Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

An Actionability Assessment Tool for Explainable AI

Ronal Singh, Tim Miller, Liz Sonenberg, Eduardo Velloso, Frank Vetere, Piers Howe, Paul Dourish

TL;DR

Actionability in explainable AI is ill-defined and often relies on researchers' intuition. The authors propose a seven-question Actionability Assessment Tool, drawing on instruments from patient education, management research, SDM, and cybersecurity, and validate it with two user studies in credit scoring and employee turnover using prototypical, counterfactual, and directive explanations. Results show the tool can discriminate actionability across explanations and that context shapes assessments, with directive explanations rated most actionable. The work advances human-centered evaluation for actionable explainability and recourse, while highlighting the need for domain adaptation and validation with real-world user behavior.

Abstract

In this paper, we introduce and evaluate a tool for researchers and practitioners to assess the actionability of information provided to users to support algorithmic recourse. While there are clear benefits of recourse from the user's perspective, the notion of actionability in explainable AI research remains vague, and claims of `actionable' explainability techniques are based on the researchers' intuition. Inspired by definitions and instruments for assessing actionability in other domains, we construct a seven-question tool and evaluate its effectiveness through two user studies. We show that the tool discriminates actionability across explanation types and that the distinctions align with human judgements. We show the impact of context on actionability assessments, suggesting that domain-specific tool adaptations may foster more human-centred algorithmic systems. This is a significant step forward for research and practices into actionable explainability and algorithmic recourse, providing the first clear human-centred definition and tool for assessing actionability in explainable AI.

An Actionability Assessment Tool for Explainable AI

TL;DR

Actionability in explainable AI is ill-defined and often relies on researchers' intuition. The authors propose a seven-question Actionability Assessment Tool, drawing on instruments from patient education, management research, SDM, and cybersecurity, and validate it with two user studies in credit scoring and employee turnover using prototypical, counterfactual, and directive explanations. Results show the tool can discriminate actionability across explanations and that context shapes assessments, with directive explanations rated most actionable. The work advances human-centered evaluation for actionable explainability and recourse, while highlighting the need for domain adaptation and validation with real-world user behavior.

Abstract

In this paper, we introduce and evaluate a tool for researchers and practitioners to assess the actionability of information provided to users to support algorithmic recourse. While there are clear benefits of recourse from the user's perspective, the notion of actionability in explainable AI research remains vague, and claims of `actionable' explainability techniques are based on the researchers' intuition. Inspired by definitions and instruments for assessing actionability in other domains, we construct a seven-question tool and evaluate its effectiveness through two user studies. We show that the tool discriminates actionability across explanation types and that the distinctions align with human judgements. We show the impact of context on actionability assessments, suggesting that domain-specific tool adaptations may foster more human-centred algorithmic systems. This is a significant step forward for research and practices into actionable explainability and algorithmic recourse, providing the first clear human-centred definition and tool for assessing actionability in explainable AI.
Paper Structure (30 sections, 1 equation, 4 figures, 3 tables)

This paper contains 30 sections, 1 equation, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Counterfactual, prototypical and directive explanations
  4. Actionability instruments
  5. The patient education materials assessment tool
  6. Actionability of management research
  7. Relevant instruments from shared decision-making literature
  8. Perceived Actionability of Security Advice Available on the Web
  9. Actionability Assessment Tool
  10. Clarity of information
  11. Decision understanding
  12. Personalisation/relevance/contextualisation
  13. Correction of misunderstandings
  14. Action
  15. Study Design
  16. ...and 15 more sections

Figures (4)

  • Figure 1: Study procedure: We run the two studies in parallel to allocate different sets of participants to the two studies.
  • Figure 2: Results of pairwise comparison between explanation types. The y-axis is the number of times each explanation type was perceived to be more actionable (d = directive; c = counterfactual; and p = prototype).
  • Figure 3: Ratings for explanations in the credit domain. Explanation types: d = directive; c = counterfactual; p = prototype.
  • Figure 4: Ratings for the explanation types in the employee turnover domain. Explanation types: d = directive; c = counterfactual; p = prototype.