Editorial: What “ethics” has to do with it (ethics-by-design)?

Abstract

After a year as director of JICES, I returned to observing the discussions in computer ethics. It is interesting to note that, fortunately, these debates, although often restricted to a specialist audience, no longer concern only the Western academic community. They now involve scholars not only in computer science, information systems and artificial intelligence (AI), but also in the pervasive applications of these disciplines.Among the many debates I have encountered while reading manuscripts received by the academic journal, the one that has prompted me most to reflect, and to write this editorial, is that there has been little discussion in AI of how ethics is conceived in this discipline. Even though it is at the centre of heated debates regarding its application, little reflection has been given to the epistemological justification of how ethical systems are implemented in AI. Are they really ethical systems? Now, my argument in this editorial is not to deny ethics per se, but to understand whether ethical methods – and in particular the philosophical legacy of the twentieth century – have been correctly used and whether the term “ethics” has been misused. At first glance, my argument may appear to be purely terminological and of no practical use. However, if you have the patience to read on, you will see that my objection actually has very significant implications.In AI ethics the scene seems to be dominated by the Ethics by Design approach. In 2026, the Ethics by Design (EbD) approach has established itself as the dominant operating methodology and fundamental requirement for the development of AI, transforming itself from a simple guideline to a binding regulatory standard: it has become the driving force behind compliance with the European Union’s AI Act. Ethics by Design, in fact, requires the integration of ethical, moral and regulatory requirements from the earliest stages of system design and development, rather than adding them as after-the-fact controls. This process is now an integral part of standard DevOps and MLOps protocols.Very importantly, this approach translates into concrete practices such as regular evaluation of training data to identify and correct discriminatory patterns. Second, the use of Explainable AI (XAI), which ensures that model decisions are explainable and transparent to users and regulators (which also means making it clear how a rule has been applied, not because that rule is right or should apply). Finally, human supervision (human-in-the-loop), which serves to maintain direct human control, especially for high-risk decisions in sectors such as healthcare and finance (which, coincidentally, together with the military and energy sectors, are the most sensible sectors from an economic point of view). The values included in Ethics By Design are not the result of random (or not) data collection or collection of people’s perception of what is right or wrong, but rather a deliberate selection of legal norms, philosophical principles and technical requirements that developers incorporate into the system from day one. In short, these are not just ethical principles, but a mix of moral norms, legal norms, corporate codes of ethics and technical standards.Given the heterogeneity of the sources, it is therefore difficult to see that the Ethics-By-Design is a kind of ethics as usually understood. In fact, sources of a different nature are being brought together. Let us proceed in order: first of all, ethics is confused with morality. In contemporary philosophy (albeit with all the necessary distinctions), “ethics” (since Moore, G.E., Principia Ethica, Cambridge, CUP, 1903) is used to refer to the study of values, rational thinking about values, prescriptiveness, and universalizability (Hare, R.M., The Language of Morals, Oxford: Oxford University Press, 1952). While “morals” is used to refer to the set of practical norms (values) applied by an individual or a group, that is the set of substantive rules that a society adopts. Again, ethics is the reflection on those values and critical thinking about them (i.e. in case of incompatibility between rules which one overwhelm others and on which basis?). Morals, on the other hand, concern the rigid norms systems of norms, and codes of conduct that, at that historical moment or in that society, tell us what is right or wrong in a normative sense. The distinction is indeed not universally accepted (in fact, in common, non-technical language, ethics and morals are interchangeable), but it is adopted here for analytical reasons. From this point of view Ethics by Design is not about ethics, but, essentially, a normative governance practice (reflection of a subject whatsoever) that seeks to anticipate morality (the laws and customs that will come). While EbD is operationally legitimate, conceptually it is not ethics in the strict sense; neither grasps the way in which ethics has been understood in modern study of ethics. EbB, in particular, does not establish duties, but translates them into technical specifications. This belongs to instrumental reasoning and is also very close to what in jurisprudence are called technical rules (which take the form of “if you want to achieve x, you must do y” – von Wright, G.H., Norm and Action. A logical enquiry, London: Routledge & Kegan Paul, 1963).Very importantly, if we accept such distinction, then the ethical elements are associated with other elements derived from power structures (laws, codes of ethics), inequalities (incorporated into many laws, and institutional and corporate behaviour) and social impacts, which are studied by sociology or cultural studies applied to technology, not by ethics. Inflating ethics weakens normativity. If everything is “ethical”, nothing really is (and it is mainly politics). As L. Introna (Management, Information and Power: A Narrative of the Involved Manager, London: Macmillan/Palgrave, 1997) has already pointed out, information systems (in which EbDs are implemented) are not neutral tools but incorporate normative (moral) assumptions, decision-making structures and power asymmetries, which are not only used through the systems but are also inscribed in their design and procedures. Therefore, Ethics by Design merely reinforce these asymmetries and power structures, even giving rise to suspicions of “ethical washing” (see Metzinger, T. (2019). Ethics washing made in Europe. Der Tagesspiegel. Available at Link to the cited article.).In ethical tradition (from Aristotle to Kant, up to contemporary metaethics), an ethical principle claims general validity that is also maintained by contextual or situated ethics, which requires a claim to extensibility (at least in similar conditions). An ethics, that structurally renounces universalizability (which is quite different from cultural imposition, but rather something that can be formulated as a principle valid beyond the specific case), does not provide general criteria of rightness, nor does it allow for intersubjectively binding moral judgements and, finally, it cannot establish obligations (or just apparently establishes them), but only assertions (i.e. a law prescribes a certain behaviour then you ought obey the law) in which the idea of justice pave the way to power.If a problem is truly ethical, then the way it is solved must be universalizable, even outside the technological domain in which it arises. Therefore, if an “AI ethics problem” cannot be generalised to similar cases in other areas (health, work, bureaucracy, education), then it is not an ethical problem, but a legal or organisational policy problem that fragments knowledge in an even more segmented way. Many issues currently framed as problems of AI or Information Ethics are in fact instances of broader legal, political or institutional problems. If their proposed solutions are not universalizable across domains, then their ethical framing is according to our proposal, conceptually mistaken. The ethical framing of technological issues should be justified, not assumed. When proposed solutions cannot be universalized beyond the technological domain in which they arise, the ethical label obscures rather than clarifies the normative nature of the problem. In this sense J. Moor’s famous proposal (Moor, J. H., What Is Computer Ethics? Metaphilosophy, 16(4), 1985, 266–275) of computers’ logical malleability (which now extends beyond the simple malleability of software) should be understood in this sense; universalizability becomes a central factor in the demand for (ethical) justice, not only in the field of computer science, and that must be achieved by striving to find a new operating method that goes beyond the much abused and intuitively easy-to-implement normative code of conducts.