully try different schemes in order to achieve its goals. ƒ Cooperative: the agent is capable of coordinating with other agents in order to achieve a common purpose. In addition to those general features, there are some other attributes that are associated with some agents: ƒ Mobility: an agent is sometimes able to move via network connections from one platform to another. ƒ Rational: an agent may have some degree of computational intelligence that is employed in order to achieve its goals. A popular rational agent subtype is a BDI agent, which employs an internal agent architecture that embodies “Beliefs”, “Desires”, and “Intentions” [Bratman, 1999; Rao and Georgeff, 1995]. ƒ Adaptive: an agent may change its behaviour or learn new behaviours based on experiences it has with its environment. Naturally, all of these features are recognisable to an outside, non-technical observer as attributes of biological agents. But these terms have further, specific semantics in the domain of information science with respect to their instantiation on real computing systems. The specific choice for each of these terms is essentially metaphorical, and the true meaning of 175

Otago Forum 2: Academic Papers

the terms can only be characterised and understood by examining the behaviour of real, executable multi-agent systems. In other words all of these agent attributes have operational definitions within the context of computing information systems. For example, the term “Autonomous” does not mean that the agent actually has free will, but merely signifies that the agent operates within its own thread of execution. This essentially means that it runs in its own separate computing process and according to its own defined semantics – it is not being controlled by some external puppet-master control module or by the underlying operating system. Similarly the term “Reflective” (other terms for this are also used) only indicates that the agent maintains some modifiable state information. This means that the intelligent agent may react differently on separate occasions when presented with the same stimuli or environmental conditions.

Agents versus objects Given the semantics of a computer software agent, one might ask how it differs from an “Object” in the sense of “Object-oriented programming”. Since the elucidation of the distinction between these two will be useful for later discussion in this paper, we offer some brief comment. Object-oriented programming languages, such as C++, Java, and CLOS (the Common Lisp Object System), provide an implementation of the object programming paradigm on computer systems. To the software programmer who uses these languages, the metaphor of a persistent “object” is offered by an object-oriented programming system. An Object in this context maintains state (has variable information that can be updated) and offers “services”. The software object’s services are implemented by means of executable procedures, subroutines, or “methods”, and these can be invoked by “sending a message” to the object (i.e. calling the procedure). In actual fact, an object is realised by maintaining some variables and procedures within the scope of a particular software object name. But what is accomplished is the embodiment of a higher-level abstraction – a software object. At this higher level, the object maintains its memory (its internal state) and offers a published set of services (i.e. it “advertises” them) to other objects. For any external object that knows about our given object to avail itself of these advertised services, it must send a message to our advertising object, asking it to invoke one of advertised its methods on some data that is also included in the message. Thus object-oriented programming offers a metaphor of an environment comprising objects which send messages to each other in order to have services performed. This is the same metaphor that is invoked by the term “Service-Dominant” logic. However, there is a significant technical implementation difference between the way objects interact and the way agents interact. Although metaphorically we can think of objects “sending messages” to each other, the way this is implemented is by a procedural or method call. This means that the calling object must wait for the result from the receiving object. Such a system of interaction requires a highly reliable interaction medium, and the interaction is said to be “synchronous”. Agents, on the other hand, interact asynchronously. An agent may send a message, such as a request for information, and then immediately attend to other activities, without waiting around for the response. The sending agent will expect a response 176

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

message sometime later from the receiving agent containing the requested information. But in the meantime, the sender agent can attend to other matters. Thus agents engage in conversations comprising sequences of several messages. Agents provide a higher-level abstraction beyond that of software objects, and indeed agentbased platforms are often developed on top of object-oriented technology. Moreover, agents typically consist of a number of objects. Instead of simply a metaphorical environment for service invocation, MAS technology offers a metaphorical environment for a community of interacting, goal-oriented autonomous agents. These agents can invoke services, but they can do much more: they can engage in extended interactions (conversations) in order to realise plans to achieve complex goals. This facilitates an Interactionist view of information technology.

