Aggregative Semantics for Quantitative Bipolar Argumentation Frameworks

A team of researchers including Yann Munro, Isabelle Bloch, and Marie-Jeanne Lesot has published a new paper titled 'Aggregative Semantics for Quantitative Bipolar Argumentation Frameworks' on arXiv. The work tackles a core challenge in AI reasoning: how to formally model and weigh potentially conflicting pieces of information, known as arguments, within a structured framework. Their focus is on Quantitative Bipolar Argumentation Frameworks (QBAFs), where arguments have intrinsic strengths and can either attack or support one another. The novel contribution is a family of 'aggregative semantics' designed to handle situations where attackers and supporters do not play a symmetric role, a limitation of previous 'modular semantics' approaches.

The proposed method decomposes the computation into three distinct, interpretable steps. First, it calculates a global weight for all attacking arguments. Second, it computes a separate global weight for all supporting arguments. Finally, these two aggregated values are combined with the argument's own intrinsic weight to determine its final acceptability. This separation maintains the 'bipolarity' of the framework—the distinct treatment of attack and support—for longer in the reasoning process, leading to more nuanced and understandable outcomes. The authors discuss the properties required for the aggregation functions and illustrate the framework's flexibility by testing 500 different aggregative semantics on a final example, showcasing a wide range of possible, configurable behaviors for AI systems.