What Is Machine Ethics?

Question

What is machine ethics, and why does it matter now? The question arises because machines increasingly make decisions with moral consequences: an autonomous vehicle decides who to harm in an unavoidable collision, a medical algorithm decides who receives scarce treatment, an automated system decides who is denied credit or flagged for surveillance. Machine ethics is the discipline that asks how these decisions should be made — and who, or what, is responsible for them.

Quick Answer

Machine ethics is the field devoted to the moral behavior of artificial agents. It has two questions. First, the engineering question: how can we build machines that behave ethically? Second, the philosophical question: can machines be moral agents at all, and if not, what follows? The answers range from treating machines as tools whose ethics is entirely the responsibility of their designers and users, to treating sufficiently sophisticated machines as artificial moral agents (AMAs) whose behavior can meaningfully be judged as right or wrong. The practical consensus is that regardless of the philosophical status of machines, the decisions they make have moral consequences, and the design of those decisions is an ethical task that cannot be avoided.

Historical Wisdom

The question of whether machines can be moral is older than computers. The philosophical tradition has asked whether moral agency requires freedom, reason, consciousness, and the capacity to be moved by reasons — and if so, whether anything non-human can possess these. Aristotle defined virtue as a settled state of character acquired through practice and issuing in deliberate choice; Kant located morality in the rational will and the capacity to act from the moral law; the utilitarian tradition located it in the capacity to experience pleasure and pain. Each of these accounts raises different questions for machines: can a machine have character, a will, or interests? What the tradition supplies is a set of criteria for moral agency — autonomy, rationality, intentionality, responsibility — against which the claims of machine ethics must be measured.

Philosophical Perspectives

The central debate in machine ethics concerns artificial moral agency. The skeptical position holds that machines cannot be moral agents, because moral agency requires capacities machines lack: consciousness (what it is like to be the machine), intentionality (meaning what one does), and responsibility (being the appropriate target of praise and blame). On this view, a machine's "decisions" are really the decisions of its designers and operators, and machine ethics reduces to human ethics — the ethics of building and using machines. This is the position of many critics of autonomous weapons and algorithmic decision systems: the machine is a tool, and the moral responsibility lies with the humans who deploy it.

The functionalist position holds that moral agency is a matter of behavior, not inner experience. If a machine reliably produces morally acceptable behavior — if it stops at crosswalks, avoids harming people, treats patients fairly — then it is a functional moral agent, and the philosophical question of whether it "really" means anything by its behavior is secondary. This position is natural for the engineering community, and it motivates the two main approaches to building ethical machines.

The top-down approach implements ethical principles directly: program the machine with rules (a version of deontology), or with a value function (a version of consequentialism), and let it reason from the principles to decisions. The bottom-up approach trains moral behavior: give the machine examples of good and bad decisions, reward good behavior, and let it learn patterns of ethical conduct the way a child learns social norms. Each approach has known weaknesses: top-down systems cannot handle the complexity and conflict of real moral life; bottom-up systems inherit the biases and failures of their training data. The frontier of machine ethics is the attempt to combine them.

The deepest philosophical question is raised by the very idea of a machine that decides. If a machine is deterministic, in what sense can it choose? If it is stochastic, in what sense is its randomness free? Philosophers such as the researchers who study "moral algorithms" argue that we should not think of machines as moral agents at all, but as components in a socio-technical system of responsibility: the machine makes the decision, but the responsibility is distributed across designers, deployers, operators, and regulators. On this view, machine ethics is the discipline of designing that distribution well.

Lessons From Thinkers

The lessons of machine ethics are these. From the skeptics: do not be distracted by talk of machines "deciding" — the question is always who built the machine, who deployed it, and who benefits and suffers from its behavior. From the functionalists: behavior is what matters in practice, and a machine that reliably does the right thing is morally preferable to a machine that means well and fails. From the top-down tradition: principles are necessary but not sufficient — the moral life is not a program. From the bottom-up tradition: machines learn ethics from us, and a machine trained on our actual behavior will learn our actual vices. And from the distribution-of-responsibility literature: the task is not to make machines responsible but to make the human systems around them responsible — transparent, accountable, and corrigible.

Practical Application

Machine ethics has immediate applications. In autonomous vehicles, it faces the trolley-style question of how the vehicle should allocate unavoidable harm — and, more importantly, the question of who decides the allocation and how it is made public. In healthcare, it governs the use of algorithms for diagnosis, triage, and resource allocation. In the criminal justice system, it governs risk assessment tools that influence bail and sentencing decisions. In finance, it governs automated trading and credit decisions. In the military, it governs the development and use of autonomous weapons. The practical lesson is that every deployment of a decision-making machine is an occasion for machine ethics: the values encoded in the machine, the accountability structures around it, and the feedback mechanisms for correcting its errors are all ethical design decisions that must be made explicitly.

Quotes

The aspiration of machine ethics is stated by its founders, Michael and Susan Leigh Anderson: "We want machines that are not just good at achieving their objectives, but that also behave ethically in doing so." The skeptical warning is stated by the computer scientist Joseph Weizenbaum: "There are some things that a computer should never be asked to do." And the practical urgency is captured in the observation that "the question is not whether machines can think, but whether they can decide — and they already decide."

Related Questions

• What Is the Alignment Problem? — the broader problem of ensuring AI acts in accordance with human values.
• What Is the Turing Test? — the classic question of machine intelligence.
• What Is the Chinese Room Argument? — Searle's challenge to machine understanding.
• What Is Ethics? — the foundations of the moral theories machine ethics applies.