Explaining preferences with argument positions

Sylvie Doutre, Trevor J. M. Bench-Capon, Paul E. DunneIJCAI 2005

Abstract

This paper presents a formal framework for practical reasoning where value preferences are not assumed a priori but emerge through the process of constructing a defensible argument. By introducing the DOR-partitioned value-based argument framework (DOR-VAF), the authors demonstrate how agents can discover their own priorities by partitioning arguments into desired, optional, and rejected sets. This approach bridges the gap between rigid abstract argumentation and prescriptive value-based models, offering a more realistic representation of human deliberation in fields such as law and politics.

Introduction

In classical logic, reasoning is primarily concerned with determining what is true. However, "practical reasoning"—the logic of determining what we should do—is inherently subjective because it depends on an individual's values. Existing models, such as Bench-Capon’s Value-based Argumentation Frameworks (VAFs), allow for rational disagreement by ranking values (e.g., prioritizing privacy over security).

Despite their utility, these models typically assume that an agent possesses a fixed hierarchy of values before a debate begins. The authors argue that this is often unrealistic. In real-world scenarios, such as a politician forming a platform or a lawyer building a case, agents often lack a pre-defined value system. Instead, they have specific goals they wish to achieve and certain outcomes they wish to avoid. Their preferences are a consequence of trying to make their desired goals defensible against opposition.

The DOR-VAF Framework

To address this, the authors propose an extension to the VAF called the DOR-partitioned value-based argument framework (DOR-VAF). This framework shifts the focus from pre-existing values to the construction of a "position" through three key components:

  1. The DOR Partition: Arguments are categorized into three distinct sets:
  2. D (Desired): Arguments the agent must include in their final position.
  3. O (Optional): Arguments the agent may use if they assist in defending the desired set.
  4. R (Rejected): Arguments the agent refuses to accept under any circumstances.
  5. Emergent Audiences: Rather than requiring a value ranking as input, the framework identifies "audiences" (specific value rankings) that make the set of desired arguments defensible. Preferences are discovered, not dictated.
  6. Dialogue Construction: The process of building a position is modeled as a game between a Proponent and an Opponent. The Proponent defends Desired arguments by strategically incorporating Optional arguments or by "committing" to specific value preferences to nullify attacks.

The Hierarchy of Commitment

A central insight of the paper is that value preferences act as constraints that are only activated when necessary. When an opponent attacks a position, the proponent follows a hierarchy of responses, ordered from the least to the most "expensive" in terms of commitment:

  • Level 1: Use an optional argument that requires no new value preferences.
  • Level 2: State a new value preference to defeat the attack directly.
  • Level 3: Use an optional argument that requires a new value preference (without conflicting with existing ones).
  • Level 4: Use an optional argument that creates a conflict, requiring additional value preferences to resolve.

This hierarchy captures the "room for maneuver" essential to human reasoning; a skilled debater avoids committing to a rigid value system until the pressure of the argument forces their hand.

Conclusion and Formal Results

The paper provides a formal definition for a successful "position" within a DOR-VAF. A position is considered successful if there exists at least one audience where the set is conflict-free, every argument is defended, and all optional arguments included serve the specific purpose of defending a desired argument.

While the framework successfully bridges the gap between Dung’s abstract argumentation and Bench-Capon’s VAF, the authors acknowledge several areas for future research. These include the computational complexity of resolving high-level conflicts, the transition from static to dynamic sets of arguments (paradigm shifts), and the eventual development of software implementations to assist human reasoners in navigating complex DOR-VAFs.

Summary updated

BibTeX
@inproceedings{doutre2005explaining,
  author = {Sylvie Doutre and Trevor J. M. Bench-Capon and Paul E. Dunne},
  title = {{Explaining preferences with argument positions}},
  booktitle = {International Joint Conference on Artificial Intelligence},
  year = {2005},
}
Figure