Transparency and Explainability of AI Systems: Ethical Guidelines in Practice

Abstract

Context and Motivation]Recent studies have highlighted transparency and explainability as important quality requirements of AI systems. However, there are still relatively few case studies that describe the current state of defining these quality requirements in practice.[Question] The goal of our study was to explore what ethical guidelines organizations have defined for the development of transparent and explainable AI systems. We analyzed the ethical guidelines in 16 organizations representing different industries and public sector.[Results] In the ethical guidelines, the importance of transparency was highlighted by almost all of the organizations, and explainability was considered as an integral part of transparency. Building trust in AI systems was one of the key reasons for developing transparency and explainability, and customers and users were raised as the main target groups of the explanations. The organizations also mentioned developers, partners, and stakeholders as important groups needing explanations. The ethical guidelines contained the following aspects of the AI system that should be explained: the purpose, role of AI, inputs, behavior, data utilized, outputs, and limitations. The guidelines also pointed out that transparency and explainability relate to several other quality requirements, such as trustworthiness, understandability, traceability, privacy, auditability, and fairness. [Contribution] For researchers, this paper provides insights into what organizations consider important in the transparency and, in particular, explainability of AI systems. For practitioners, this study suggests a structured way to define explainability requirements of AI systems.

Norms or fun? The influence of ethical concerns and perceived enjoyment on the regulation of deepfake information

“…This can help identify the reasons for improper application, which in turn can help prevent potential safety hazards in the future (European Commission, 2019). Traceability helps realize the realization of accountability and auditability of deepfake information, and these principles are key to ensure the reliability of such content (Balasubramaniam et al, 2022). Hence, we propose the following hypothesis:…”

“…Traceability plays a crucial role in identifying and distinguishing deepfake information and is essential for post-event accountability (Balasubramaniam et al, 2022). The data sets, data markers and algorithms used in the production of deepfake information should be recorded for traceability, as it reduces the deceptive nature of the fabricated content by revealing its source (Di Domenico et al, 2021a).…”

Interpretable Medical Imagery Diagnosis with Self-Attentive Transformers: A Review of Explainable AI for Health Care

“…However, as AI applications become more ubiquitous, concerns have arisen regarding their transparency, accountability, and trustworthiness [24]. When AI model is used in mission critical applications like medical diagnosis [25], Cybersecurity [26], or autonomous driving [27], transparency enables stakeholders to comprehend the rationale behind the model's predictions, ensuring accountability, identifying potential biases, and facilitating trust in high-stakes scenarios.…”

Requirements Framework for Engineering Human-centered Artificial Intelligence-Based Software Systems

“…Explainability requirements might occur during the building process or after the AI-based model is deployed. In [51] explainability requirements are divided into four components to include: Who the explanation is addressed to, what needs to be explained, when should the explanation happen, and who explains? Also, explanations should consist of consequences that might occur due to an action performed by the user.…”