Trustworthy AI Research Articles

Cyber—Physical Attack Detection in Water Distribution Systems with Temporal Graph Convolutional Neural Networks

Prompt detection of cyber–physical attacks (CPAs) on a water distribution system (WDS) is critical to avoid irreversible damage to the network infrastructure and disruption of water services. However, the complex interdependencies of the water network’s components make CPA detection challenging. To better capture the spatiotemporal dimensions of these interdependencies, we represented the WDS as a mathematical graph and approached the problem by utilizing graph neural networks. We presented an online, one-stage, prediction-based algorithm that implements the temporal graph convolutional network and makes use of the Mahalanobis distance. The algorithm exhibited strong detection performance and was capable of localizing the targeted network components for several benchmark attacks. We suggested that an important property of the proposed algorithm was its explainability, which allowed the extraction of useful information about how the model works and as such it is a step towards the creation of trustworthy AI algorithms for water applications. Additional insights into metrics commonly used to rank algorithm performance were also presented and discussed.

Hasta la vista, baby: reflections on the risks of algocracy, killer robots, and artificial superintelligence

<p>The neologism Algocracy may mean government or governance by algorithms. Architects of artificial intelligence have perspectives on killer robots and government by artificial superintelligence and are engaged in public debate on both themes. The risks of being dominated by artificial superintelligence and of being subjected to undemocratic, unconstitutional or illegal algo norms inspires our reflection. Institutions should organize rules of the game that prevent machine learning algorithms from learning how to dominate humans. Algorithms need new design requirements to incorporate responsibility, transparency, auditability, incorruptibility, and predictability. The algorithmic responsibility of the state, national public policies for developing a trustworthy AI, and the algorithmic law of killer robots and artificial superintelligence could reduce the risks of algocracy. The particular character of algorithms demands a special discipline to control their power, architecture, and commands. law and government can channel the development and use of killer robots, eventually even setting a global prohibition of autonomous weapons. Likewise, the threat of government by algorithms posed by the emergence of an artificial superintelligence that dominates humankind also requires the development of a new algorithmic law that establishes checks and balances and controls the technological system.<em></em></p>

A Leap of Faith: Is There a Formula for "Trustworthy" AI?

Trust is one of the big buzzwords in debates about the shaping of society, democracy, and emerging technologies. For example, one prominent idea put forward by the High-Level Expert Group on Artificial Intelligence appointed by the European Commission is that artificial intelligence should be trustworthy. In this essay, we explore the notion of trust and argue that both proponents and critics of trustworthy AI have flawed pictures of the nature of trust. We develop an approach to understanding trust in AI that does not conceive of trust merely as an accelerator for societal acceptance of AI technologies. Instead, we argue, trust is granted through leaps of faith. For this reason, trust remains precarious, fragile, and resistant to promotion through formulaic approaches. We also highlight the significance of distrust in societal deliberation, as it is relevant to trust in various and intricate ways. Among the fruitful aspects of distrust is that it enables individuals to forgo technology if desired, to constrain its power, and to exercise meaningful human control.

Ethics-Based Auditing to Develop Trustworthy AI

A series of recent developments points towards auditing as a promising mechanism to bridge the gap between principles and practice in AI ethics. Building on ongoing discussions concerning ethics-based auditing, we offer three contributions. First, we argue that ethics-based auditing can improve the quality of decision making, increase user satisfaction, unlock growth potential, enable law-making, and relieve human suffering. Second, we highlight current best practices to support the design and implementation of ethics-based auditing: To be feasible and effective, ethics-based auditing should take the form of a continuous and constructive process, approach ethical alignment from a system perspective, and be aligned with public policies and incentives for ethically desirable behaviour. Third, we identify and discuss the constraints associated with ethics-based auditing. Only by understanding and accounting for these constraints can ethics-based auditing facilitate ethical alignment of AI, while enabling society to reap the full economic and social benefits of automation.

Opening the path to ethics in artificial intelligence

In an attempt to address the ethical challenges in AI, we currently have several ethical AI frameworks in place worldwide, with more being released or in development around the world every day. From the European Commission's Guidelines for Trustworthy AI, to the Asilomar AI Principles, the message is usually similar: more transparency and explicability. However, navigating the broad number of resources currently available is not a simple process. So, how do we find common ground when it comes to ethics in AI?

