Reasoning about the Appropriateness of Proponents for Arguments

Anthony HunterAAAI 2008

Abstract

In classical formal argumentation, the focus is typically on the internal logical consistency of an argument. However, in real-world discourse, the credibility and expertise of the proponent are often as critical as the logic itself. This paper proposes a meta-level logic framework that formalizes reasoning about the appropriateness of proponents. By integrating metadata regarding expertise, trustworthiness, and roles, the system distinguishes between logically sound arguments and those that should be accepted based on the credibility of the source.

Introduction

The evaluation of arguments in formal systems usually occurs at the "object-level," where the primary concern is whether premises logically entail a claim and if counterarguments exist. Yet, in many domains—such as medicine or law—laypeople must rely on experts because they lack the specialized knowledge to verify arguments directly. Standard logic systems struggle here because they treat all sources as equal; a logically sound argument from an unreliable source is given the same weight as one from a recognized expert.

Anthony Hunter addresses this by introducing a Meta-level Argumentation framework. This approach allows for reasoning about the argumentation process, enabling the system to handle "source-based" attacks (e.g., "the proponent lacks expertise") alongside traditional "content-based" attacks.

Meta-level Argumentation Framework

The framework distinguishes between the object-level, containing the actual arguments and claims, and the meta-level, which uses separate logical axioms to reason about those arguments.

Meta-Predicates

The system utilizes specific predicates to manage proponent metadata:

  • ast(p, a): Proponent p asserted argument a.
  • app(p, a): Proponent p is appropriate for argument a.
  • acc(a): Argument a is acceptable.

Axiomatizing Appropriateness

The core of the method is a "metabase" of rules that define appropriateness. For instance, a proponent is considered appropriate if the argument's topic falls within their professional scope (Expertise) or if they lack conflicts of interest (Trustworthiness). The framework also supports an outrank predicate, allowing the system to prefer arguments from more qualified experts, such as a specialist over a general practitioner.

Key Insights: The Gatekeeper Predicate

The app(p, a) predicate acts as a logical gatekeeper. An argument is only "warranted" if its proponent is deemed appropriate at the meta-level. This allows for "meta-level undercuts," where an argument is defeated by proving the proponent is unqualified or untrustworthy, without needing to challenge the argument's internal logic. This formalizes Douglas Walton’s heuristic-based "Arguments from Expert Opinion" into a rigorous mathematical structure.

Results and Filtering Mechanisms

The paper provides mathematical proofs (Propositions 1 and 2) showing that meta-level trees are isomorphic to standard argument trees, ensuring backward compatibility. Proposition 3 links "warranted" arguments to "good expert arguments," asserting that an argument is only warranted if it is proposed by a qualified, confident proponent and lacks superior competing expert arguments.

To implement this, the author defines a filtering mechanism $(Q, C, B, S)$ that categorizes arguments based on whether the proponent is Qualified ($Q$), Confident ($C$), provides the Best argument ($B$), or a Safe one ($S$).

Table illustrating the filtering mechanism (Q, C, B, S) for different argument sets.

As shown in the table above, the filters allow for a granular evaluation of argument sets ($E_i$) based on the meta-level criteria. For example, an argument might be categorized as "Confident" but fail the "Qualified" filter if the proponent's credentials do not match the subject matter.

Conclusion

This meta-level framework provides a robust method for incorporating proponent credibility into formal argumentation. While it currently relies on a pre-defined "metabase" of facts and classical logic, it lays the groundwork for more sophisticated AI systems that can reason about trust and expertise in complex, real-world environments. Future work may explore the use of defeasible logic to better handle the fluid nature of these meta-level assessments.

Summary updated

BibTeX
@inproceedings{hunter2008reasoning,
  author = {Anthony Hunter},
  title = {{Reasoning about the Appropriateness of Proponents for Arguments}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2008},
}
Figure