Abstract
Value-Based Argumentation Frameworks (VAFs) allow for the resolution of conflicts by considering the underlying values of arguments. However, when multiple agents' preferences are aggregated to form a collective audience, the system becomes vulnerable to strategic manipulation. This paper explores the formal properties of VAF-MANIPULATION, proving that VAFs are inherently more susceptible to cheating than traditional voting systems. We demonstrate that while strategy-proofness is unattainable for fair aggregation rules, computational complexity can serve as a deterrent.
Introduction
The problem of collective decision-making in multi-agent systems often involves reconciling divergent value priorities. Value-Based Argumentation Frameworks (VAFs) provide a formal mechanism for this, where the validity of an attack between arguments depends on an audience's preference ranking of values. A critical challenge in this setting is the aggregation of individual preferences into a single collective ranking. This paper addresses the vulnerability of such aggregation processes to strategic manipulation, where agents misreport their values to influence the set of accepted arguments.
Theoretical Framework
The authors combine Dung’s abstract argumentation with social choice theory. In a VAF, arguments are associated with values, and an attack succeeds only if the attacker’s value is preferred over the victim’s.
Individual value rankings are aggregated using Preference Aggregation Functions to determine the collective grounded extension. The study applies the Gibbard-Satterthwaite theorem to analyze the strategy-proofness of these functions within the VAF context.
Analysis of Manipulation
The research reveals that VAFs are strictly more vulnerable to manipulation than standard voting. This "More Vulnerable" Paradox implies that even strategy-proof voting rules can be manipulated when used to aggregate audiences in argumentation. This vulnerability is exacerbated by the dichotomous nature of agent preferences, where the primary goal is the acceptance or rejection of a specific decisive argument.
Key Results
- Manipulability Inheritance: Any rule manipulable in voting is also manipulable in VAFs.
- Strategy-Proofness Failure: Strategy-proof voting rules do not guarantee strategy-proofness in VAFs, as shown by counter-examples in the paper.
- Complexity Barrier: The authors establish a formal reduction showing that the VAF-MANIPULATION problem is at least as hard as the strategic voting problem for the same aggregation rule. If the underlying voting rule is NP-hard to manipulate, so is the VAF.
- Dictatorship Necessity: Following the logic of the Gibbard-Satterthwaite theorem, complete prevention of manipulation requires a dictatorial aggregation rule, which is generally undesirable for collective systems.
Conclusion and Future Work
While theoretically vulnerable, the practical difficulty of manipulation may be mitigated by computational complexity and incomplete information. Future work should extend these findings to other argumentation semantics (such as preferred or stable semantics) and explore structural properties of graphs that might be naturally resistant to strategic behavior. Additionally, the concept of "minimal manipulation"—finding the smallest lie to achieve a goal—remains an open area for exploration.
Summary updated