Protocols for Multiparty Argumentation
Whereas in the WP2 the focus is on a one-step aggregation, here we plan to explore the iterative aspect of the aggregation of argumentation systems. Rather than aggregating the individual inputs to immediately determine a group outcome, agents initially submit some information, and then have the possibility (in a way defined by a specific protocol) to revise their inputs or to provide more information in view of the other participants’ submitted inputs. If agents reach a consensus, they come to agree on the issues under discussion. This is different from a collective outcome defined by an aggregation rule, where the participants agreed a priori to accept the result of the employed aggregation procedure, that typically satisfy a number of desirable conditions. For example, the Condorcet Jury Theorem shows that, under the majority rule, a group is likely to reach the correct decision. Still, it is not always sensible to expect an agent who disagrees with the opinions of the majority to conform his views to the group position. The acceptance of other people statements may force the agent to revise beliefs that he holds strongly [Pettit06]. But agents are more likely to endorse a position reached by a deliberative process [LW81, RCM06].
Early works in multiagent argumentation were confined to the study of specific bilateral protocols, see e.g. [PSJ98,AMP00]. Recently, different properties of protocols for persuasion [Prakken06], which regulate the exchange of arguments to arbitrate among conflicting viewpoints, have been studied with the help of game-theoretical concepts (see [RL09] for a survey). Bonzon and Maudet [BM11] studied the properties of a multiparty deliberation protocol, in the case where agents may have different views on the attack relations among arguments. Then, they studied the influence of experts in such deliberation protocols in order to stabilize those debates [KBMM12].
The multiparty protocols that guide the deliberation process aim at realistically modelling debates among agents. The study of the properties that the aggregation of several argumentation systems should satisfy is the object of the WP2, which provides the theoretical framework for the WP3. Similarly to WP2, defining new deliberation protocols is not enough and must come along with a formal study of the properties of these protocols.
1. Definition of new deliberation protocols
The first task of this WP is to propose some multiparty protocols for deliberation. This raises many design issues. One obvious question is the turn-taking policy. Another question is to know if the protocol allows agents to revise or update their point of view, and how. Can we assume that agents share the same point of view on the attack relations between arguments and the same argumentation semantics? On what subject do the agents deliberate? Are they focused on a single issue, or do they want to deliberate about several issues?
2. Properties of deliberation protocols
Once the protocols are defined, we need to study the properties that one could expect from these deliberation protocols, in order to justify the obtained results. A first interesting question is about the efficiency of the protocol: is it for instance possible to obtain a “correct” collective outcome with a multiparty protocol? To do so, it would be interesting to compare the results obtained with such a protocol with the results of centralized aggregation procedures (as defined in the WP2). Another important question is about the fairness of the protocol: does every agent have a chance to put forward its own arguments and defend its views “fairly” with the proposed protocol? The question of the communication complexity should also be raised: as the process is at least partly distributed. Obviously, some properties may contradict each other: by constraining the protocol one reduces the communication complexity, but one makes also more likely to dismiss the position of some agents and make the protocol less “fair”. A source of inspiration to deal with this trade-off is to build on the idea of relevance-based protocols proposed by [Prakken05].
3. Resistance to manipulation
The step-by-step aggregation allows agents to strategize. On the one hand, this strategization can allow the debate not to be entirely predetermined: agents may have a chance to influence its outcome depending on how they play. On the other hand, this can lead to two different kinds of manipulation. The first one is the manipulation by an agent: is it possible for an agent to manipulate the result by lying, or by omitting some information? The second one is the manipulation by the designer: if the protocol uses a chair, is this chair able to influence the result by some strategic choice? For example, by choosing the agenda of the protocol?
[AMP00] L. Amgoud, N. Maudet, Simon Parsons. Modeling Dialogues Using Argumentation. ICMAS 2000: 31-38
[BM11] E. Bonzon and N. Maudet. On the Outcomes of Multiparty Persuasion. In AAMAS'11, p.47-54, 2011.
[KBMM12] D. Kontarinis, E. Bonzon, N. Maudet, P. Moraitis. Picking the Right Expert to Make a Debate Uncontroversial, in COMMA'12, p. 486--497, 2012.
[LW81] K. Lehrer, C. Wagner. Rational Consensus in Science and Society. Reidel, 1981.
[PSJ98] S. Parsons, C. Sierra and N. R. Jennings. Agents that reason and negotiate by arguing. Journal of Logic and Computation, 8, 261--292, 1998.
[Pettit06] P. Pettit. When to defer to majority testimony - and when not. Analysis, 66(3): 179-187, 2006.
[Prakken05] H. Prakken. Coherence and Flexibility in Dialogue Games for Argumentation. Journal of Logic and Computation, 15(6): 1009-1040, 2005.
[Prakken06] H. Prakken. Formal systems for persuasion dialogue. Knowledge Engineering Review, 15:1009—1040, 2005.
[RL09] I. Rahwan and K. Larson. Argumentation and Game Theory, chapter Argumentation in Artificial Intelligence, pages 321—339. Springer, 2009.
[RCM06] H. Regan, M. Colyvan, L. Markovchick-Nicholls. A formal model for consensus and negotiation in environmental management. Journal of Environmental Management, 80: 167-176, 2006.