Abstract
Abstract Argumentation (AA) frameworks typically rely on the principle of admissibility, requiring a set of arguments to defend itself against all attackers. However, this classical approach often fails when faced with self-defeating arguments—"zombies" that are logically invalid yet still capable of excluding reasonable arguments. This paper introduces weak admissibility and weak defense, foundational concepts that allow argumentation systems to disregard such unreasonable attackers. By redefining defense through a recursive reduct-based approach, the authors establish a new family of semantics—weak complete, preferred, and grounded—that are more intuitive, insensitive to self-loops, and fully compatible with Dung’s original framework.
The "Zombie" Argument Problem
In Phan Minh Dung’s seminal 1995 work on Abstract Argumentation, the gold standard for an acceptable set of arguments is admissibility: a set must be conflict-free and able to defend itself against every possible attacker. While mathematically elegant, this "all-or-nothing" defense leads to a conceptual glitch when dealing with self-contradictory arguments.
Consider a scenario where an argument $a$ attacks itself ($a \to a$) and also attacks argument $b$ ($a \to b$). In classical AA, $b$ can never be part of an admissible set. Because $a$ is the only attacker of $a$, and $a$ cannot be part of any admissible set due to its self-attack, $b$ remains undefended and thus excluded.
The authors argue that $a$ is a "zombie"—it is self-contradictory and effectively "dead" on arrival. There is no logical reason why a reasonable argument like $b$ should be discarded simply because a "dead" argument attacks it. This paper seeks to rectify this by weakening the requirement for defense.
Methodology: Weak Admissibility and the E-reduct
The authors move away from Dung's absolute defense and introduce a recursive, reduct-based approach to evaluate the "seriousness" of an attacker. Instead of requiring a counter-attack for every threat, they introduce the E-reduct ($F^E$).
To evaluate a set $E$, the framework is reduced to a sub-graph containing only arguments whose status is still "undecided"—specifically, those that are not in $E$ and are not already defeated by $E$.
Based on this reduct, the authors define two core concepts:
- Weak Admissibility: A set $E$ is weakly admissible if it is conflict-free and any argument $y$ attacking $E$ is "not serious." An attacker is deemed "not serious" if it cannot be part of any weakly admissible set within the reduct.
- Weak Defense: A set $E$ weakly defends an argument $x$ if it either counter-attacks the attacker (classical defense) or if the attacker is not "trustworthy" based on the recursive evaluation of the reduct.
Using these foundations, the authors redefine the standard semantics of AA, creating Weak Complete ($co^w$), Weak Preferred ($pr^w$), and Weak Grounded ($gr^w$) semantics.
Key Insights and Results
The transition to weak admissibility yields several significant theoretical advantages:
- Semantic Insensitivity to Self-Loops: The new semantics are "insensitive" to self-attacking arguments. In classical AA, adding or deleting a self-attack $(a, a)$ can radically change which other arguments are accepted. In this framework, the "zombie" status of $a$ is recognized regardless of the specific syntax of the self-attack.
- Generalization of Dung’s Framework: Weak admissibility is a true generalization. Every classically admissible set is also weakly admissible, ensuring that the new system expands upon rather than replaces existing logic.
- Universal Definedness: For any finite argumentation framework, a weakly admissible set (at least the empty set) is guaranteed to exist, ensuring the system is computationally stable.
- The Grounded Core: While weak grounded extensions are no longer necessarily unique, they all share a "common core" that includes the classical grounded extension.
Conclusion and Future Work
By redefining the foundations of defense, this work provides a more robust logic for handling self-contradictory information in automated reasoning. While the current study focuses on finite frameworks, future work remains to explore the computational complexity of these semantics and their extension to infinite frameworks. Additionally, integrating these concepts into existing argumentation solvers could significantly improve the reliability of AI systems in legal, medical, and debate-centric applications.
Summary updated