Action-Based Alternating Transition Systems for Arguments about Action

Katie Atkinson, Trevor J. M. Bench-CaponAAAI 2007

Abstract

This paper formalizes practical reasoning in multi-agent systems by mapping informal argument schemes onto Action-Based Alternating Transition Systems (AATS). By incorporating social values and joint actions into the transition model, the authors provide a rigorous mathematical foundation for agents to generate, critique, and defend proposals for action. The framework identifies 17 critical questions that serve as formal attacks against proposed actions, bridging the gap between qualitative argumentation and computational multi-agent modeling.

Introduction

In artificial intelligence, "practical reasoning" refers to the process of deciding which action to take to achieve a specific goal. Traditionally, this has been addressed through planning or logic-based approaches. However, real-world reasoning is inherently defeasible—where new information can overturn previous conclusions—and subjective, as different agents often prioritize different social values.

A significant challenge in multi-agent systems is that the outcome of an individual agent's action often depends on the simultaneous choices of others. While existing argumentation frameworks utilized "argument schemes" (e.g., "In circumstances R, perform action A to achieve goal G"), these were frequently ad hoc and lacked a mathematical basis to verify their validity within a dynamic, multi-agent environment.

Methodology: The AATS Framework

The authors address these challenges by building upon Action-Based Alternating Transition Systems (AATS). The framework consists of several key components:

  1. The Formal Model: An AATS defines a set of states and agents. Transitions between states are driven by joint actions, representing the combination of actions taken by all agents simultaneously.
  2. Social Values: The standard AATS is extended by labeling transitions with social values. A transition can promote, demote, or remain neutral toward specific values (e.g., "efficiency" vs. "safety").
  3. Formalized Argument Schemes: Informal linguistic schemes are translated into formal structures. An argument is defined as: "In state $q_x$, agent $i$ should perform action $\alpha$ because it is part of a joint action leading to state $q_y$, which promotes value $V$."
Figure 1: A complex AATS state-transition diagram illustrating the interplay between states, joint actions (j), and value labels (v).

The core contribution is the formalization of 17 Critical Questions (CQs). These act as mathematical checks to attack a proposal, categorized into epistemic concerns (e.g., "Are we in the assumed state?"), action selection (e.g., "Does this demote a value?"), and multi-agent coordination (e.g., "Will the other agent perform their part of the joint action?").

Figure 2: A simplified AATS model, demonstrating how agents might reason over a subset of the state space or from a specific perspective.

Key Insights

The paper conceptualizes practical reasoning as a three-stage process:

  • Problem Formulation: Establishing the shared "map" or AATS structure.
  • Epistemic Reasoning: Determining the current state within that map.
  • Action Selection: Debating the optimal path based on value preferences.

By centering joint actions, the formalism forces agents to consider the "grand coalition" of all participants' choices. Furthermore, subjectivity is formalized: agents may disagree due to differing world models (AATS structures), differing beliefs about the current state, or differing value hierarchies.

Results and Conclusion

The primary result is a mathematical mapping of 17 informal linguistic challenges to specific AATS properties. This transforms qualitative argumentation into a computational problem suitable for autonomous agents. The framework allows agents to systematically generate attacks against proposals, ensuring that selected actions are robust against criticism.

While the framework provides a rigorous foundation, limitations remain regarding computational complexity as state spaces grow. Future work involves addressing the state-space explosion and developing algorithms for the evolution of value rankings during negotiation.

Summary updated

BibTeX
@inproceedings{atkinson2007action,
  author = {Katie Atkinson and Trevor J. M. Bench-Capon},
  title = {{Action-Based Alternating Transition Systems for Arguments about Action}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2007},
}
Figure