The societal and ethical implications of the use of opaque artificial intelligence systems for consequential decisions, such as welfare allocation and criminal justice, have generated a lively debate among multiple stakeholder groups, including computer scientists, ethicists, social scientists, policy makers, and end users. However, the lack of a common language or a multi-dimensional framework to appropriately bridge the technical, epistemic, and normative aspects of this debate prevents the discussion from being as productive as it could be. Drawing on the philosophical literature on the nature and value of explanations, this paper offers a multi-faceted framework that brings more conceptual precision to the present debate by (1) identifying the types of explanations that are most pertinent to artificial intelligence predictions, (2) recognizing the relevance and importance of social and ethical values for the evaluation of these explanations, and (3) demonstrating the importance of these explanations for incorporating a diversified approach to improving the design of truthful algorithmic ecosystems. The proposed philosophical framework thus lays the groundwork for establishing a pertinent connection between the technical and ethical aspects of artificial intelligence systems.

Explainable NLP (ExNLP) has increasingly focused on collecting human-annotated explanations. These explanations are used downstream in three ways: as data augmentation to improve performance on a predictive task, as a loss signal to train models to produce explanations for their predictions, and as a means to evaluate the quality of model-generated explanations. In this review, we identify three predominant classes of explanations (highlights, free-text, and structured), organize the literature on annotating each type, point to what has been learned to date, and give recommendations for collecting ExNLP datasets in the future.

In recent years, there has been an explosion of AI research on counterfactual explanations as a solution to the problem of eXplainable AI (XAI). These explanations seem to offer technical, psychological and legal benefits over other explanation techniques. We survey 100 distinct counterfactual explanation methods reported in the literature. This survey addresses the extent to which these methods have been adequately evaluated, both psychologically and computationally, and quantifies the shortfalls occurring. For instance, only 21% of these methods have been user tested. Five key deficits in the evaluation of these methods are detailed and a roadmap, with standardised benchmark evaluations, is proposed to resolve the issues arising; issues, that currently effectively block scientific progress in this field.

The vast field of Artificial Intelligence(AI) has experienced enormous growth in recent years. With newer and more complex models coming each year, AI models have started to surpass human intellect at a pace that no one could have predicted. But as we get more accurate and precise results, it's becoming harder to explain the reasoning behind the complex mathematical decisions these models take. This mathematical abstraction also doesn't help the users maintain their trust in a particular model's decisions. Though the model might have given the correct diagnosis, a doctor can't really advise a patient confidently as he/she doesn't know the reasoning behind the said model's diagnosis.

The widespread adoption of black-box models in Artificial Intelligence has enhanced the need for explanation methods to reveal how these obscure models reach specific decisions. Retrieving explanations is fundamental to unveil possible biases and to resolve practical or ethical issues. Nowadays, the literature is full of methods with different explanations. We provide a categorization of explanation methods based on the type of explanation returned. We present the most recent and widely used explainers, and we show a visual comparison among explanations and a quantitative benchmarking.

Logic-based argumentation is a well-established formalism modelling nonmonotonic reasoning. It has been playing a major role in AI for decades, now. Informally, a set of formulas is the support for a given claim if it is consistent, subset-minimal, and implies the claim. In such a case, the pair of the support and the claim together is called an argument. In this paper, we study the propositional variants of the following three computational tasks studied in argumentation: ARG (exists a support for a given claim with respect to a given set of formulas), ARG-Check (is a given set a support for a given claim), and ARG-Rel (similarly as ARG plus requiring an additionally given formula to be contained in the support). ARG-Check is complete for the complexity class DP, and the other two problems are known to be complete for the second level of the polynomial hierarchy (Parson et al., J. Log. Comput., 2003) and, accordingly, are highly intractable. Analyzing the reason for this intractability, we perform a two-dimensional classification: first, we consider all possible propositional fragments of the problem within Schaefer's framework (STOC 1978), and then study different parameterizations for each of the fragment. We identify a list of reasonable structural parameters (size of the claim, support, knowledge-base) that are connected to the aforementioned decision problems. Eventually, we thoroughly draw a fine border of parameterized intractability for each of the problems showing where the problems are fixed-parameter tractable and when this exactly stops. Surprisingly, several cases are of very high intractability (paraNP and beyond).

Argumentation provides a representation of arguments and attacks between these arguments. Argumentation can be used to represent a reasoning process over evidence to reach conclusions. Within such a reasoning process, understanding the value of information can improve the quality of decision making based on the output of the reasoning process. The value of an item of information is inherently dependent on the available evidence and the question being answered by the reasoning. In this paper we introduce a value of information on argument frameworks to identify the most valuable arguments within the finite set of arguments in the framework, and the arguments and attacks which could be added to change the output of an evaluation. We demonstrate the value of information within an argument framework representing an intelligence analysis in the maritime domain. Understanding the value of information in an intelligence analysis will allow analysts to balance the value against the costs and risks of collection, to effectively request further collection of intelligence to increase the confidence in the analysis of hypotheses.

Previous research in Explainable Artificial Intelligence (XAI) suggests that a main aim of explainability approaches is to satisfy specific interests, goals, expectations, needs, and demands regarding artificial systems (we call these stakeholders' desiderata) in a variety of contexts. However, the literature on XAI is vast, spreads out across multiple largely disconnected disciplines, and it often remains unclear how explainability approaches are supposed to achieve the goal of satisfying stakeholders' desiderata. This paper discusses the main classes of stakeholders calling for explainability of artificial systems and reviews their desiderata. We provide a model that explicitly spells out the main concepts and relations necessary to consider and investigate when evaluating, adjusting, choosing, and developing explainability approaches that aim to satisfy stakeholders' desiderata. This model can serve researchers from the variety of different disciplines involved in XAI as a common ground. It emphasizes where there is interdisciplinary potential in the evaluation and the development of explainability approaches.

With recent progress in the field of Explainable Artificial Intelligence (XAI) and increasing use in practice, the need for an evaluation of different XAI methods and their explanation quality in practical usage scenarios arises. For this purpose, we present VitrAI, which is a web-based service with the goal of uniformly demonstrating four different XAI algorithms in the context of three real life scenarios and evaluating their performance and comprehensibility for humans. This work reveals practical obstacles when adopting XAI methods and gives qualitative estimates on how well different approaches perform in said scenarios.

Explainable Artificial Intelligence (XAI) has re-emerged in response to the development of modern AI and ML systems. These systems are complex and sometimes biased, but they nevertheless make decisions that impact our lives. XAI systems are frequently algorithm-focused; starting and ending with an algorithm that implements a basic untested idea about explainability. These systems are often not tested to determine whether the algorithm helps users accomplish any goals, and so their explainability remains unproven. We propose an alternative: to start with human-focused principles for the design, testing, and implementation of XAI systems, and implement algorithms to serve that purpose. In this paper, we review some of the basic concepts that have been used for user-centered XAI systems over the past 40 years of research. Based on these, we describe the "Self-Explanation Scorecard", which can help developers understand how they can empower users by enabling self-explanation. Finally, we present a set of empirically-grounded, user-centered design principles that may guide developers to create successful explainable systems.