AI Governance Research Articles

AI Governance and the Policymaking Process: Key Considerations for Reducing AI Risk

This essay argues that a new subfield of AI governance should be explored that examines the policy-making process and its implications for AI governance. A growing number of researchers have begun working on the question of how to mitigate the catastrophic risks of transformative artificial intelligence, including what policies states should adopt. However, this essay identifies a preceding, meta-level problem of how the space of possible policies is affected by the politics and administrative mechanisms of how those policies are created and implemented. This creates a new set of key considerations for the field of AI governance and should influence the action of future policymakers. This essay examines some of the theories of the policymaking process, how they compare to current work in AI governance, and their implications for the field at large and ends by identifying areas of future research.

Should We Regulate A.I.? Can We?

Artificial intelligence is viewed by many as the defining technology of the 21st century. But how can we ensure that its benefits outweigh the potential risks? This think piece takes the one-year anniversary of the first pedestrian death caused by an autonomous vehicle to consider the question of whether and how regulation can mitigate the dangers of AI. It discusses in particular the Model AI Governance Framework adopted in January 2019 by Singapore’s Personal Data Protection Commission, and the possibilities and limitations of its guiding principles of human-centricity and transparency.

Data Trusts as an AI Governance Mechanism

This paper is a response to the Singapore Personal Data Protection Commission consultation on a draft AI Governance Framework. It analyses the five data trust models proposed by the UK Open Data Institute and identifies that only the contractual and corporate models are likely to be legally suitable for achieving the aims of a data trust. The paper further explains how data trusts might be used as in the governance of AI, and investigates the barriers which Singapore's data protection law presents to the use of data trusts and how those barriers might be overcome. Its conclusion is that a mixed contractual/corporate model, with an element of regulatory oversight and audit to ensure consumer confidence that data is being used appropriately, could produce a useful AI governance tool.

Toward the Agile and Comprehensive International Governance of AI and Robotics [point of view

Rapidly emerging technologies, such as AI and robotics, present a serious challenge to traditional models of government regulation. These technologies are advancing so quickly that in many sectors, traditional regulation cannot keep up, given the cumbersome procedural and bureaucratic procedures and safeguards that modern legislative and rulemaking processes require. Consequently, regulatory systems will predictively fail to put in place appropriately tailored regulatory measures by the time new applications of fast-moving technologies begin to affect society. Perhaps even worse, if a regulatory system does somehow manage to rush into place new regulations for an emerging technology, they will likely be obsolete by the time the ink dries on the enactment. Given this so-called “pacing problem,” traditional regulatory approaches will either produce no regulation or bad regulation [1] .

The middle-out approach: assessing models of legal governance in data protection, artificial intelligence, and the Web of Data

ABSTRACTAll models of legal governance and most regulatory options have to do with ‘top-down’ solutions as an essential ingredient of the approach. Such models may include ‘bottom-up’ forms of self-regulation, such as in forms of ex post regulation, or unenforced self-regulation. This paper focuses on what lies in between such top-down and bottom-up approaches, namely, the middle-out interface of the analysis. Within the EU legal framework, this middle-out layer is mainly associated with forms of co-regulation, as defined by Recital 44 of the 2010 AVMS Directive and Article 5(2) of the GDPR. However, there are also additional models on how we should grasp the middle-out layer of legal regulation, as shown by the debates on the governance of AI and the Web of Data. For example, the debates on issues such as monitored self-regulation, coordination mechanisms for good AI governance, and ‘wind-rose’ models for the Web of Data make it clear that co-regulation is not the only alternative to both bottom-up and top-down approaches. From a methodological viewpoint, the middle-out approach sheds light on three different kinds of issues that regard (i) how to strike a balance between multiple regulatory systems; (ii) how to align primary and secondary rules of the law; and (iii) how to properly coordinate bottom-up and top-down policy choices. The increasing complexity of technological regulation recommends new models of governance that revolve around this middle-out analytical ground.

AI Governance: Digital Responsibility as a Building Block

AI Governance: Digital Responsibility as a Building Block

Transformative AI Governance and AI-Empowered Ethical Enhancement Through Preemptive Simulations

As the problem-solving ability of AI increased significantly during this decade, its scope of application has been extended to various areas including ethically relevant fields such as the development of autonomous systems. In this contribution, we evince why for the purpose of effective AI governance, humans have to quantitatively specify their ethical conceptions within a consequential framework. Thereby, the implementation of advanced AI systems not only forces society to provide machine-understandable ethical goal functions, but it also simultaneously facilitates a new transformative socio-technological feedback-loop with the potential for a dynamic ethical enhancement at the societal level. Furthermore, we exemplify why a common objection to consequentialism related to impossibility theorems does not represent an argument against the feasibility of such a consequential framework with ethical goal functions despite the soundness of these theorems. Finally, we elaborate on how AI (and broader science and technology) might equip humans with a novel particularly powerful preemptive tool within a socio-technological feedback-loop: the ability to get access to a simulated first-person experience of future states of the world and the estimation of the related – as we term it – artificially simulated future instant utility.

