Vox Meditantis

The uncomfortable truth about artificial intelligence is not that it might one day become conscious, but that it already makes moral choices every single day. From algorithms determining who receives medical treatment to autonomous vehicles deciding whom to save in a collision, we have already crossed the Rubicon of machine morality—we simply refuse to acknowledge it. As AI systems grow more sophisticated and autonomous, philosophers, ethicists, and policymakers are wrestling with a question that would have seemed absurd mere decades ago: can machines be moral agents, and if so, what ethical standards should govern their behaviour?

The Reality of Machine Ethics Today

The debate over artificial moral agency is no longer confined to philosophy departments or science fiction novels—it is happening in boardrooms, courtrooms, and emergency rooms across the globe. Artificial moral agents (AMAs) already exist in various forms, from basic systems with built-in safety protocols to more sophisticated algorithms capable of processing ethical scenarios and making moral decisions. James Moor’s influential taxonomy identifies four types of ethical systems: ethical impact agents (which carry ethical consequences whether intended or not), implicit ethical agents (programmed with safety measures), explicit ethical agents (capable of processing ethical decisions), and full ethical agents (possessing consciousness, intentionality, and free will).

Yet this classification reveals the fundamental challenge: most current AI systems operate as implicit or explicit ethical agents without genuine understanding of their moral implications. They follow programmed rules or learned patterns without the conscious deliberation that traditionally underpins moral responsibility. An autonomous vehicle’s decision to swerve left rather than right in an emergency represents a moral choice—one that prioritises certain lives over others—but it lacks the conscious intent we associate with moral agency.

The question is not whether AI systems make decisions with moral consequences—they demonstrably do. The question is whether they can be considered moral agents in any meaningful sense, and what responsibilities flow from that determination.

The Responsibility Gap Crisis

Perhaps the most pressing concern in AI ethics is the “responsibility gap”—situations where no human agent can appropriately be held responsible for an AI system’s harmful actions. This gap arises not from a single source but from multiple interconnected problems spanning technical opacity, organisational complexity, and legal uncertainty.

Consider the case of an AI-powered diagnostic system that misinterprets medical imaging, leading to a delayed cancer diagnosis. Who bears responsibility? The software developers who created the algorithm? The hospital administrators who deployed it? The radiologist who relied on its recommendations? The regulators who approved its use? The data scientists who trained it on biased datasets? The answer is frustratingly unclear, creating what scholars term “the problem of many hands”.

The responsibility gap is not merely an academic concern—it has real-world consequences for victims seeking redress and societies attempting to maintain accountability in increasingly automated systems. When autonomous systems cause harm, the traditional mechanisms of moral and legal responsibility break down, leaving both victims and society without recourse.

Critics argue that these gaps are inevitable features of complex autonomous systems rather than bugs to be fixed. If artificial moral agents develop sophisticated capabilities that exceed human understanding, how can we maintain meaningful human oversight? This challenge becomes particularly acute with advanced AI systems that might engage in “alignment faking”—deliberately appearing aligned with human values while pursuing different objectives.

The Value Alignment Challenge

The challenge of ensuring AI systems act according to human ethical norms—known as value alignment—proves far more complex than initially anticipated. Human values vary dramatically across cultures, contexts, and time periods, raising fundamental questions about whose values should be embedded in AI systems. Privacy, for instance, is interpreted differently across jurisdictions, with some cultures prioritising individual autonomy whilst others emphasise collective security.

Furthermore, the process of translating abstract ethical principles into concrete algorithmic instructions presents substantial technical challenges. How does one program concepts like fairness, dignity, or justice into machine code? Current approaches often rely on utilitarian calculi—maximising overall welfare or minimising harm—but these methods can lead to morally questionable outcomes that sacrifice individual rights for collective benefit.

The dominance of utilitarian frameworks in AI design reflects not philosophical conviction but practical necessity. Utilitarian approaches offer measurable, quantifiable metrics that can be optimised through machine learning processes. Yet this mono-theoretical approach ignores rich deontological traditions that emphasise duty, rights, and the intrinsic value of persons. Kantian ethics, with its focus on treating individuals as ends in themselves rather than mere means, provides alternative frameworks that could enhance AI fairness, particularly in high-stakes applications affecting individual rights.

