Despite its enormous economical and societal impact, lack of human-perceived control and safety is re-defining the design and development of emerging AI-based technologies. New regulatory requirements mandate increased human control and oversight of AI, transforming the development practices and responsibilities of individuals interacting with AI. In this paper, we present the SPATIAL architecture, a system that augments modern applications with capabilities to gauge and monitor trustworthy properties of AI inference capabilities. To design SPATIAL, we first explore the evolution of modern system architectures and how AI components and pipelines are integrated. With this information, we then develop a proof-of- concept architecture that analyzes AI models in a human-in-the- loop manner. SPATIAL provides an AI dashboard for allowing individuals interacting with applications to obtain quantifiable insights about the AI decision process. This information is then used by human operators to comprehend possible issues that influence the performance of AI models and adjust or counter them. Through rigorous benchmarks and experiments in real- world industrial applications, we demonstrate that SPATIAL can easily augment modern applications with metrics to gauge and monitor trustworthiness, however, this in turn increases the complexity of developing and maintaining systems implementing AI. Our work highlights lessons learned and experiences from augmenting modern applications with mechanisms that support regulatory compliance of AI. In addition, we also present a road map of on-going challenges that require attention to achieve robust trustworthy analysis of AI and greater engagement of human oversight.@en

Various AI systems have taken a unique space in our daily lives, helping us in decision-making in critical as well as non-critical scenarios. Although these systems are widely adopted across different sectors, they have not been used to their full potential in critical domains such as the healthcare sector enabled by the Internet of Things (IoT). One of the important hindering factors for adoption is the implication for accountability of decisions and outcomes affected by an AI system, where the term accountability is understood as a means to ensure the performance of a system. However, this term is often interpreted differently in various sectors. Since the EU GDPR regulations and the US congress have emphasised the importance of enabling accountability in AI systems, there is a strong demand to understand and conceptualise this term. It is crucial to address various aspects integrated with accountability and understand how it affects the adoption of AI systems. In this paper, we conceptualise these factors affecting accountability and how it contributes to a trustworthy healthcare AI system. By focusing on healthcare IoT systems, our conceptual mapping will help the readers understand what system aspects those factors are contributing to and how they affect the system trustworthiness. Besides illustrating accountability in detail, we also share our vision towards causal interpretability as a means to enhance accountability for healthcare AI systems. The insights of this paper shall contribute to the knowledge of academic research on accountability, and benefit AI developers and practitioners in the healthcare sector. @en

Different explainable AI (XAI) methods are based on different notions of ‘ground truth’. In order to trust explanations of AI systems, the ground truth has to provide fidelity towards the actual behaviour of the AI system. An explanation that has poor fidelity towards the AI system’s actual behaviour can not be trusted no matter how convincing the explanations appear to be for the users. The Contextual Importance and Utility (CIU) method differs from currently popular outcome explanation methods such as Local Interpretable Model-agnostic Explanations (LIME) and Shapley values in several ways. Notably, CIU does not build any intermediate interpretable model like LIME, and it does not make any assumption regarding linearity or additivity of the feature importance. CIU also introduces the value utility notion and a definition of feature importance that is different from LIME and Shapley values. We argue that LIME and Shapley values actually estimate ‘influence’ (rather than ‘importance’), which combines importance and utility. The paper compares the three methods in terms of validity of their ground truth assumption and fidelity towards the underlying model through a series of benchmark tasks. The results confirm that LIME results tend not to be coherent nor stable. CIU and Shapley values give rather similar results when limiting explanations to ‘influence’. However, by separating ‘importance’ and ‘utility’ elements, CIU can provide more expressive and flexible explanations than LIME and Shapley values.@en

Reinforcement Learning performs well in many different application domains and is starting to receive greater authority and trust from its users. But most people are unfamiliar with how AIs make their decisions and many of them feel anxious about AI decision-making. A result of this is that AI methods suffer from trust issues and this hinders the full-scale adoption of them. In this paper we determine what the main application domains of Reinforcement Learning are, and to what extent research in those domains has explored explainability. This paper reviews examples of the most active application domains for Reinforcement Learning and suggest some guidelines to assess the importance of explainability for these applications. We present some key factors that should be included in evaluating these applications and show how these work with the examples found. By using these assessment criteria to evaluate the explainability needs for Reinforcement Learning, the research field can be guided to increasing transparency and trust through explanations.@en

Robots and digital agents find their way in an increasing number of areas in our everyday lives. In this paper we look at the history of coaching devices and their impact on our health and lifestyle, as well as the emergence of coaching robots. We explore the worry that a growing entanglement with coaching technology that quantify our lives may reduce or invalidate the user's personal experiences and preferences. We propose that the kind of bond that should be formed between user and technology is one that engenders trust while maintaining contextual and user-focused perspectives, in order to preserve personal values and autonomy, as well as the possibility of coaching robots being able to provide these kinds of bonding interactions.@en

Cognitive science and artificial intelligence are interconnected in that developments in one field can affect the framework of reference for research in the other. Changes in our understanding of how the human mind works inadvertently changes how we go about creating artificial minds. Similarly, successes and failures in AI can inspire new directions to be taken in cognitive science. This article explores the history of the mind in cognitive science in the last 50 years, and draw comparisons as to how this has affected AI research, and how AI research in turn has affected shifts in cognitive science. In particular, we look at explainable AI (XAI) and suggest that folk psychology is of particular interest for that area of research. In cognitive science, folk psychology is divided between two theories: theory-theory and simulation theory. We argue that it is important for XAI to recognise and understand this debate, and that reducing reliance on theory-theory by incorporating more simulationist frameworks into XAI could help further the field. We propose that such incorporation would involve robots employing more embodied cognitive processes when communicating with humans, highlighting the importance of bodily action in communication and mindreading.@en