…matters concerned with conduct and questions of what is good for us have no fixity…

Aristotle - Nicomachean Ethics, Book II Chapter 2

No course of action could be determined by a rule, because every course of action can be made out to accord with the rule.

Ludwig Wittgenstein - Philosophical Investigations §201

Modern society increasingly tries to make ethics fit its own structure, while quietly abandoning the activity that once made ethics meaningful. What now passes as moral seriousness, responsibility, or care is more often a question of alignment: alignment with targets, standards, models, risk frameworks, and approved objectives whose legitimacy is largely assumed rather than judged. This shift is rarely experienced as loss. It is made to feel like progress, as it is packaged as improvement, control and efficiency.

Alignment is formalizable, auditable and scalable, while judgment is contextual, embodied and cannot be made procedural without making it something other. Alignment promises firm ground. Judgment begins where the ground will not hold still. The problem is not the existence of forms, rules, or procedures, but that they are asked to bear ethical weight they cannot sustain once scale and speed sever them from lived judgment. Formalization does not erase judgment by necessity, but by accumulation, by gradually displacing where judgment is expected to occur. Under conditions where speed and coordination do dominate, judgment gradually loses its place, until ethical life comes to consist primarily in being correctly positioned within systems that has already decided what counts as correct or ethical. Meanwhile, the space for situationally lived ethics and wisdom, of which judgment is part, contracts.

In Book II of the Nicomachean Ethics, Aristotle argues that ethical inquiry does not admit of the same precision as geometry or mathematics, because the matters it addresses are not fixed in the way its objects are. This is a claim easily breezed past or overlooked, but one demanding renewed attention seen in the context of our predicament. Questions of conduct and of what is good for us unfold in situations that vary, that resist stabilization, and that change their character as circumstances shift. To ask for exactness here is already to misunderstand the domain, because the good life is not an object that can be derived from first principles and then applied. It is embodied, worked out in the midst of life, through deliberation that must remain sensitive to particulars that cannot be enumerated in advance.

If ethical life lacks fixity, then no set of rules, however refined, can substitute for the activity of judgment itself. Practical wisdom, phronesis, is not the application of general laws to specific cases, but the cultivated ability to see what matters here, now, under these conditions, with these stakes. This capacity cannot be externalized without distortion, nor compressed into procedures without loss, because it is inseparable from experience, responsibility, and the acceptance of uncertainty. Aristotle’s point is not merely that ethics is messy, but that its messiness is essential, that the space in which judgment operates is precisely the space that resists formal closure. It is in this context that his conclusion lands with its full weight: for the things about which decisions are made are those that cannot be determined by exact rules.

Alignment, by contrast, presumes that what matters has already been decided elsewhere, translated into objectives, constraints, or metrics against which action can be optimized. Where judgment interprets the situation, alignment enforces coherence with a prior abstraction. The appeal of alignment lies in its promise to remove the burden of judgment without abandoning the appearance of ethics, translating moral responsibility into a problem of specification and compliance. In a landscape of accelerating complexity, alignment offers a way to act decisively without having to decide meaningfully. With this shift, disagreement ceases to be a site of deliberation and becomes instead a deviation from alignment, and moral conflict comes to register as an implementation error, to be resolved through specification and tighter alignment, rather than judgment.

The temptation to replace judgment with alignment rests on something older than modern technology or bureaucracy, the thought that correctness can be guaranteed by adherence to rules if those rules are specified with sufficient care. This assumption feels reasonable because it mirrors how formal systems operate at small scales, where instructions appear to determine outcomes cleanly. Yet Wittgenstein (1953) shows that this confidence is misplaced in contexts outside the formalizable. His rule-following argument does not claim that rules are useless, but that rules are never self-sufficient. A rule, however precise, does not contain its own application: at some point, someone, some agency, must decide what counts as following it correctly. This means a moment of judgment that no rule or system of rules can absorb without remainder, and it is precisely this moment that systems of alignment tend to eliminate or conceal. Who is the “someone” once we scale the system past anyone’s ability to see what it is doing?

Wittgenstein’s point is that any course of action can be made to fit a rule if one is determined to interpret it in a certain way. If a rule truly dictated its application, there would be no room for disagreement about what it requires, no need for training, correction, or shared practice. Yet in reality, learning a rule always involves being initiated into a way of going on, a form of life in which certain responses are treated as appropriate and others are not. What guides action is therefore not the rule alone, but a background of judgment that the rule presupposes without being able to articulate. This is one of the hallmarks of a holistic view: we cannot separate something from its context, the whole, without loss.

This becomes decisive for ethical and institutional reasoning. When systems are built on the premise that alignment to rules or objectives can replace judgment, they quietly ignore the fact that interpretation has not disappeared, it has merely been displaced. Instead of being exercised openly and responsibly by agents situated in concrete contexts, interpretation is frozen into specifications, models, and procedures that are treated as neutral. The moment of judgment is concealed rather than eliminated, and correctness is attributed to the system itself, while the human activity that makes any rule meaningful is rendered invisible. Importantly, this does not constitute the actual neutralization of morality, but the creation of moral surplus with no corresponding ownership.

What Wittgenstein exposes, then, is not a failure of specification due to insufficient complexity in rules of alignment, but a category mistake. Rules function only within practices that already know how to go on, and they cannot generate that knowledge themselves. Alignment systems invert this relation, treating rules and objectives as primary and judgment as a residual source of error. The consequence is a form of moral life in which action can remain impeccably aligned while drifting ever further from the lived situations that give ethical reasoning its purpose.