Agent communication languages In order for a group of agents to interact effectively, they need to have a shared communication language. Of course since agents can have an almost infinite range of applications, one might ask how they can ever arrive at a common vocabulary if we desire to build open distributed information systems? The technical response so far on the part of the MAS community to this challenge has been to fashion an agent communication language (ACL) that employs the linguistic conception of “speech acts” (“illocutionary acts”) Searle [1969]. This recognises the fact that when we employ ordinary language in conversation, our utterances usually contain an implied action: when we say something, we are actually doing something. For example, when someone hears, “look out!”, that could be interpreted as a warning of danger. Other speech acts can be questions, invitations, commands, declarations, etc. Searle set up a simple classification scheme that has been adopted by the Foundation for Intelligent Physical Agents (FIPA)2 [2008] for their ACL. ƒ ƒ ƒ ƒ ƒ

Assertives: a speaker commits to the truth of a statement. Directives: a speaker wants the hearer to take a particular action, e.g. requests or commands. Commissives: a speaker commits to a future action, such as a promise. Expressives: a speaker expresses an attitude, such as congratulations, sorrow, or thanks. Declaratives: statements that effect a changed reality, e.g. pronouncing a couple man and wife.

The FIPA ACL is a two-tiered message system. The outer layer of a message is made up of one of the FIPA-defined communicative acts (speech acts) that represent the type of action that is being requested or effected. The inner layer consists of the message proposition expressed in terms of an ontology that is shared by the two communicative participants. Individual agents may be implemented in different programming languages and on different platforms and still communicate, as long as they can exchange messages using the FIPA communicative acts. Simple example FIPA communicative acts are inform (an assertive) and 2

FIPA is an organisation that facilitates the development of agent-based system standards. 177

Otago Forum 2: Academic Papers

request (a directive). Figure 3 shows an example of what a simple FIPA request message looks like.

(request :sender Agent-007 :receiver Agent-008 :content "open \"db.txt\" for input" :language vb :ontology X-files

)

Figure 3. A FIPA request message From Figure 3 it is evident that the FIPA ACL specifies the outer wrapper and that the actual message content can be expressed in any language and make reference to any ontology, assuming that these are understood by the communicating parties. In open distributed systems, where new agents may come and go, the maintenance of a shared ontology is still a significant research problem, and much effort has been devoted to the development of techniques for representing ontologies and for reasoning about messages that have been expressed in terms of them [Tamma, Cranefield, Finin, and Willmott, 2005]. Rather than rely only on computationally burdensome, GOFAI-based mechanical reasoning procedures to understand the inner content of the message, though, we can reduce some of this burden by doing what ordinary people do – employ “conversation policies” (also called “interaction protocols”). For example, when a person enters a restaurant, he (or she) doesn’t have to worry about all the possible statements that might be made about food. Instead, he (or she) expects to be given a menu and to place an order. Later the food will be brought, and only afterwards (for this particular restaurant, anyway) will he (or she) be expected to pay the bill. This may be called a “restaurant interaction protocol”, and the existence of such a protocol greatly reduces the search space of possible responses required, which is limited to the responses appropriate to the particular point that one has reached in the protocol. Each of these individual interactions involve the provision of a service, and the overall interaction, the “conversation”, entails the multiple services provided by multiple parties. The customer, the waiter, the cook, and the cashier all know this protocol and keep track of where they are in terms of it3. In our own work we have developed a new technology employing Coloured Petri Nets [Purvis, Hwang, Purvis, Cranefield, and Schievink, 2002; Purvis, Nowostawski, Oliveira, and Cranefield, 2003]. This technology is used for the efficient computer representation and transfer of resident interaction protocols to new agents when they enter an already existing society of interacting agents. Examples of this representation for the simplest conversation, the request4, is shown in Figures 4 and 5. The interaction protocol technology facilitates the ability of software agents to engage in multiple, asynchronous conversations with other 3

Note that the waiter and the cashier may be holding many simultaneous conversations with various customers, all using the same protocol. 4 Here ‘request’ refers to the interaction protocol, not the communicative act. 178

