AI2030 Responsible AI Tools Marketplace

Key Takeaways

Explainability goes beyond feature importance: Traditional XAI methods tell you which features matter, but this paper's approach explains why predictions happen by uncovering causal relationships between variables—moving from "this feature was used" to "this feature caused this outcome."
Causality + argumentation creates richer explanations: The method combines causal discovery with an "argumentation framework" that maps how features support or oppose each other, similar to how arguments in a debate can strengthen or weaken one another, providing more nuanced reasoning for model decisions.
Practical for real-world accountability: By translating model logic into causal arguments with supporting and opposing factors, this approach could help organizations better explain decisions to stakeholders, regulators, or affected individuals—particularly important for high-stakes domains like lending, hiring, or healthcare.
Tested on real data: The authors validated their method on benchmark datasets and compared it against existing explainability approaches, suggesting the technique is grounded in empirical evaluation rather than theory alone.
Addresses a genuine gap in current XAI: Since many deployed models use post-hoc explanation methods that can be misleading about causation, this work tackles the important problem that current explainability tools often obscure why decisions are actually made.

Abstract

Explainable AI (XAI) methods identify which features are relevant to a model's predictions but often fail to clarify why certain decisions are made. In this work, we present a novel method that integrates causality with argument-based reasoning to explain why models may be making predictions. Our approach first identifies causal relationships among variables using causal discovery methods and then translates these into a Bipolar Argumentation Framework (BAF) to represent supportive and opposing interactions among features. By using semi-stable semantics, we find extensions of features that explain why certain outcomes may have been chosen. We demonstrate our method on two benchmark datasets and compare its results against standard post-hoc explainability approaches.

To be published in The 4th World Conference on eXplainable Artificial Intelligence

Key Takeaways

Explainability goes beyond feature importance: Traditional XAI methods tell you which features matter, but this paper's approach explains why predictions happen by uncovering causal relationships between variables—moving from "this feature was used" to "this feature caused this outcome."
Causality + argumentation creates richer explanations: The method combines causal discovery with an "argumentation framework" that maps how features support or oppose each other, similar to how arguments in a debate can strengthen or weaken one another, providing more nuanced reasoning for model decisions.
Practical for real-world accountability: By translating model logic into causal arguments with supporting and opposing factors, this approach could help organizations better explain decisions to stakeholders, regulators, or affected individuals—particularly important for high-stakes domains like lending, hiring, or healthcare.
Tested on real data: The authors validated their method on benchmark datasets and compared it against existing explainability approaches, suggesting the technique is grounded in empirical evaluation rather than theory alone.
Addresses a genuine gap in current XAI: Since many deployed models use post-hoc explanation methods that can be misleading about causation, this work tackles the important problem that current explainability tools often obscure why decisions are actually made.

Abstract

To be published in The 4th World Conference on eXplainable Artificial Intelligence

A Causal Argumentation Method for Explainability of Machine Learning Models

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

A Causal Argumentation Method for Explainability of Machine Learning Models

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links