Maintenance of Social Commitments in Multiagent Systems

Pankaj R. Telang, Munindar P. Singh, Neil Yorke-SmithAAAI 2021 · DOI

Introduction

In the field of Multiagent Systems (MAS), social commitments serve as the primary mechanism for coordinating autonomous agents. Traditionally, research has focused on achievement commitments, where a debtor agent promises a creditor that a specific state will be reached (e.g., "I will deliver the package"). However, many real-world interactions are centered on maintenance: sustaining a condition over time rather than reaching a one-off goal. Examples include an airline operator requiring an engine manufacturer to keep an aircraft "fit to fly" or a gardener committing to keeping a lawn green.

Treating maintenance as a special case of achievement (using temporal logic like "always $p$") often fails to capture the nuances of real-world scenarios. If a maintained condition is temporarily violated, a simple "always" condition is permanently failed. In practice, maintenance involves a reactive loop: if the condition becomes false, the agent must act to re-establish it. This paper, presented at AAAI-21, introduces "maintenance commitments" as a first-class construct, providing the formal machinery to handle these "negated and resurrected" states.

Formalizing Maintenance Commitments

The authors define a maintenance commitment as a tuple $\langle x, y, l, m, d \rangle$, where agent $x$ (the debtor) commits to agent $y$ (the creditor) that if an antecedent $l$ holds, they will sustain condition $m$ until a discharge condition $d$ is met.

To operationalize this, the paper introduces formal life cycle models. Unlike achievement commitments, maintenance commitments must account for ongoing monitoring and the potential for temporary failures that do not necessarily terminate the commitment.

The life cycle of a maintenance commitment, illustrating the transitions between Conditional, Detached, and Sustain states.

As shown in the life cycle diagram, a commitment can move between being Detached (triggered by the antecedent) and Sustain (where the condition $m$ is actively maintained). This structure allows for a "consequent failure" to trigger corrective action rather than an immediate, irreversible violation.

Practical Rules and Coherence

A key contribution of the work is the development of "practical rules"—if-then templates that guide a rational agent's behavior. These rules ensure coherence, defined as the alignment between an agent's internal cognitive state (its own goals) and its social state (its commitments to others). For instance, if a maintenance commitment is detached, the rules dictate that the agent should adopt internal achievement goals to restore the condition if it fails.

The authors prove through mathematical convergence theorems that if agents follow these prescribed rules, an initially incoherent system will eventually reach a coherent state and maintain it over time, even as environmental conditions fluctuate.

Case Study: Aerospace Aftermarket

The framework is evaluated through a qualitative case study of the "Aerospace Aftermarket" scenario. This scenario involves complex coordination between an airline operator, an engine manufacturer, and a parts supplier to ensure engine health and operational readiness.

The Aerospace Aftermarket scenario, demonstrating the coordination between Operator, Engine Manufacturer, and Parts Manufacturer.

In this model, the Engine Manufacturer has a maintenance commitment to the Operator to keep the engines functional. When monitoring indicates a potential issue, the manufacturer requests maintenance (an unscheduled event), leading to a sequence of actions—removing the engine, refurbishing it, and supplying parts—to restore the "fit to fly" condition. This case study demonstrates how the formal life cycles and practical rules map to complex, multi-party industrial processes.

Conclusion and Future Work

This research fills a critical gap in MAS by providing a rigorous foundation for maintenance commitments. By distinguishing maintenance from achievement and providing operational semantics for reactive restoration, the authors enable more resilient and realistic agent coordination.

Future work includes extending the framework from propositional to first-order logic to handle quantified commitments (e.g., maintaining an entire fleet) and incorporating explicit temporal intervals. Additionally, addressing "disputes" where agents disagree on the state of a maintained condition remains an open challenge for building robust normative systems.

Summary updated

BibTeX
@inproceedings{telang2021maintenance,
  author = {Pankaj R. Telang and Munindar P. Singh and Neil Yorke-Smith},
  title = {{Maintenance of Social Commitments in Multiagent Systems}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/aaai.v35i13.17355},
}
Figure