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

agents over time. Related techniques that have been explored in this connection include dialogue games [McBurney and Parsons, 2002] and agent narratives Purvis [2004]. In Not-understood Request

Not-understood Rec. request answer

Start

Request sent

In

Receive request

Receive request result

answer Process answer

Process request

Send not understood Refuse

Fail

Refuse

Result Send agree

Result Done

Agree sent Agree

Agree

Send request

Fulfill request

Send refuse

Done Send result

Fail

Send done

Out

Figure 4. Request: Initiator

Out

Send fail

Figure 5. Request: Participant

Interaction protocols are related to games, and there has been interesting MAS work extended into this area, as well. When agents engage in an economic transaction, such as an ecommerce auction, they are effectively playing a game that has its rules of interaction, with the goal of maximising their own profits. The game has its specific rules of play, and interactions need not be linear sequences; they can involve multiple, concurrent actions as long as they conform to the rules. Besides ontologies and interaction protocols, there are other MAS technologies that have been developed to facilitate extended interactions, including commitments, trust, institutions, and norms. ƒ Commitments. We mentioned that one type of speech act is a commissive. Some agent messages imply a commitment to a future action (or sometimes a refraining from some action). Commitments can be modelled as parts of an interaction protocol, but in complicated interactive situations, it can be more efficient to separate out commitments as separate concepts [De Oliveira, Cranefield and Purvis, 2007; Desai, Narendra, and Singh, 2008]. ƒ Norms and policies Agents can be expected to conform to social norms, which are more general than specific interaction policy rules. Sometimes a norm can be somewhat arbitrary, such as driving down the left-hand side of a road, but it is useful if all the agents conform to it. There is research on agent norm emergence and norm propagation, in addition to research investigating mechanisms for sanctioning uncooperative agents that violate norms [Boella, 2003; Savarimuthu, Cranefield, Purvis and Purvis, 2007]. ƒ Trust. If agents disobey the interaction protocol rules or violate norms, they can be 179

Otago Forum 2: Academic Papers

deemed untrustworthy. In addition, some agents, even if they don’t specifically engage in rule infractions, can be deemed to be “free riders”. This is a difficult problem in ecommerce, because malevolent operators can always dispose of existing agents branded as malefactors and create new agents with new identities that are difficult to link to previous harmful agents. Reputation scores and referrals are used to maintain a web of trust about known agents in a group [Ramchurn, Huynh, and Jennings, 2004]. ƒ Institutions. When agents engage collectively in some cooperative activity, they can operate according to the standards of an institution. An institution is a society of agents that has various mechanisms for keeping track of the members, maintaining reputations, archiving relevant ontologies and interaction protocols, and keeping track of norms and policies. This information needs to be made available to new agents that enter the institution, and there can be special institutional agents that enforce policies and perhaps expel uncooperative agent members [Nowostawski, Purvis, De Oliveira and Cranefield, 2006].

Multi-agent systems and S-D logic – two interactionist domains Since, as remarked above, S-D logic represents the Interactionist pole of the G-D/S-D divide, it corresponds with the MAS work vis-a-vis information science. Crucial testimony in this connection is the statement by Vargo and Lusch that value creation is “phenomenological and experiential in nature” [Vargo and Lusch, 2008a]. This lies at the heart of Interactionism, and we take this to be an overriding principle of S-D logic thought. As a consequence, we think that much of the distributed multi-agent work relates to the issues discussed in the servicedominant logic marketing literature, and we think that some MAS practices and results may be of interest to the service-dominant logic community. At this stage we propose a dialogue between the two communities which may be mutually beneficial. As an initial gesture in this proposed dialogue, we offer here some observations concerning our perception of the S-D literature: 1. While we appreciate that there are undoubtedly carefully considered reasons for the choices of terminology in the S-D logic literature, we wonder if the use of the word ‘service’ in the phrase, “service dominant logic” gives rise to some obscurity. As Vargo and Lusch have pointed out [Vargo and Lusch, 2008b], ‘service’ in the G-D logic sphere has a different meaning from how they have used that term in connection with S-D logic. In G-D logic terms, ‘service’ suggests something ancillary to goods and is offered by the provider to enhance access or utility of the value-laden entities, i.e. the goods. Services, in G-D logic terms, are simply presented as an intangible form of goods. But in S-D logic, ‘service’ is recast to be fundamental. Thus there are two quite different interpretations of the term, depending on whether the Objectivist or Interactionist standpoint is taken. The same kind of problem appears in information science in connection with the term ‘knowledge’. In the Objectivist realm of information science, ‘knowledge’ refers to an objective, quantifiable entity. This could take the form of either data accessed by table180