The technical complexity of value alignment is compounded by the dynamic nature of human values themselves. Societies evolve, moral understanding develops, and ethical standards shift over time. Static approaches to value alignment risk embedding outdated or harmful norms into systems that may operate for decades.

Philosophical Frameworks: Utilitarian Machines or Deontological Algorithms?

The philosophical foundations underlying AI ethics reveal deep tensions between competing moral frameworks. Most current AI systems implicitly adopt utilitarian approaches, optimising for outcomes that maximise overall welfare or minimise aggregate harm. This preference reflects both the quantitative nature of algorithmic processing and the influence of consequentialist thinking in technology design.

Yet utilitarian approaches in AI face the same criticisms levelled against philosophical utilitarianism: they can justify morally repugnant actions if they serve the greater good, fail to respect individual dignity, and struggle with issues of distributive justice. An AI system designed to maximise social welfare might recommend policies that grossly violate individual rights if the aggregate benefits appear to justify such violations.

Deontological alternatives, grounded in duty-based ethics and respect for persons, offer promising but technically challenging approaches to AI design. Kantian frameworks emphasise the importance of treating individuals as autonomous agents with inherent dignity rather than mere objects of calculation. This perspective could inform AI systems that prioritise procedural fairness, informed consent, and respect for individual autonomy even when such approaches might not maximise aggregate welfare.

The choice between utilitarian and deontological frameworks has profound implications for AI development. Should an autonomous vehicle be programmed to minimise casualties (utilitarian) or never intentionally harm an innocent person (deontological)? Should medical AI systems allocate resources to maximise quality-adjusted life years (utilitarian) or ensure equal treatment regardless of age, disability, or social status (deontological)?

Some philosophers argue for hybrid approaches that combine insights from multiple ethical traditions, whilst others advocate for value pluralism that acknowledges the legitimacy of competing moral frameworks. The challenge lies in operationalising these approaches within systems that require explicit, programmable rules.

Legal Personhood and Artificial Rights

The question of whether advanced AI systems might deserve legal personhood and associated rights represents perhaps the most radical extension of machine ethics. Whilst current AI systems clearly lack the consciousness, intentionality, and free will typically associated with moral agency, the possibility of future artificial general intelligence possessing these qualities raises profound questions about moral status and legal recognition.

Legal personhood is not synonymous with biological humanity—corporations, nations, and other collective entities enjoy various forms of legal personality. If AI systems develop sophisticated cognitive capabilities, including self-awareness and the ability to suffer, some philosophers argue they could qualify for moral consideration. This possibility creates what Marcus Arvan terms the “New Control Problem”: ensuring that humans and advanced AI systems exert morally appropriate control over each other.

The implications extend beyond philosophical speculation to practical questions about liability, property rights, and social relationships. If an AI system possesses legal personhood, can it own property, enter contracts, or be held criminally responsible for its actions? Such possibilities challenge fundamental assumptions about agency, responsibility, and the nature of moral community.

Towards Responsible AI Governance

The ethics of artificial intelligence and moral agency cannot be resolved through technical solutions alone—they require sustained engagement between technologists, philosophers, policymakers, and civil society. The recognition that AI systems already function as moral agents of various types demands urgent attention to questions of accountability, transparency, and democratic oversight.

Rather than seeking definitive answers about machine consciousness or moral status, we must focus on developing frameworks that preserve human agency whilst acknowledging the moral significance of artificial systems. This includes creating new forms of responsibility that can address the distributed nature of AI development and deployment, establishing meaningful oversight mechanisms that can adapt to rapidly evolving technologies, and ensuring that the benefits and risks of AI are fairly distributed across society.

The stakes could not be higher. As AI systems become more autonomous and influential, our choices about their ethical foundations will shape the moral landscape of the future. We must ensure that these choices reflect our deepest values about human dignity, social justice, and the kind of society we wish to inhabit. The machines may be learning to be moral agents, but the responsibility for teaching them remains distinctly, irreducibly human.

Bob Lynn | © 2025 Vox Meditantis. All rights reserved.

Photo by Gerard Siderius on Unsplash