A multi-component framework for the analysis and design of explainable artificial intelligence

Abstract

The rapid growth of research in explainable artificial intelligence (XAI) follows on two substantial developments. First, the enormous application success of modern machine learning methods, especially deep and reinforcement learning, which have created high expectations for industrial, commercial and social value.Second, the emergence of concern for creating trusted AI systems, including the creation of regulatory principles to ensure transparency and trust of AI systems.These two threads have created a kind of "perfect storm" of research activity, all eager to create and deliver any set of tools and techniques to address the XAI demand.As some surveys of current XAI suggest, there is yet to appear a principled framework that respects the literature of explainability in the history of science, and which provides a basis for the development of a framework for transparent XAI. Here we intend to provide a strategic inventory of XAI requirements, demonstrate their connection to a history of XAI ideas, and synthesize those ideas into a simple framework to calibrate five successive levels of XAI.

STEP-EZ: Syntax Tree guided semantic ExPlanation for Explainable Zero-shot modeling of clinical depression symptoms from text

“…Here we discuss the performance of our models based on two broad categories of performance measures, such as, (1) accuracy and (2) explainability [2] as follows, Depression Symptoms Detection (DSD) Task Accuracy We observe that NLI models are the best and Sentence Embedding Familty (SEF) models are on par indicates, NLI and sentence embedding models with their semantic similarity pre-training, are inherently better in ZSL tasks. With respect to top-k we can see gradual decrease of Micro-F1 with the increase of top-k with pick at 1, see Figure 3.…”

“…Most of the earlier work in text based Depression classification can be divided into two broad categories such as, (1) Post level signs of Depression detection [23,5] and (2) User-level signs of Depression detection [24,8]. It is to be noted that task (1) is often an important pre-requisite of task (2). Even importantly, for clinically meaningful user-level signs of Depression detection, we need to have models that can identify post level signs of clinical Depression symptoms.…”

On Explainable AI Solutions for Targeting in Cyber Military Operations

“…Accordingly, the quality of explanations depends on how the stakeholder/audience perceives them depending on their background, goals, expectations, context etc. (Atakishiyev et al, 2020). Hence, based on (Swartout & Moore, 1993;Walsh et al, 2021), the following criteria are considered to evaluate explanations of AI models:…”

Opportunities in Real Time Fraud Detection: An Explainable Artificial Intelligence (XAI) Research Agenda

“…This concept of interpretability representing models that can be decomposed by an appropriately skilled audience is becoming more widely recognised amongst contemporary authors. Specifically, authors identify linear models, decision trees, rule-based models and constrained variants of black-box models as interpretable models [21]- [25]. Such models are often referred to as "inherently" interpretable [22], [23], [26], [27] or "intrinsic" [28], with the advantage that they are able to provide accurate and undistorted [26] explanations for the model output.…”

“…This definition brings together four key elements of context frequently discussed in the literature. First it leverages the centrality of the audience [25], [87]- [90] since it is the audience who determines whether the explanation is a good one or not [16]. Second it captures the importance of understanding the goals of the audience, [85], [91] since it is the goals that drive the ML model design [85], [91], [92].…”