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

lookup procedures or a set of if-then rules that are accessed by rule-based engineers. But in the Interactionist realm, knowledge can encompass “know-how” and refers more generally to knowing how to engage in an operation. For example, knowing how to ride a bicycle can be a form of knowledge, but it is not quantifiable in a set of rules and data. As a consequence, the term ‘knowledge’ has become a source of confusion in information science, and we avoid using the term altogether. In addition, ‘service’ suggests an action provided by a provider to a recipient. There is a directional polarity implied by the term. But in general interactions, there may not be such an obvious polarity. When two people shake hands, it is possible, but somewhat artificial, to view the action as comprising two services: each person providing a service to the other. As an illustration, Figure 6, which presents our interaction protocol diagram for an electronic trading game participant’s role, shows there are multiple operable ‘services’, and it may be more convenient to think of the entire description as an interaction. Vargo and Lusch have recast ‘service’ to encompass general interactions, and their references to “value co-creation” (FP6 in Figure 1), which implicitly retreats from the directional notion of ‘service’, and therefore incline us to suggest that the word ‘interaction’ be considered as an alternative.

Rec. bid accpt.

Receive corner announce

In

Rec. bid from dealer

Rec. traded cards

Accpt.

Process rec'd cards Cards rec'd Discarded rec'd. cards

Cards sent + bid info.

Rec. cards

Rec. start signal

Cards in

Restore hand

(end)

Cards dealt Start play

Bid rec'd

Cards Make bid

Send timeout Cards left

Discarded rec'd. bid

Corner?

Bid timeout Corner

Cards offered Accept bid Bid accpt. timeout Bid accepted + cards to send

Send cards to player

Out

Figure 6. Interaction protocol for ‘Participant’ role in Pit trading game 181

Otago Forum 2: Academic Papers

2. We note that reference is commonly made in the S-D logic literature to the “produce/consumer” dyad, and we offer as a suggestion that it might be more fruitful to refer to this as the “initiator/participant” dyad in connection with agent roles. We find that in an extended MAS operational protocol, there is always an initiator role, which can attract the participation of other, possibly many, participants. During the interaction, as mentioned above, there may be various brief “producers” and “consumer” roles, but there is often a persistent meaning associated with the agent that initiated the interaction, as distinguished from the subsequent participants. 3. We have argued above that the Interactionist perspective of information is more comprehensive than the Objectivist perspective. It can thus encompass and employ Objectivist models in appropriate circumstances. This suggests that the S-D logic thinking, too, can encompass and employ G-D logic concepts and terminology if due care is given to this employment. Thus the use of the term “resource” carries an Objectivist connotation for us, and therefore it seems to be derived from a G-D logic view. This suggests to us that “resource” may be a concept that has a narrower scope than the full domain of S-D logic thinking and its usage should perhaps be restricted accordingly. 4. Fundamental to the service interactions are the participating agents. Vargo and Lusch have recently placed a greater emphasis on this aspect, which they refer to as the operants. We strongly endorse this inclusion, since it conforms more closely to the conceptual basis of the MAS research community. 5. The work of Ballantyne and Varey on dialogical models of S-D logic [Ballantyne and Varey, 2006a, 2006b] strikes us as very much in harmony with the extended agent “conversation” research that has been done in the multi-agent community. There is considerable MAS work here on constructing dialogical models that may be used to characterise extended interactions, and we believe that this a promising area for future work. We are confident that this work can be extended in the direction of narrative modelling. For example, recent developments in microblogging [Java, Song Finin and Tseng, 2007] is proving to be an interesting direction for future investigation into low threshold “micro” interaction-expressed narratives. 6. The MAS research in the areas of commitments, trust, institutions, and norms is likely to have direct connections with the interests of the S-D logic community. We believe that these are areas that would extend the power and scope of the S-D logic approach to move beyond the “object-oriented” (service) paradigm and correspond to the more comprehensive “agent-oriented” (interaction) paradigm.

