Explainability fact sheets: a framework for systematic assessment of explainable approaches

By Kacper Sokol, Peter A. Flach. FAT*, 2019

361 citations recorded by the source corpus.

Explanations in Machine Learning come in many forms, but a consensus regarding their desired properties is yet to emerge. In this paper we introduce a taxonomy and a set of descriptors that can be used to characterise and systematically assess explainable systems along five key dimensions: functional, operational, usability, safety and validation. In order to design a comprehensive and representative taxonomy and associated descriptors we surveyed the eXplainable Artificial Intelligence literature, extracting the criteria and desiderata that other authors have proposed or implicitly used in their research. The survey includes papers introducing new explainability algorithms to see what criteria are used to guide their development and how these algorithms are evaluated, as well as papers proposing such criteria from both computer science and social science perspectives. This novel framework allows to systematically compare and contrast explainability approaches, not just to better understand their capabilities but also to identify discrepancies between their theoretical qualities and properties of their implementations. We developed an operationalisation of the framework in the form of Explainability Fact Sheets, which enable researchers and practitioners alike to quickly grasp capabilities and limitations of a particular explainable method. When used as a Work Sheet, our taxonomy can guide the development of new explainability approaches by aiding in their critical evaluation along the five proposed dimensions.

Subfields: Artificial Intelligence, Safety Research

View Paper: https://www.semanticscholar.org/paper/c021955e06618511f58df4e41882d7c1ac96a459

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Safety Research

Citations in the corpus

The Dataset Nutrition Label (2nd Gen): Leveraging Context to Mitigate Harms in Artificial Intelligence (2022, background): “Finally, there are several related projects in the dataset transparency space, including the manually-generated Datasheets for Datasets work which our project leverages [9], data statements [30], and several automated tools [31, 23].”
Teach Me to Explain: A Review of Datasets for Explainable NLP (2021, neutral): “While explainability fact sheets have been proposed for models (Sokol and Flach, 2020), they have not been proposed for datasets.”
From Anecdotal Evidence to Quantitative Evaluation Methods: A Systematic Review on Evaluating Explainable AI (2022, background): “Continuity also adds to generalizability beyond a particular input [166, 237] or generalizability to new contexts [238].”
What Do We Want From Explainable Artificial Intelligence (XAI)? - A Stakeholder Perspective on XAI and a Conceptual Model Guiding Interdisciplinary XAI Research (2021, background): “This finding is in line with calls for experiments involving human participants in proxy tasks in order to show that given explanatory information not only elicits understanding when simulating a context, but also under real world conditions (that is, to investigate whether given explanatory informa”
Trustworthy Artificial Intelligence: A Review (2022, background): “Sokol and Flach [186] proposed a fact sheet for systematic evaluation of interpretable approaches.”
Teach Me to Explain: A Review of Datasets for Explainable Natural Language Processing (2021, background): “Explainability fact sheets have been proposed for models [114], but not for datasets.”
Overcoming Failures of Imagination in AI Infused System Development and Deployment (2020, background): “Some of these calls have been realized through structured documentation and checklists aiming at codifying a range of responsible AI principles (Arnold et al., 2019; Mitchell et al., 2019; Gebru et al., 2018; Sokol and Flach, 2020; Stoyanovich and Howe, 2019; Madaio et al., 2020; Kiran et al., 2015)”
AI Transparency in the Age of LLMs: A Human-Centered Research Roadmap (2023, background): “” For a long list of goodness criteria, we point interested readers to Sokol and Flach [172] and Carvalho et al.”
Opportunities and Challenges in Explainable Artificial Intelligence (XAI): A Survey (2020, background): “Each explanation should be consistent across similar data points and generate stable or similar explanation on the same data point over time [39].”
Interactive Model Cards: A Human-Centered Approach to Model Documentation (2022, background): “We propose a novel concept for an interactive model card (IMC) that: (1) lowers barriers to accessing key information about model behavior; (2) supports deeper interrogation of models; and (3) surfaces some attendant risks and limitations without additional work on the part of model developers.”
Explainable Artificial Intelligence (XAI) 2.0: A Manifesto of Open Challenges and Interdisciplinary Research Directions (2023, background): “Expanded lists with further terms and concepts can be found in [4,5,18,27,28].”