Managing Artificial Intelligence within a Digital Density Framework

AI is a new technology that enables organizations to leverage their data to create new value propositions. At the same time, the introduction of AI into the organization translates into new challenges, since the data-both as input and as output of AI models-require specific governance. This chapter outlines the holistic guidelines that general management should follow to achieve good AI governance within a digital density framework.

Natural Law and its Implications for AI Governance

Natural Law and its Implications for AI Governance

The Necessity of Auditing Artificial Intelligence Algorithms

Artificial intelligence (“AI” ) is the industrial revolution of the information age. Just as the industrial revolution automated physical tasks, AI automates cognitive processes. Governments sought to limit the harms and maximize the benefits of the industrial revolution by regulating the vehicle through which tasks were automated—namely, product manufacturing. Because today’s AI revolution automates cognitive processes through algorithms, any effective AI governance framework will require that regulators be able to audit the algorithms that automate these cognitive processes. Rather than propose a full AI governance framework, this paper explains why audits of AI algorithms will be a necessary component of any successful governance framework. First, the opaque nature of AI algorithms shields its results and processes from meaningful review. Second, such a review is necessary to identify and correct the root causes of problematic behavior. Third, the need for meaningful review outweighs the intellectual property interests in the AI. Although the final version of AI governance may take many forms, each proposal will need to include a mechanism through which AI algorithms can be tested and monitored for validity.

Before Machines Consume the Consumers: High-Level Takeaways from the Artsy Project

Business is using artificial intelligence in essentially all sectors of the economy: machine learning is employed to generate knowledge out of big (consumer) data, to optimize processes and to undertake new, previously impossible tasks. This might come with benefits for the consumers and the society, but definitely poses numerous challenges. In this short note, we provide an overview of challenges for EU consumer law and policy stemming from the business’s use of AI, and sketch a plan for action. We argue that AI will need to be tackled case-by-case, bottom-up, though with the big picture in mind. It needs to be tackled soon, but we do need to take our time to reconsider the assumptions that have been challenged, and not rush to political conclusions. Moreover, we argue that the role of law is not just to minimize the risks, but also to enable developments in consumer-empowering AI tools. These tools, however, will not be given to us by business. Civil society must take action and fight for them. We cluster the challenges and takeaways by the type of actors that are affected by the business’s use of AI. Consumers face the risk of undue influence on their behavior (by targeted personalized commercial practices), exclusion from access to goods and services (ad delivery & price discrimination) and lower quality of information and services in the interaction with artificial agents. Regulators need to revise their governance toolbox, taking into account the specificity of AI’s operations (stealth infringement, wide-spread-by-minor damage, automation of reasoning). Also, regulation needs to strike a correct balance between specific cases and the bigger picture, and between commanding and enabling. In EU consumer law, the concepts of unfair commercial practices and unfair contractual terms should be revisited, to take into account the reality of business using AI. In addition, we should consider the usefulness of adopting special data protection rules to supplement the GDPR, by stating what purposes of data processing are lawful in what markets. Civil society should strive to seize the opportunities of AI in the medium-term, making the best use of the exiting legal instruments (UCPD, UCTD, GDPR) in the short-term, and lobby for societal and legal change in the long-term. Finally, academia, in particular legal scholars, must re-consider their role in the debate on AI governance: they should ground their research in empirical findings, acknowledge the limitations of sectoral knowledge and remedy such limitations by engaging in an interdisciplinary and multi-stakeholder dialogue. We argue that the competitive advantage of scholars goes beyond offering concrete policy recommendations. Instead, it concerns a critical reflection on the ways in which the mass deployment of AI challenges the basic assumptions and presuppositions of the existing legal and regulatory theory and practice.

Governing AI

Ian Fish FBCS, Acting Chair of the Information Security Specialist Group, considers current and future AI developments and the security concerns around AI governance.

A Layered Model for AI Governance

AI-based systems are “black boxes,” resulting in massive information asymmetries between the developers of such systems and consumers and policymakers. In order to bridge this information gap, this article proposes a conceptual framework for thinking about governance for AI.