Evaluating Architectural Safeguards For Uncertain Ai Black Box Components

Although tremendous progress has been made in Artificial Intelligence (AI), it entails new challenges. The growing complexity of learning tasks requires more complex AI components, which increasingly exhibit unreliable behaviour. In this book, we present a model-driven approach to model architectural safeguards for AI components and analyse their effect on the overall system reliability.

In this work, the authors analysed the co-dependency between models and analyses, particularly the structure and interdependence of artefacts and the feature-based decomposition and composition of model-based analyses. Their goal is to improve the maintainability of model-based analyses. They have investigated the co-dependency of Domain-specific Modelling Languages (DSMLs) and model-based analyses regarding evolvability, understandability, and reusability.

This work addresses the problem of how to capture the dynamics of maneuvering objects for visual tracking. Towards this end, the perspective of recursive Bayesian filters and the perspective of deep learning approaches for state estimation are considered and their functional viewpoints are brought together.

This book assesses the normative and practical challenges for artificial intelligence (AI) regulation, offers comprehensive information on the laws that currently shape or restrict the design or use of AI, and develops policy recommendations for those areas in which regulation is most urgently needed. By gathering contributions from scholars who are experts in their respective fields of legal research, it demonstrates that AI regulation is not a specialized sub-discipline, but affects the entire legal system and thus concerns all lawyers. Machine learning-based technology, which lies at the heart of what is commonly referred to as AI, is increasingly being employed to make policy and business decisions with broad social impacts, and therefore runs the risk of causing wide-scale damage. At the same time, AI technology is becoming more and more complex and difficult to understand, making it harder to determine whether or not it is being used in accordance with the law. In light of this situation, even tech enthusiasts are calling for stricter regulation of AI. Legislators, too, are stepping in and have begun to pass AI laws, including the prohibition of automated decision-making systems in Article 22 of the General Data Protection Regulation, the New York City AI transparency bill, and the 2017 amendments to the German Cartel Act and German Administrative Procedure Act. While the belief that something needs to be done is widely shared, there is far less clarity about what exactly can or should be done, or what effective regulation might look like. The book is divided into two major parts, the first of which focuses on features common to most AI systems, and explores how they relate to the legal framework for data-driven technologies, which already exists in the form of (national and supra-national) constitutional law, EU data protection and competition law, and anti-discrimination law. In the second part, the book examines in detail a number of relevant sectors in which AI is increasingly shaping decision-making processes, ranging from the notorious social media and the legal, financial and healthcare industries, to fields like law enforcement and tax law, in which we can observe how regulation by AI is becoming a reality.

Visual perception is one of the most important sources of information for both humans and robots. A particular challenge is the acquisition and interpretation of complex unstructured scenes. This work contributes to active vision for humanoid robots. A semantic model of the scene is created, which is extended by successively changing the robot's view in order to explore interaction possibilities of the scene.

When models of a system change, analyses based on them have to be reevaluated in order for the results to stay meaningful. In many cases, the time to get updated analysis results is critical. This thesis proposes multiple, combinable approaches and a new formalism based on category theory for implicitly incremental model analyses and transformations. The advantages of the implementation are validated using seven case studies, partially drawn from the Transformation Tool Contest (TTC).

The development of “intelligent” systems that can take decisions and perform autonomously might lead to faster and more consistent decisions. A limiting factor for a broader adoption of AI technology is the inherent risks that come with giving up human control and oversight to “intelligent” machines. For sensitive tasks involving critical infrastructures and affecting human well-being or health, it is crucial to limit the possibility of improper, non-robust and unsafe decisions and actions. Before deploying an AI system, we see a strong need to validate its behavior, and thus establish guarantees that it will continue to perform as expected when deployed in a real-world environment. In pursuit of that objective, ways for humans to verify the agreement between the AI decision structure and their own ground-truth knowledge have been explored. Explainable AI (XAI) has developed as a subfield of AI, focused on exposing complex AI models to humans in a systematic and interpretable manner. The 22 chapters included in this book provide a timely snapshot of algorithms, theory, and applications of interpretable and explainable AI and AI techniques that have been proposed recently reflecting the current discourse in this field and providing directions of future development. The book is organized in six parts: towards AI transparency; methods for interpreting AI systems; explaining the decisions of AI systems; evaluating interpretability and explanations; applications of explainable AI; and software for explainable AI.