Conclusions In this paper we have outlined correspondences and affinities between multi-agent systems research in information science and service-dominant logic research in marketing. Both approaches are domain-specific applications of Interactionist thinking. We are enthusiastic about the opportunities for fruitful cross-fertilisation between the two disciplines. Because both disciplines overlap in the application domain of electronic commerce, we have made some suggestions concerning terminology and future research directions which may contribute to future collaborative possibilities. 182

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

References Ballantyne, D., & Varey, R. J. (2006a). Introducing a dialogical orientation to the servicedominant logic of marketing. In Toward a Service Dominant Logic of Marketing: Dialog, Debate and Directions, R. F. Lusch, & Vargo, S. L., eds.pp. 224–235. Armonk, N.Y.: M. E. Sharpe. Ballantyne, D., & Varey, R. J. (2006b). Creating value-in-use through marketing interaction: the exchange logic of relating, communicating and knowing. Marketing Theory, 6(3), 335– 348. Boella, G., (2003) Norm governed multiagent systems: the delegation of control to autonomous agents, Proceedings. of the IEEE/WIC Intelligent Agent Technology Conference. Bratman, M. E. (1999) Intention, Plans, and Practical Reason. CSLI Publications, Stanford, CAL. Calero, C., Ruiz, F., & Piattini, M. Ontologies for Software Engineering and Software Technology, Springer, New York, 2006. Chalmers, D. J., (1994) A computational foundation for the study of cognition, ftp://cogsci.indiana.edu/pub/chalmers.computation.ps. Constantin, J. A. &.Lusch, R. F. (1994), Understanding Resource Management Oxford, OH:, The Planning Forum, Oxford, OH. Copeland, B. J. (2002) The Church-Turing thesis. Stanford Encyclopaedia of Philosophy, Metaphysics Research Lab, CSLI, Stanford University, Stanford, CAL, http://plato.stanford.edu/entries/church-turing/. Copeland, B. J. (2002) Hypercomputation. Minds and Machines, Volume 12, Number 4, Springer, pp 461-502. De Oliveira, M., Cranefield, S., & Purvis, M. K. (2007) Normative spaces in institutional environments by the means of commitments, reputation and colored petri nets. Proceedings of the 8th International Workshop on Agent Oriented Software Engineering (AOSE 2007). Honolulu, HI. Desai, N., Narendra, N. C., & Singh, M. P. (2008) Checking correctness of business contracts via commitments. Proceedings of the 7th International Joint Conference on Autonomous Agents and MultiAgent Systems (AAMAS). Foundation for Intelligent Physical Agents (2008), Institute of Electrical and Electronics Engineers, http://www.fipa.org/. 183