Towards the Trustworthy AI

After decades of theoretical deliberations, the rapid development of advanced information technology has allowed machine learning as a first practical step towards artificial intelligence to enter widespread commercial and government use. The transition into a post-industrial, information society has revealed the value of data as an important resource whose processing is the basis of the new innovative information society services. The European Union has enacted several important regulations and directives in the recent past to protect the recognized fundamental rights of individuals and to regulate the obligations of service providers to ensure safe and secure processing. The Charter of Fundamental Rights as the legal basis of the European system of human rights contains significant checks and limitations to the effect and purpose of future EU AI regulation. Whenever and however this regulation is adopted, it will need to comply with and contain existing European legal standards regarding the fundamental rights of individuals in the EU. The European Commission’s ethical guidelines establish ethical principles based on the recognized fundamental rights that future AI systems need to adhere to in order to be recognized as trustworthy. The purpose of this paper is to present and analyse the mechanisms present in existing European regulations in the fields of data protection and information security and in the European Union documents regarding the future artificial intelligence regulation and to offer suggestions for future regulations. The research methodology includes a comparative analysis of available regulations and policy documents of the European Union, national laws, legal literature, and other sources.

Regulating Transformative Technology in The Quantum Age: Intellectual Property, Standardization & Sustainable Innovation

Regulating Transformative Technology in The Quantum Age: Intellectual Property, Standardization & Sustainable Innovation

Algorithm, Machine Learning and Artificial Intelligence

We discuss the concepts of algorithm, machine learning and artificial intelligence. We start by noting that different definitions of algorithm may serve different epistemic purposes, and by highlighting the specificities of computer science algorithms together with a high-level overview of their different types. We continue by discussing machine learning as a discipline and commenting on its algorithms, with a focus on artificial neural networks due to their role in the deep learning revolution. We discuss the epistemic and ethical challenges arising from the widely spread use of machine learning algorithms in the applications. Then, we discuss the artifacts of the artificial intelligence domain, clarifying their relation with machine learning algorithms. Finally, we provide the reader with comments on the emergence of the “trustworthy AI” paradigm, highlighting some of the theoretical challenges affecting the discussions on trust in human-AI interactions.

How the EU Can Achieve Legally Trustworthy AI: A Response to the European Commission’s Proposal for an Artificial Intelligence Act

This document contains the response to the European Commission’s Proposal for an Artificial Intelligence Act from members of the Legal, Ethical & Accountable Digital Society (LEADS) Lab at the University of Birmingham. The Proposal seeks to give expression to the concept of ‘Lawful AI.’ This concept was mentioned, but not developed in the Commission’s High-Level Expert Group on AI’s Ethics Guidelines for Trustworthy AI (2019), which instead confined its discussion to the concepts of ‘Ethical’ and ‘Robust’ AI. After a brief introduction (Chapter 1), we set out the many aspects of the Proposal which we welcome, and stress our wholehearted support for its aim to protect fundamental rights (Chapter 2). Subsequently, we develop the concept of ‘Legally Trustworthy AI,’ arguing that it should be grounded in respect for three pillars on which contemporary liberal democratic societies are founded, namely: fundamental rights, the rule of law, and democracy (Chapter 3). Drawing on this conceptual framework, we first argue that the Proposal fails to reflect fundamental rights as claims with enhanced moral and legal status, which subjects any rights interventions to a demanding regime of scrutiny and must satisfy tests of necessity and proportionality. Moreover, the Proposal does not always accurately recognise the wrongs and harms associated with different kinds of AI systems and appropriately allocates responsibility for them. Second, the Proposal does not provide an effective framework for the enforcement of legal rights and duties, and does not ensure legal certainty and consistency, which are essential for the rule of law. Third, the Proposal neglects to ensure meaningful transparency, accountability, and rights of public participation, thereby failing to reflect adequate protection for democracy (Chapter 4). Based on these shortcomings in respecting and promoting the three pillars of Legally Trustworthy AI, we provide detailed recommendations for the Proposal’s revision (Chapter 5).

Varieties of AI Explanations Under the Law. From the GDPR to the AIA, and Beyond

The quest to explain the output of artificial intelligence systems has clearly moved from a mere technical to a highly legally and politically relevant endeavor. In this paper, we provide an overview of legal obligations to explain AI and evaluate current policy proposals. In this, we distinguish between different functional varieties of AI explanations - such as multiple forms of enabling, technical and protective transparency - and show how different legal areas engage with and mandate such different types of explanations to varying degrees. Starting with the rights-enabling framework of the GDPR, we proceed to uncover technical and protective forms of explanations owed under contract, tort and banking law. Moreover, we discuss what the recent EU proposal for an Artificial Intelligence Act means for explainable AI, and review the proposal’s strengths and limitations in this respect. Finally, from a policy perspective, we advocate for moving beyond mere explainability towards a more encompassing framework for trustworthy and responsible AI that includes actionable explanations, values-in-design and co-design methodologies, interactions with algorithmic fairness, and quality benchmarking.

