Developing a Catalogue of Explainability Methods to Support Expert and Non-expert Users

Abstract

Organisations face growing legal requirements and ethical responsibilities to ensure that decisions made by their intelligent systems are explainable. However, provisioning of an explanation is often application dependent, causing an extended design phase and delayed deployment. In this paper we present an explainability framework formed of a catalogue of explanation methods, allowing integration to a range of projects within a telecommunications organisation. These methods are split into low-level explanations, high-level explanations and co-created explanations. We motivate and evaluate this framework using the specific case-study of explaining the conclusions of field engineering experts to non-technical planning staff. Feedback from an iterative co-creation process and a qualitative evaluation is indicative that this is a valuable development tool for use in future company projects.

Contextualization and Exploration of Local Feature Importance Explanations to Improve Understanding and Satisfaction of Non-Expert Users

“…However, for now there is no consensus on what it means in practice to provide contextual information for ML explanations. In the literature, the main approach consists in extracting automatically the context from the ML system or the data [2,12,19,23,27,27,29,35,37]. Indeed, context can first be directly extracted from the ML system, e.g.…”

“…However, they can be relevant for non-expert users as well. Martin et al [23] show that these users are more likely to ask for concrete examples of output from the training set. Gomez et al [12] propose to contextualize local explanations by adding a visual representation of the dataset.…”

“…Contextualization is considered an essential part of the explanation process by the social science field [22,24]. In the XAI community, several works showed the positive impact of contextualizing explanations by adding domain knowledge [17,27,29] or by displaying relevant training examples [12,23,35]. However, most of these works target expert users, even though non-expert users are likely to also benefit from additional contextual information.…”

Energy democracy: A digital future?

“…However, in an evolving field where the sociotechnical context and expert understanding of the issues continue to change rapidly, the knowledge required to participate will remain a moving target. With similar challenges echoed across the digital economy [102,103], there is clear demand for policy and practitioner development in this area.…”

Explaining Predictions by Approximating the Local Decision Boundary

“…Over the last few years, explaining opaque machine learning (ML) models has become a topic of increasing attention [2,25,8,23,37,26]. This attention arises from multiple needs of ML users, such as ensuring model trustworthiness, detecting and removing unwanted biases (fairness) and understanding causal relationships [2].…”