Using Arguments for Making Decisions: A Possibilistic Logic Approach

Leila Amgoud, Henri PradeUAI 2004

TL;DR

This paper introduces a novel framework for decision-making that leverages argumentation and possibilistic logic. It formalizes how humans use arguments (pros and cons) to evaluate choices, especially under uncertainty, by building arguments from knowledge bases and prioritized goals. The approach offers both pessimistic and optimistic decision-making strategies, handling potential inconsistencies in knowledge and goals.

What Problem Does This Paper Address?

Humans naturally use arguments to justify their decisions and evaluate potential choices. Each choice often comes with a set of advantages and disadvantages, each with varying degrees of strength. While this intuitive approach is common, formalizing it within artificial intelligence has been challenging. Existing methods often lack a robust logical foundation for handling uncertainty and inconsistency, or they rely on empirical calculations rather than principled reasoning. This paper aims to bridge this gap by proposing a formal, logic-based framework for decision-making that directly models the process of argumentation, incorporating uncertainty and prioritized goals.

Methods

The core of this work lies in a possibilistic logic framework for decision-making. The authors define two key components:

  • Knowledge Base (K): A set of propositions with associated certainty levels (e.g., "it is raining" with certainty 0.8). This is represented as a possibilistic logic base, where (kj, pj) means the necessity of kj is at least pj.
  • Goal Base (G): A set of prioritized goals, where (gi, λi) means goal gi has priority λi.

From these bases, the framework constructs arguments for and against potential decisions.

  • Arguments PRO (for a decision): These are triples <S, C, d>, where S is a set of knowledge propositions supporting the decision d, and C is a set of satisfied goals. The strength of these arguments is measured by their level (minimum certainty of knowledge used) and weight (degree of satisfaction of goals).
  • Arguments CON (against a decision): These are triples <S, C, d>, where S is knowledge supporting the negation of a goal, and C represents the unsatisfied goals. Their weakness is measured by their level (certainty of knowledge used) and degree (priority of unsatisfied goals).

The paper then defines how to compare arguments and, consequently, how to compare decisions based on these arguments. Two main decision criteria are proposed:

  • Pessimistic Criterion: This approach focuses on the worst-case scenario, evaluating decisions based on the minimum strength of their supporting arguments. It's risk-averse, prioritizing decisions where even the least favorable outcomes are acceptable. This is formalized using a pessimistic qualitative utility E*(d).
  • Optimistic Criterion: This approach looks for the best-case scenario, evaluating decisions based on the maximum strength of their supporting arguments. It's risk-seeking, favoring decisions where at least one highly favorable outcome is possible. This is formalized using an optimistic qualitative utility E*(d).

The framework also outlines how to handle inconsistent knowledge or goal bases, extending the argumentation process to manage conflicts and determine acceptable arguments and decisions.

Datasets

Not discussed in the paper. The paper uses illustrative examples rather than formal datasets.

Key Insights and Ideas

  • Formalizing Human Argumentation: The paper provides a principled logical framework for how humans use pros and cons to make decisions, moving beyond ad-hoc reasoning.
  • Possibilistic Logic for Uncertainty: By using possibilistic logic, the framework effectively models and reasons with uncertain knowledge and preferences, which is crucial for real-world decision-making.
  • Pessimistic vs. Optimistic Attitudes: The explicit distinction between pessimistic (risk-averse) and optimistic (risk-seeking) decision-making strategies is a significant contribution, allowing for flexible decision-making tailored to user preferences.
  • Argument Strength as a Dual Measure: Defining argument strength by both the certainty of the supporting knowledge (level) and the degree of goal satisfaction (weight) provides a nuanced way to compare arguments.
  • Handling Inconsistency: The extension to handle inconsistent knowledge and goal bases is vital, as real-world scenarios rarely present perfectly consistent information.

Key Results

The paper presents theoretical results and definitions rather than empirical quantitative results. Key outcomes include:

  • Definitions of Arguments PRO and CON: Formal definitions for constructing arguments supporting and opposing decisions.
  • Strength/Weakness Measures: Definitions for quantifying the strength of arguments PRO and the weakness of arguments CON, based on knowledge certainty and goal satisfaction.
  • Argument and Decision Comparison: Criteria for comparing arguments and subsequently comparing decisions based on their associated arguments.
  • Theorems relating argument strength to qualitative utility: Theorem 1 and Corollary 1 link the pessimistic utility to the strength of supporting arguments, while Theorem 2 and Corollary 2 do the same for the optimistic utility.
  • Framework for Inconsistent Bases: Definitions for argumentation systems that can handle conflicting arguments arising from inconsistent knowledge or goals, leading to the identification of rejected, acceptable, and abeyant arguments and decisions.

Limitations and Future Work

  • Computational Complexity: While the paper outlines the logical framework, the computational complexity of constructing and comparing arguments, especially in the case of inconsistent bases, is not deeply explored.
  • Refining Argument Strength: The authors suggest refining argument strength measures by using vector-valued strengths instead of scalar ones to better capture multiple weak points in an argument.
  • Extension to Multi-Criteria Decision Making: The paper notes a planned extension of the approach to multi-criteria decision making, leveraging the relationship between decision under uncertainty and multi-criteria problems.
  • Practical Implementation: The paper focuses on the theoretical framework; further work would be needed for practical implementation and evaluation on real-world problems.

Summary updated

BibTeX
@inproceedings{amgoud2004using,
  author = {Leila Amgoud and Henri Prade},
  title = {{Using Arguments for Making Decisions: A Possibilistic Logic Approach}},
  booktitle = {Conference on Uncertainty in Artificial Intelligence},
  year = {2004},
}
Figure