Autonomous Vehicle Regulation and Trust

The autonomous vehicle (AV) industry works very hard to create public trust in both AV technology and its developers. Building trust is part of a strategy to permit the industry itself to manage the testing and deployment of self-driving vehicle technology without regulatory interference. The AV industry supplements its appeal for trust by extolling a plethora of public benefits that AV technology supposedly will bring to society. In this article, we examine the behavior of certain AV industry participants in various concrete situations, considering whether actions by these AV industry participants promote trust or, conversely, create concerns. To evaluate whether a particular action promotes trust, we consider both common sense intuitions as well as recommendations contained in three publications with intended global applicability: (i) the recently published IEEE Standard Model Process for Addressing Ethical Concerns during System Design (IEEE 7000); (ii) Ethics Guidelines for Trustworthy AI; and (iii) Policy and Investment Recommendations for Trustworthy AI. These publications inform our analysis of practices that merit trust (and helps us identify those that do not). Our analysis identifies certain trust destroying practices, suggesting that the public should not accept either AV industry appeals for trust or take at face value the proffered narrative of benefits. We perceive a need for more regulation by federal, state and local governments. Despite our concerns, we discuss how the AV industry might earn trust by changing its adversarial approach to law, regulation and disclosure, starting by compliance with standards promulgated by the engineering community. AV industry participants should shift to cooperation with lawmakers and regulators by embracing engineering standards relating to safety, such as SAE J3018, and ethical principles which promote trust, such as identified in IEEE 7000. SAE J3016 should be replaced as a regulatory standard by a simpler rubric which promotes safety with less complexity. Cooperation might take place by use of negotiated rulemaking.

Learning Fine-Grained Fact-Article Correspondence in Legal Cases

Automatically recommending relevant law articles to a given legal case has attracted much attention as it can greatly release human labor from searching over the large database of laws. However, current researches only support coarse-grained recommendation where all relevant articles are predicted as a whole without explaining which specific fact each article is relevant with. Since one case can be formed of many supporting facts, traversing over them to verify the correctness of recommendation results can be time-consuming. We believe that learning fine-grained correspondence between each single fact and law articles is crucial for an accurate and trustworthy AI system. With this motivation, we perform a pioneering study and create a corpus with manually annotated fact-article correspondences. We treat the learning as a text matching task and propose a multi-level matching network to address it. To help the model better digest the content of law articles, we parse articles in form of premise-conclusion pairs with random forest. Experiments show that the parsed form yielded better performance and the resulting model surpassed other popular text matching baselines. Furthermore, we compare with previous researches and find that establishing the fine-grained fact-article correspondences can improve the recommendation accuracy by a large margin. Our best system reaches an F1 score of 96.3%, making it of great potential for practical use. It can also significantly boost the downstream task of legal decision prediction, increasing the F1 score by up to 12.7%. The dataset and code will be released upon acceptance. Code and dataset are available at https://github.com/gjdnju/MLMN.

Strategies, Policies, and Standards in the EU Towards a Roadmap for Robust and Trustworthy AI Certification

Strategies, Policies, and Standards in the EU Towards a Roadmap for Robust and Trustworthy AI Certification

Is It Time to Consider EU Criminal Law Rules on Robotics?

This paper is devoted to issues which have not been sufficiently explored in European literature, and which have had fragmented consideration in comparative literature. These issues raise the question of whether the EU legislator should develop a framework of criminal law rules which would regulate the use of Artificial Intelligence (hereinafter: AI) in the near future, and what such rules should specifically address. The authors recognise two issues of particular importance for the future regulation of AI development within the EU, and offer their perspective on the areas which should be subjected to regulation in this regard. In order to provide a systematic overview of this topic, the paper starts with a description of the recent regulatory action of the EU in the field of AI, with special reflection on the Ethics Guidelines for Trustworthy AI. The authors then describe what are, in their opinion, the most important intersections of AI and criminal law in the broader sense, and in conclusion present their views of which areas should be specifically regulated by EU legislature in this context. Keywords: robot, Artificial Intelligence, criminal law, criminal procedure, autonomous, sanctions, European Union. This work is licensed under the Creative Commons Attribution − Non-Commercial − No Derivatives 4.0 International License. Suggested citation: I Vuletic and T Petrasevic, ‘Is it Time to Consider EU Criminal Law Rules on Robotics?’ (2020) 16 CYELP 225.