Otago Forum 2: Academic Papers

Gruber, T. (1993). A translation approach to portable ontology specifications". In: Knowledge Acquisition. 5:199-199. Hesse, M. Forces and Fields, Dover, New York, 1962., pp 29-50. Lakoff, G. Women, Fire, and Dangerous Things, University Of Chicago Press, Chicago, 1987. Java, A., Song, S., Finin, T. & Tseng, B. (2007) Why we twitter: understanding microblogging usage and communities. Proceedings of the 9th Webkdd and 1st Sna-kdd 2007 Workshop on Web Mining and Social Network Analysis, San Jose, CA, pp 56-65. Lewis, H. R. & Papadimitrious, C. H. Elements of the Theory of Computation, Prentice Hall, New York, 1981, pp 222-224. McBurney, P. & Parsons, S. (2002) Dialogue games in multi-agent systems. Informal Logic,. 22 (3): 257-274. Merleau-Ponty, M. Phenomenology of Perception, Routledge & Kegan Paul, London, 1962, pp viii-ix. Nowostawski, M., Purvis, M. K., De Oliveira, M., & Cranefield, S. (2006) Institutions in the opal multi-agent system”, Journal of Intelligent and Fuzzy Systems, 17 (3): 191-207. Purvis, M. K., Hwang, P., Purvis, M. A., Cranefield, S. J.,& Schievink, M.(2002) Interaction protocols for a network of environmental problem solvers”, Proceedings of the 2002 iEMSs International Meeting: Integrated Assessment and Decision Support (iEMSs 2002), Volume 3, Andrea E. Rizzoli and Anthony J. Jakeman, eds., The International Environmental Modelling and Software Society, Lugano, Switzerland pp 318-323. Purvis, M. K., Nowostawski, M., Oliveira, M., & Cranefield, S. (2003) Multi-agent interaction protocols for e-business. Proceeding of the 2003 IEEE/WIC International Conference on Intelligent Agent Technology, J. Liu, B. Faltings, N. Zhong, R. Lu, and T. Nishida, eds., IEEE Press, Los Alamitos, CA, pp 318-324. Purvis, M. K. (2004) Narrative structure for multi-agent interaction”, Proceedings IEEE/WIC/ACM International Conference on Intelligent Agent Technology (IAT 2004), N. Zhong, J. Bradshaw, S. K. Pal, D. Talia, D., J. Liu, and N. Cercone, eds., IEEE Press, Los Alamitos, CA, pp 223-238. Rao, A. S. & Georgeff, M. P., (1995) BDI agents: from theory to practice. Proceedings of the First Intl. Conference on Multiagent Systems, San Francisco, 1995. Ramchurn, S. D., Huynh, D., & Jennings, N. R. (2004) Trust in multi-agent systems. The Knowledge Engineering Review, 19 (1): 1–25. Russell, B. (1913) On the notion of cause, Proceedings of the Aristotelian Society, 13:1-26. 184

Wider perspectives on service dominant logic Martin Purvis & Andrew Long

Savarimuthu, B. T. R., Cranefield, S., Purvis, M. A., & Purvis, M. K., (2007) Role model based mechanism for norm emergence in artificial agent societies. Coordination, Organizations, Institutions, and Norms in Agent Systems III, J. S. Sichman, J. Padget, S. Ossowski, and P. Noriega, eds.Springer-Verlag, Berlin, pp 203-217. Searle, J., Speech Acts: An Essay in the Philosophy of Language, Cambridge University Press, Cambridge, UK, 1969. Shapiro, S. Encyclopedia of Artificial Intelligence, vol. 2, John Wiley & Sons, New York, 1987, p 1624. Syropoulos, A. Hypercomputation: Computing Beyond the Church-Turing Barrier, Springer, 2008 Tamma, T., Cranefield, S., Finin, T. W. & Willmott, S. Ontologies for Agents: Theory and Experiences, Birkhäuser Basel, 2005. Tynan, D. (2006) The 25 worst tech products of all time, PC World, 26 May 2006, http://www.pcworld.com/article/125772-3/the_25_worst_tech_products_of_all_time.html. Vargo, S. L., & Lusch, R. F. (2004a). Evolving to a new dominant logic for marketing. Journal of Marketing, 68: 1-17. Vargo, S. L.& Lusch, R. F., (2004b). The four services marketing myths: remnants from a manufacturing model. Journal of Service Research, 6 (4) 324-335. Vargo, S. L.& Lusch, R. F., (2006). Service-dominant logic: what it is, what is is n ot, what it might be, in The service-dominant logic of marketing: dialog, debate, and directions, S. L. Vargo & R. F. Lusch, eds., ME Sharpe, Armonk, New York, 43-56. Vargo, S. L.& Lusch, R. F., (2008a). Service-dominant logic: continuing the evolution. Journal of the Academy of Marketing Science, 36:1-10. Vargo, S. L.& Lusch, R. F., (2008b). Why service. Journal of the Academy of Marketing Science, 36:25-38.

185