'Reconstructing AI Ethics Principles: Rawlsian Ethics of Artificial Intelligence' by Salla Westerstrand in (2024) 30 Science and Engineering Ethics comments
The recent popularisation of Artificial Intelligence (AI) technologies has prompted discussion about their ethical implications. Many have speculated over the potential of AI systems to threaten societal structures and quality of life (Bostrom, 2016, 2017; Coeckelbergh, 2022a, 2024; Russell, 2019). This development has pushed organisations to steer the development and application of AI systems towards a more ethical direction through principles and guidelines (see Jobin et al., 2019; Hagendorff, 2020; Ayling & Chapman, 2022). This trend also known as principalism (Clouser & Gert, 1990) has raised questions about the applicability of these principles in practice (Hickok, 2021; Mittelstadt, 2019), which has led to an increasing volume of research about operationalisation of ethics principles (Bleher & Braun, 2023; Morley et al., 2021, 2023; Stix, 2021). Whereas research around operationalising is essential in order for the principles to effectively guide AI development and deployment, I argue that the work around actionable ethics principles needs revisiting.
Many of the principles in existing guidelines are overlapping (Ashok et al., 2022; Hagendorff, 2020; Hunkenschroer & Luetge, 2022; Jobin et al., 2019), but they also differ in ways that bring forth challenges. They seem to differ in how principles are interpreted (Jobin et al., 2019) and seem to rarely guide the reader through the reasoning leading to the principles (Franzke, 2022; Jobin et al., 2019). Whereas the reviews such as those of Jobin et al., (2019) and Hagendorff (2020) map principles and values found in the guidelines, the very use of ethics behind the guidelines remains obscured (Franzke, 2022). In common language, one might talk about ethical questions while referring to predefined rights and wrongs in the given normative context. As Stahl (2022) points out, such issues could be better described as social concerns rather than issues of ethics. An ethicist, in contrast, could ask: Why are the suggested principles the most ethical ones? If the guidelines do not offer justifications, are the guidelines truly ethics guidelines, or just values or opinions, or a result of a political processes? As Bleher and Braun (2023) put it, if we do not ground our principles into rigorous ethical reasoning, they and the tools for their operationalisation risk being”either inappropriate, meaningless, or merely an end in themselves” (p. 10).
Moreover, Hagendorff (2020) observed that most AI ethics guidelines he reviewed tend to neglect impacts of AI on democracy, governance, and political deliberation. Meanwhile, several studies imply that the current developments in AI might threaten democracy at large (Coeckelbergh, 2022a, 2024) and harm democratic governance by distorting political opinion formation and elections (Alnemr, 2020; Chesney & Citron, 2019; Feezell et al., 2021; König & Wenzelburger, 2020; Manheim & Kaplan, 2019; Nemitz, 2018; Paterson & Hanley, 2020), eroding trust towards democratic institutions (Chesney & Citron, 2019; Manheim & Kaplan, 2019; Paterson & Hanley, 2020), and violating fundamental democratic values, such as equality and justice (Hacker, 2018; Janssen et al., 2022; König & Wenzelburger, 2020; Tolan, 2019). As AI systems with ever broader collective impacts on societies get popularised, such as generative AI systems, a need for shift in perspective from human-in-the-loop towards society-in-the loop suggested by Rahwan (2018) gains ever more relevance. Ethics as an approach seems to have potential for increasing our understanding of the relationship between AI and democracy (Westerstrand, 2023), and Stahl also noted a connection between the ongoing AI ethics discourse with regulation, such as human rights (Stahl, 2022). This implies that to improve the state of AI systems and their alignment with collective values, it might be wise to inspect the larger ecosystem and the ethical foundations behind its guiding principles.
In this paper, I contribute to the discussion through the perspective of John Rawls’s theory of justice as fairness. Rawls intended his contractarian theory to build “the most appropriate moral basis for a democratic society” (Rawls, 1971, p. viii), which makes it a theoretically interesting foundation for discussing the ethicality of AI development that comes with an increasing collective ethical and societal implications. By taking a relatively unpopular direction and contributing to the principalist paradigm in contrast to several other applications of Rawls’s theory concentrating on algorithmic applications (e.g., Leben, 2017, 2018; Heidari et al., 2019; see also Keeling, 2018), this paper aims to strengthen the ethical rigour of the principalist discourse that keeps informing the technical development of algorithms, the spheres of policy and regulation, as well as strategic decision-making regarding which technologies we should develop and deploy in the first place, and in which contexts.
The paper begins with an overview of Rawls’s theory of justice in the context of AI, which serves as a theoretical starting point for revisiting AI ethics guidelines. Then, a suggestion is drafted for a set of ethics guidelines that is based on Rawls’s principles of justice as fairness. Finally, conclusions are drawn, the academic and practical potential of the work are discussed, and possibilities for future research and application are identified.
For fairness
Developers and deployers of an AI system must ensure that the AI system does not threaten the basic liberties of any individual. AI systems should not endanger but support the freedom of thought and liberty of conscience. AI systems should not compromise but support political liberties and freedom of association, such as the right to vote and to hold public office. AI systems should not harm but support the liberty and integrity of the person, including freedom from psychological oppression and physical assault and dismemberment. ➵ All AI systems should be aligned with the principle of rule of law.
The use and development of AI systems should not negatively impact people’s opportunities to seek income and wealth. If an AI system is used in distribution of advantageous positions, such as recruitment, performance evaluation, or access to education, it needs to be ensured that. The tool is trained with non-biased training data, or appropriate tools are used to mitigate the biases in the final product if no non-biased training data is available (data bias mitigation), The outcome of the use of the tool includes an explanation of the grounds for the outcome it produces (explainability), and. The algorithms used shall encourage neither biased results nor the systematic repetition and amplification thereof in, e.g., the feedback loops of a machine learning system (algorithmic bias mitigation). If these conditions cannot be met, AI should not be used in the process.
All inequalities affected by AI systems, such as acquiring a position of power or accumulation of wealth, must be to the greatest benefit of the least advantaged members of society.
