AI2030 Responsible AI Tools Marketplace

Key Takeaways

Power indices offer a systematic framework for explaining AI decisions by quantifying how much each input contributes to an AI system's output. Originally developed in game theory to fairly distribute resources and analyze voting systems, these mathematical tools are now being adapted to make AI systems more transparent and accountable.
Choosing the right explanation method matters significantly for results. The paper demonstrates through concrete examples that different power indices can produce substantially different attributions for the same AI decision. This means practitioners need guidance on which method to use—a choice that currently lacks clear standards in the field.
The paper organizes power indices into three distinct categories (single-player, set-based, and cardinality-based) to help practitioners navigate options. This systematic organization addresses a gap in the literature by providing the first comprehensive guide for selecting appropriate explanation methods based on specific application contexts and fairness principles.
Some explanation methods preserve more information than others. The research shows that when moving to cardinality-based approaches, certain details about individual inputs are lost, though some mathematical distinctions between methods remain. This trade-off is important for practitioners to understand when deciding how granular their AI explanations need to be.
Formal principles matter for responsible AI. The paper provides mathematical proofs showing which power indices satisfy key fairness properties, establishing a rigorous foundation for choosing explanation methods that align with ethical principles like consistency and symmetry in how AI systems justify their decisions.

Abstract

Power indices, originating in cooperative game theory, quantify each player's influence on the outcome of a given game. Originally designed to distribute profits or costs among players and to analyse the fairness of voting systems, power indices have recently gained prominence as methods for attributing outputs of AI-based systems to inputs, thus facilitating explainability. However, selecting the appropriate power index for a given explanation task is an understudied problem. To address this, we organise power indices along three attribution dimensions: single-player, set-based, and cardinality-based. For each dimension, we review the corresponding power indices, generalise existing ones where applicable, and analyse which formal principles they satisfy. We provide proofs for properties that are missing in the literature and show that moving to the cardinality-based setting removes player-identity information while preserving some index-level distinctions. Using concrete examples, we illustrate how the choice of dimension and index affects the resulting attributions in practice, and offer guidance for practitioners seeking to select a suitable power index for a given application context.

Key Takeaways

Power indices offer a systematic framework for explaining AI decisions by quantifying how much each input contributes to an AI system's output. Originally developed in game theory to fairly distribute resources and analyze voting systems, these mathematical tools are now being adapted to make AI systems more transparent and accountable.
Choosing the right explanation method matters significantly for results. The paper demonstrates through concrete examples that different power indices can produce substantially different attributions for the same AI decision. This means practitioners need guidance on which method to use—a choice that currently lacks clear standards in the field.
The paper organizes power indices into three distinct categories (single-player, set-based, and cardinality-based) to help practitioners navigate options. This systematic organization addresses a gap in the literature by providing the first comprehensive guide for selecting appropriate explanation methods based on specific application contexts and fairness principles.
Some explanation methods preserve more information than others. The research shows that when moving to cardinality-based approaches, certain details about individual inputs are lost, though some mathematical distinctions between methods remain. This trade-off is important for practitioners to understand when deciding how granular their AI explanations need to be.
Formal principles matter for responsible AI. The paper provides mathematical proofs showing which power indices satisfy key fairness properties, establishing a rigorous foundation for choosing explanation methods that align with ethical principles like consistency and symmetry in how AI systems justify their decisions.

Abstract

Dimensions of Power: A Systematic Guide to Power Indices for Explainable AI

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links

Dimensions of Power: A Systematic Guide to Power Indices for Explainable AI

Key Takeaways

Abstract

Paper Info

Responsible AI Topics

Links