Cases for Explainable Software Systems: Characteristics and Examples

Abstract

Human explanations are often contrastive, meaning that they do not answer the indeterminate "Why?" question, but instead "Why P, rather than Q?". Automatically generating contrastive explanations is challenging because the contrastive event (Q) represents the expectation of a user in contrast to what happened. We present an approach that predicts a potential contrastive event in situations where a user asks for an explanation in the context of rule-based systems. Our approach analyzes a situation that needs to be explained and then selects the most likely rule a user may have expected instead of what the user has observed. This contrastive event is then used to create a contrastive explanation that is presented to the user. We have implemented the approach as a plugin for a home automation system and demonstrate its feasibility in four test scenarios. CCS CONCEPTS• Software and its engineering → Extra-functional properties;• Human-centered computing → Interactive systems and tools; Ubiquitous and mobile computing.

Explaining the Unexplainable: The Impact of Misleading Explanations on Trust in Unreliable Predictions for Hardly Assessable Tasks

“…This theory helps to elucidate the results of studies such as [54,69] belonging to Q1, wherein the effect of explanations on trust is found to be negligible 1 . A plausible assumption is that individuals tend to bypass explanations when the judgment process is straightforward and they perceive the system to work effectively [59]. This rationale finds support in the study conducted by Bansal et al [4] where participants claimed that they mostly ignored AI in easily assessable tasks (the sentiment analysis) with up to tripled ratio with respect to more hardly assessable tasks in that study (the LSAT question answering).…”

Correlating contexts and NFR conflicts from event logs

“…The authors proposed a model to analyze the impacts of explainability across different quality dimensions. Sadeghi et al [16] present a taxonomy of explanation needs that classify scenarios that require explanations. The taxonomy can be used to guide the requirements elicitation for explanation capabilities of interactive intelligent systems.…”

Explanation Needs in App Reviews: Taxonomy and Automated Detection

“…Constructing taxonomies provides numerous benefits, including supporting the communication of complex concepts, revealing relationships between entities, and uncovering knowledge gaps. In a similar approach for a different domain, Sadeghi et al [3] developed a taxonomy of reasons for Explanation Needs. They primarily distinguish between four categories of situations requiring explanations: Training, Interaction, Debugging, and Validation, yet the authors focused on Interaction.…”

“…There are several approaches to generating explanations for different algorithmic paradigms. However, there has been relatively little focus in the literature on what users actually need explanations for [3]. This lack of knowledge limits our ability to effectively elicit explainability requirements and apply existing explanation generation methods.…”