1.4 Thesis Structure
2.3.2 Extracting Arguments
2.3.2.3 Argumentation Scheme Based Approaches
Both the above approaches (i.e., the heuristic and the logic based) advocate agents to maintain belief representations about their counterparts and perform some form of rea- soning on these in order to extract the arguments to use within the dialogue. This allows agents to derive arguments most dedicated to their counterparts. However, within a multi-agent context, this requires them to maintain respective beliefs models about each of their counterparts within the society and perform reasoning on these representations every time they need to generate an argument. Thus, most of these techniques are stud- ied either at a theoretical level or implemented in a two agent context. Furthermore, their computational complexity has never been implemented or tested in a larger multi- agent scenario.
In this context, the concept ofargumentation schemes [Walton, 1996] is increasingly emerging as a mechanism for systematically identifying arguments within multi-agent literature. In essence, argumentation schemes capture stereotypical patterns of reason- ing upon which communication structures can be built. In more detail, these schemes represent patterns of human reasoning, especially defeasible ones, that have proved troublesome to view deductively or inductively. To illustrate this, consider the follow- ing scheme forargument from expert opinionextracted from [Walton, 2005]:
In more detail, the above scheme consists of three premises. These represent the as- sumptions that, if justified as acceptable, warrant the inference of the conclusion. How- ever, if the respondent is sceptical about the inference, he can challenge and critically
Scheme for Argument from Expert Opinion E is an expert in domain D.
E asserts that A is known to be true. A is within D.
Therefore, A may plausibly be taken to be true.
question the different elements within the scheme to establish this justification. To this end, the scheme acts as a stencil for both the participants to direct their challenges to one another and, thereby, engage in a dialogue to establish the validity of the conclu- sion. For example, in the above particular case, authors have identified the following six possible ways that a respondent can attack the above schema (as per [Walton, 1997]):
• Expertise: How credible is E as an expert source?
• Field: Is E an expert in the field that A is in?
• Opinion: What did E assert that implies A?
• Trustworthiness: Is E personally reliable as a source?
• Consistency: Is A consistent with what other experts assert?
• Backup Evidence: Is E’s assertion based on evidence?
More recently, a number of authors have argued for the use of argumentation schemes in computational contexts, including multi-agent systems, since they hold potential for sig- nificant improvements in reasoning and communication abilities in such systems [Reed and Walton, 2004; Walton, 2005]. One of the more recent efforts by Atkinson et al. [2004], highlights the use of this technique for extracting arguments in a computational context. In particular, their work extends Walton’s scheme forpractical reasoningin a manner to suit a computational context. This is stated as follows:
Argument Scheme for Practical Reasoning In the Current Circumstances R
an agent should perform Action A to achieve the New Circumstances S which will realise a certain goal G which will promote a certain value V.
Having used the scheme to state how a rational entity would practically reason to per- form actions, they then use it as a stencil (or a schema) to identify a number of ways of attacking this scheme. In particular, they highlight the following five major ways of attacking the above scheme:
• Denial of premises
• Alternative ways to satisfy the same value
• Side effects of the action
• Interference with other actions
• Disagreements relating to impossibility
These are then expanded to extract a series of arguments that agents can use to argue about the validity of that agent’s practical reasoning. For a more detailed analysis on the use of this technique refer to [Atkinson, 2005].
Having described these different approaches of extracting arguments proposed in litera- ture, we will now explain how they inspire and contribute to our research. In particular, our method for extracting arguments benefits mainly from the argumentation schemes approach discussed above. In essence, analogous to the practical reasoning approach adopted by Atkinson et al., we represent the social behaviour of agents as a schema for social reasoning and use this, in turn, to identify social arguments that agents can use to argue within a multi-agent community (see Section 3.2). The main advantage of using this method is its ability to identify arguments in anofflinemanner. In more detail, as highlighted above, the argumentation schemes’ approach allow us to identify and ex- tract a set of possible arguments to use against a typical agent within this context. Since all agents within the context are deemed to follow this stereotypical line of reasoning, agents can use these arguments against any typical agent within the community. They only need to consider which argument to use from this identified set. Using thisoffline
method reduces the computational cost of extracting arguments during the course of the encounter. Since all agents use this common schema, they only need to represent this schema and reason within it. Thus, they no longer need to represent a dedicated belief model for each of their counterparts and do complex reasoning during the encounter. This not only reduces the space requirement for representation, but also the reasoning required by an agent to identify arguments.
Furthermore, all these approaches (apart from Sierra et al.), completely ignore the so- cial context when generating arguments. Even the work of Sierra et al. only considers authority based relationships, which we believe only capture a specialised form of so- cial contexts (i.e., institutions or formal organisations). Our work, on the other hand, explicitly considers this societal element in extracting arguments (for more details refer to Section 3.2). Moreover, unlike Sierra et al., we present a more generic way of cap- turing these social influences of roles and relationships (i.e., using social commitment
with different degrees of influence; see Section 3.1). This not only provides a simple unified mechanism to extract arguments in different social contexts with a wide array of relationships and social influences, but also allows us to experiment with our agents’ ability to argue, negotiate and resolve conflicts in such disparate social systems.
Given the literature related to the way we extract arguments within our research, we now proceed to analyse the state of the art related to exchanging these arguments, specifically those that have inspired our language and protocol elements.