What a Philosopher Learned at an AI Ethics Evaluation

AI ethics increasingly focuses on converting abstract principles into practical action. This case study documents nine lessons for the conversion learned while performing an ethics evaluation on a deployed AI medical device. The utilized ethical principles were adopted from the Ethics Guidelines for Trustworthy AI, and the conversion into practical insights and recommendations was accomplished by an independent team composed of philosophers, technical and medical experts.

NSF AI institute for research on trustworthy ai in weather, climate, and coastal oceanography

NSF AI Institutes In 2019, the National Science Foundation (NSF) launched a new national investment in Artificial Intelligence (AI) to create a network of national AI institutes. Each institute will serve as a nexus of collaboration to create next-generation theory and applications of AI and to work synergistically with academia and industry. In the fall of 2020, NSF announced 5 new NSF AI institutes and 2 additional institutes funded by the United States Department of Agriculture (USDA) and the National Institute of Food and Agriculture (NIFA). Each institute is funded for approximately $20M over 5 years to make significant advances in foundational and applied AI research.

Welcome to AI matters 6(3)

Welcome to the last issue of this year's AI Matters Newsletter! We start with a report on upcoming SIGAI Events and Conference reports by Louise Dennis, our conference coordination officer. In our regular Education column, Todd Neller introduces the next AAAI/EAAI-2022 mentored undergraduate research challenge: AI-Assisted Game Design (AIAGD). We then bring you our regular Policy column, where Larry Medsker covers ongoing discussions on AI policy, this time with a special focus on Europe and the US. Finally, we close with two article contributions. The first article is by our former co-editor in chief Amy McGovern, about a newly established NSF-funded AI Institute for Research on Trustworthy AI in Weather, Climate, and Coastal Oceanography that she now directs. The second paper discusses the very timely topic of ways that AI can aid students in post-secondary education.

Culture by design

This article investigates a moment of the big data age in which artificial intelligence became a fixed point of global negotiations between different interests in data. In particular, it traces and explicates cultural positioning as an interest in the artificial intelligence momentum with an investigation of the unfolding of a European AI policy agenda on trustworthy AI in the period 2018–2019.

ETHICAL ASPECTS OF THE USE OF INNOVATIVE INFORMATION TECHNOLOGIES IN CLINICAL TRIALS

A clinical trial, according to the WHO, “is any research study that prospectively assigns human participants or groups of humans to one or more health-related interventions to evaluate the effects on health outcomes. Interventions include but are not restricted to drugs, cells and other biological products, surgical procedures, radiological procedures, devices, behavioural treatments, process-of-care changes, preventive care, etc”.
 The application of innovative information technologies like artificial intelligence and big data analytics in clinical trial processes is a new challenge. Such systems are useful tools, and promise to enhance the healthcare management, and to optimize clinical outcomes and economic effectiveness. However, their use raises ethical and social issues.
 In this direction, the European Commission in June 2018 set up the High Level Expert Group on AI, which offers guidance on a comprehensive framework for trustworthy AI. Trustworthy AI consists of three components, which should be met during the entire life cycle of the system: (1) it should be lawful, (2) it should be ethical, and (3) it should be robust.
 In this article we used the focus group methodology to obtain information from experts about the ethical aspects raised when innovative information technologies like artificial intelligence and big data analytics are used in clinical trials. Feedback from the experts was also gathered regarding the usage of the proposed guidelines for trustworthy AI, as an evaluation tool for the particular case of clinical trials.

Artificial Intelligence in economic decision making: how to assure a trust?

Motivation: The decisions made by modern ‘black box’ artificial intelligence models are not understandable and therefore people do not trust them. This limits down the potential power of usage of Artificial Intelligence. Aim: The idea of this text is to show the different initiatives in different countries how AI, especially black box AI, can be made transparent and trustworthy and what kind of regulations will be implemented or discussed to be implemented. We also show up how a commonly used development process within Machine Learning can be enriched to fulfil the requirements e.g. of the Ethics guidelines for trustworthy AI of the High-Level Expert Group of the European Union. We support our discussion with a proposition of empirical tools providing interpretability. Results: The full potential of AI or products using AI can only be raised if the decision of AI models are transparent and trustworthy. Regulations which are followed over the whole life cycle of AI models, algorithms or the products they using these are therefore necessary as well as understandability or explainability of the decisions these models and algorithms made. Initiatives on every level of stakeholders started, e.g. international level on the European Union, country level, USA, China etc. as well on a company level. The post-hoc local interpretability methods could and should be implemented by economic decision makers to provide compliance with the regulations.