ARGUMENTATION
1. Introduction to Argumentation
The theory of argumentation is a rich, interdisciplinary area of research straddling philosophy, communication studies, linguistics, psychology and artificial intelligence. Traditionally, he focus has been on “informal” studies of argumentation and its role in natural human reasoning and dialogue. More formal logical accounts of argumentation have recently been proposed by the artificial intelligence community as a promising paradigm for modelling common-sense reasoning and communication between reasoning entities. In these works, an argument is a set of premises offered in support of a claim. For example, consider the argument:
A1 = Information I about Tony should be published
because
Tony has political responsibilities
and
I is in the national interest
and
if a person has political responsibilities and information about that person is in the
national interest then that information about the person should be published
where Information I about Tony should be published is the argument’s claim, and the italicised statements following ‘because’ are the the argument’s premises. Consider the following counter-argument to A1 that undermines a premise in A1:
A2 = Tony does not have political responsibilities because Tony resigned from parliament and if a person resigns from parliament then that person no longer has political responsibilities
Consider the following counter-argument to A2:
A3 = Tony does have political responsibilities because Tony is now middle east envoy and if a person is a middle east envoy then that person has political responsibilities
Argumentation is the the process whereby arguments are constructed and evaluated in light of their interactions with other arguments. So, in the above example, arguments A1, A2 and A3 have been constructed. A3 ‘attacks’ A2 by contradicting A2’s claim, and A2 ‘attacks’ A1 by contradicting a premise in A1 (and so undermining A1). The winning arguments can then be evaluated. A1 is attacked by A2, but since A2 is itself attacked by A3, and the latter is not attacked, we obtain that A1 and A3 are the winning arguments.
This example illustrates the modular nature of argumentation that most formal theories (models) of argumentation adopt: 1) arguments are constructed in some underlying logic that manipulates statements about the world; 2) interactions between arguments are defined; 3) given the network of interacting arguments, the winning arguments are evaluated.
The appeal of the argumentation paradigm resides in this intuitive modular characterisation that is akin to human modes of reasoning. Also, recent work in the AI, and computer science community at large, has illustrated the potential for tractable implementations of logical models of argumentation, and the wide range of application of these implementations in software systems.
Furthermore, the inherently dialectical nature of argumentation models provide principled ways in which to structure exchange of, and reasoning about, justifications/arguments for proposals and or statements between human and or automated reasoning entities (agents).
Consider the above example where instead of a single agent engaging in its own internal argumentation to arrive at a conclusion, we now have two agents involved in a dialogue. Greg proposes argument A1, Alistair argument A2, and then Greg counters with argument A3. This represents a dialogue where each participant has the goal of persuading the other to adopt a belief through the process of exchanging arguments that must interact according to the underlying model of argumentation, and where the winning arguments are evaluated according to the underlying model of argumentation.
Of course, dialogues introduce an added dimension, in the sense that realistic dialogues often involve more than simply the exchange of arguments. For example, Alistair might challenge a premise in argument A1, by asking why information I is private. The burden of proof is on Greg to provide an argument as to why I is private. Otherwise, Alistair can be legitimately be said to be ‘winning’ the argument or dialogue. The formal study of dialogue models therefore accounts for a broader range of statements or ‘locutions’ than simply those involving submission of arguments.
Furthermore, the goal of the argumentation based dialogue may not only be to persuade, but also to collaboratively decide or deliberate over an appropriate course of action, or to negotiate over resources. In these dialogues, the reasons or arguments for proposed actions, or offers and rejections, can be usefully used to further the goal of the dialogue.
The goal of the dialogue may determine a specific set of statements or allowed locutions, as well as rules for making locutions at any point in the dialogue, and rules for determining the outcome of the dialogue. These rules are encoded in a dialogue’s protocol.
Consider for example the following negotiation dialogue between a buyer and seller of cars in which locutions also involve making, accepting and rejecting offers:
Seller - Offer: Renault
Buyer - Reject: Renault
Seller - Why
Buyer - Argue: Because Renault is a French make of car, and french cars are unsafe
Seller - Argue: Renaults are not unsafe as Renaults have been given the award of safest car in Europe by the European Union.
Buyer - Accept: Renault
The above example illustrates the utility of argumentation based models of reasoning and their application to dialogues. Online negotiations involving automated software agents are a key area of research and development. In a handshaking protocol, a seller would simply successively make offers and have these either rejected or accepted. The exchange of arguments provides for agreements that would not be reached in simple handshaking protocols. In the above example, it is by eliciting the reason for the rejection, and successfully countering this reason, that the seller is then able to convince the buyer to buy the car.
2. ASPIC and Argumentation: The aims of the ASPIC project
ASPIC is a collaborative project involving AI researchers and computer scientists in both the academic and commercial sectors. The aim of this project is two-fold:
1) To bring together a number of research groups who have established themselves in formal argumentation theory, in order to develop consensus theoretical models that improve on and extend existing models;
2) To develop efficient proof procedures and software component implementations of these models for deployment in real-world applications.
