A High-Level Overview of AI Ethics

Emre Kazim - University College London;

Adriano Koshiyama - University College London

-- The ethics of artificial intelligence (AI) had risen in importance with the increased development and seemingly universal adoption of automated systems. Spurred by high-profile cases of harm, such as voter manipulation through algorithmic ‘nudging’ technologies, to bias in recidivism calculation and facial recognition in policing technologies, there is considerable social concern and calls for regulation and standards in this space (see our overview, "A Review of the ICO’s Draft Guidance on the AI Auditing Framework"). As a result, a voluminous literature has emerged from different stakeholders in the area, namely industry, academia, and civil society groups and NGOs (see our paper, "Human Centric AI: A Comment on the IEEE’s Ethically Aligned Design," with respect to NGOs). We can also think of the literature as reflecting either an engineering or a social scientific perspective, where the former considers issues of ethical by-design and the latter more policy, philosophy, law and the broader humanities (ex. sociology, anthropology, etc.). As such, we believe that ‘AI Ethics’ as a term and discipline requires clarification and unification in a manner that reflects this spectrum of interdisciplinarity.

In response to this, we authored our most recent paper, "A High-Level Overview of AI Ethics" to introduce the basic concepts and sketch the approaches and central themes currently employed in AI ethics. Our paper is divided into four parts:

  1. Definitions: In this section we offer and explicate definitions of key terms. As a point of departure, we begin by defining first ‘digital ethics’ and then ‘AI ethics’. Following this we then move on to defining how the terms ‘digital’ and ‘AI’ are being used. Further, we explain how the term ‘ethics’ is being used in this context and expand upon the dominant ethical philosophies that AI ethics draws upon. This section closes with an exploration of ‘Human Centric AI’, which we take to be the overarching value framework of AI ethics

  2. Predecessors to AI Ethics: AI ethics is an emergent field that is still in its nascent phase. However, there are a number of disciplines that have long traditions and literature from which AI ethics draws and it can be seen as, in various ways, as a continuation of them. The three bodies of literature most relevant are, i. engineering ethics, ii. philosophy of technology, and iii. science and technology studies.

  3. Three Approaches - Principles, Processes and Ethical Consciousness: the two main approaches to AI ethics have been a principles approach, which can be read broadly as an attempt to guide and structure the uses of the technology (thereby mitigating the risk of misuse) and an ethical-by-design approach, which seeks to mitigate the harms that result from design flaws. In this section we explore these two approaches and also a third, denoted as ‘ethical consciousness’, which draws from the business ethics literature and concerns a need to institute particular structures and shifts in cultures, attitudes and norms of those who use, develop and deploy AI systems.

  4. Major Themes in AI Ethics: There are many terms and phrases that have emerged within the AI ethics literature. For example, a comprehensive review of AI ethics guidelines found eleven ‘ethical principles’, which were identified through frequency of the terms (and their synonyms) in the literature. For the purposes of our overview we draw on the growing engineering expertise that overlaps with the ethics principle space. Indeed, in this section, we identify and explore six themes, these are human agency and oversight, safety, privacy, transparency, fairness, and accountability. Drawing on the European Commission’s Ethics Guidelines for Trustworthy AI , we note that these themes can be read as falling under the umbrella of ‘Trustworthy AI’.

We conclude by summary and noting future directions and debates, namely;

  • Data Ethics and AI: given the substantial literature, practice and regulation around data ethics, we anticipate that the relationship between data ethics and AI will increase in importance, both in conceptually and in terms of regulatory/practical consequences.

  • Legal Status of Algorithms: the legal status of an algorithm, with respect to responsibilities and obligations of those developing and deploying them, is likely to raise a number of complex questions regarding the nature of legal culpability and even questions of agency and personhood.

  • AI and the Economy: we believe that the relationship between AI and the economy will become a major theme of AI ethics. In addition to the current discussions of automation and the loss of labour, which allay into questions such as universal basic income, etc., there are broader questions regarding taxation of AI systems, national and international procurement standards and strategies, and the strategic importance of AI in national budgets.

Copyright © 2018 All Rights Reserved. Faculty of Law, The Chinese University of Hong Kong

The Chinese University of Hong Kong