DAMN: Defeasible Reasoning Tool for Multi-Agent Reasoning

Abdelraouf Hecham, Madalina Croitoru, Pierre BisquertAAAI 2020 · DOI

Abstract

DAMN (Defeasible Reasoning Tool for Multi-Agent Reasoning) is a web-based platform designed to facilitate collaborative decision-making among multiple stakeholders by merging knowledge bases and resolving logical conflicts. Utilizing defeasible reasoning—a logic framework that accommodates exceptions—DAMN visualizes reasoning processes through Statement Graphs. This paper describes the tool's architecture, its ability to handle incomplete or contradictory information in real-time, and its application in real-world scenarios like the EU H2020 NoAW project.

Introduction

In collaborative decision-making, experts often bring conflicting information to the table. Traditional "hard" logic systems are fragile in these environments; a single contradiction can cause the entire system to fail. Defeasible reasoning addresses this by allowing for "rules of thumb" that can be defeated by more specific information (e.g., "Birds generally fly, but penguins are birds that don't fly"). While theoretically robust, existing tools for this logic often lack collaborative features or intuitive visualizations. DAMN fills this gap by providing a multi-user, real-time environment where agents can visualize conflicts and observe how different logical semantics resolve ambiguity.

System Architecture

DAMN is built on a modern web stack to support real-time interaction between multiple agents. The front-end is developed using Angular, providing each user with a visualization module and access to their specific knowledge base. The back-end is split into a Collaboration Server, utilizing Java Spring and WebSockets for two-way communication, and a Reasoning Server. The Reasoning Server leverages the GRAAL API for query answering and implements the Statement Graph framework to handle complex logical features like "team defeat" and "ambiguity propagation."

The DAMN system architecture showing the interaction between front-end UI applications and back-end reasoning/collaboration servers.

Visualizing Logic through Statement Graphs

A core innovation of DAMN is the use of Statement Graphs (SG) to turn abstract logical rules into visual maps. This approach significantly improves explainability by allowing users to see exactly why a conclusion was reached. Nodes are color-coded to indicate their status—such as "Accepted Strictly," "Accepted Generally," or "Rejected Strictly"—providing immediate visual feedback on the state of the knowledge base. This flexibility allows logicians and domain experts alike to experiment with different reasoning semantics and observe how they impact the final graph.

A Statement Graph demonstrating the "Kowalski the penguin" logic puzzle, visualizing how conflicting rules about flight are resolved.

Evaluation and Key Results

The tool's utility was demonstrated through the EU H2020 NoAW project, which involved 21 experts collaborating on wine production waste management. DAMN successfully identified a critical conflict where a winemaker's suggestion was based on a law that had been superseded. By highlighting these friction points, the tool serves as a catalyst for human discussion rather than just an automated solver. Key results include:

  • Real-time Interaction: Successful support for multiple agents modifying a shared knowledge base simultaneously.
  • Conflict Detection: The ability to pinpoint specific logical contradictions arising from disparate expert inputs.
  • Semantic Flexibility: Users can toggle between different reasoning modes to see how ambiguity is handled across the system.

Limitations and Future Work

While effective for small to medium groups, the scalability of the visualization remains a challenge; large rule sets can lead to complex "hairball" graphs that are difficult to interpret. Future development will focus on improving UI clarity for non-logicians and implementing automated knowledge extraction from the Semantic Web to reduce the need for manual data entry via JSON or DLGP formats.

Summary updated

BibTeX
@inproceedings{hecham2020damn,
  author = {Abdelraouf Hecham and Madalina Croitoru and Pierre Bisquert},
  title = {{DAMN: Defeasible Reasoning Tool for Multi-Agent Reasoning}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2020},
  doi = {10.1609/aaai.v34i09.7093},
}
Figure