Treat ethics as alignment and ethics becomes procedural. Responsibility migrates upward into abstract goal-setting and downward into rule execution, leaving no level at which someone must answer for whether the goals themselves remain appropriate to the situation at hand. This structural relocation of responsibility creates a distinctive moral imbalance: actions can be perfectly aligned and still deeply wrong, while no one involved experiences themselves as having acted wrongly. What is lost is not ethical intent, but ethical agency, the capacity to own decisions, inhabit their consequences, and revise ends in light of what they produce. Alignment is the attempt to eliminate the irreducible ethical remainder that Aristotle and Wittgenstein show must exist, a remainder that necessitates judgment.

This echoes what Jonas (1984) identified when he argued that the technological age demands an expanded ethics of responsibility. Where earlier moral frameworks could afford to focus on proximate intentions and local effects, technological action now unfolds at scales, durations, and levels of irreversibility that exceed any stable set of rules. Power has outrun foresight. Consequences propagate beyond the horizon of calculation, and yet responsibility cannot be deferred to systems, models, or procedures without dissolving altogether. Responsibility, under technological conditions, cannot be reduced to compliance with given norms or alignment with predefined goals, because what is at stake is precisely the question of whether certain powers should be exercised at all. Judgment becomes unavoidable where action reaches beyond prediction, and ethics becomes inseparable from the capacity to restrain, interrupt, or refuse action whose consequences cannot be responsibly borne.

Judgment is as such not only the capacity to discern what ought to be done, but the capacity to refuse, to interrupt a course of action even when it appears coherent, justified, or efficient. It binds agents to the consequences of their decisions in a way that cannot be delegated or dissolved into procedure, exposing them to the possibility that an action, however well-intentioned, should not proceed at all. Alignment systems are designed to preserve continuity, to redirect or constrain action without suspending it, and to ensure that moral questions are settled in advance through specification. In doing so, alignment does not resolve ethical conflict: it prevents it from appearing. By closing questions before they arise, it ensures that moral disagreement registers only as error, noise, or deviation, while insulating agents from inhabiting the consequences of what is done on their behalf. This contracts the space in which ethical judgment might still interrupt trajectories of action that remain correct and responsive, but also disastrously unexamined.

The displacement of judgment by alignment becomes especially visible once values are translated into targets. Campbell (1976) arrived at what has become known as Campbell’s Law: when a measure is used for decision-making, it is progressively corrupted by the very role it is meant to play. Once metrics become targets, institutions are compelled to defend them, because abandoning the metric would require reopening the question of what the institution is for. Alignment thus hardens because such failures would otherwise demand renewed judgment. Indicators initially introduced to approximate complex goods, quality of care, educational attainment, institutional trust, economic value and so on, gradually become the goods themselves. As this substitution takes hold, the task of judgment shifts from evaluating whether outcomes are good to verifying whether targets have been met. Alignment appears successful precisely because it no longer has to answer to the reality it was designed to represent. The system becomes increasingly correct in its own terms while drifting from the domain of meaning that once justified its existence.

Vallor (2016) argues that moral excellence cannot be reduced to states, outputs, or rule compliance. Virtue, in her account, is a sustained practice shaped by habituation, context, and ongoing responsiveness to others. It unfolds over time and remains sensitive to particulars that resist formal capture. Habituation cannot be shortcut by optimization, because virtues are not properties of outputs, but embodied capacities forged over time and experience. Alignment systems may reproduce the surface regularities of ethical behavior, but they bypass the formative processes through which judgment is cultivated. What Vallor describes at the level of moral formation is, at the level of moral culture, the condition MacIntyre (1981) diagnosed as the fragmentation of ethical life. For MacIntyre, once moral reasoning is severed from the practices and traditions that once sustained it, ends and means are separated, and ethical discourse survives only in the thin and often misleading form of rules, sentiments, and techniques of management detached from the practices that once gave them coherence. Alignment thrives in precisely such an environment, as only optimization remains when ends are taken as given or left implicit. Judgment, which depends on contested purposes and practical reasoning about the good, becomes unintelligible. What remains is a morality of means without ends, a system capable of coordinating action at scale while increasingly unable to say what any of it is for.

This is reinforced by the temporal conditions of late modernity. Rosa (2020) describes a world characterized by acceleration across technical, social, and experiential dimensions, a world in which responsiveness becomes compulsory while genuine answerability recedes. Acceleration does not simply compress time, it reorganizes our relation to the world so that continuous responsiveness replaces the possibility of being addressed in ways that could interrupt or transform action. Under these conditions, judgment comes to appear as friction, something that slows down processes that must remain in motion. Alignment offers a substitute that preserves responsiveness while eliminating the need for resonance, the capacity to be addressed by situations rather than merely reacting to signals. Systems remain active, adaptive, and sensitive to feedback, yet increasingly insulated from the question of whether what they are doing still makes sense. Judgment requires pauses that acceleration cannot afford, and so it becomes a prime candidate to be designed out.

Current approaches to the AI alignment problem inherit, almost without exception, the substitution of judgment with specification that now structures ethical life more broadly. Alignment is framed as the task of encoding values, constraints, or preferences into systems whose behavior can then be optimized, monitored, and corrected. The difficulty here is not of an engineering kind but of a conceptual kind: the assumption that ethical judgment can be decomposed into values, constraints, and preferences without remainder. In practice, however, ethical judgment arises only where action precedes specification, where agents must answer for what cannot be decided in advance, under conditions of uncertainty, conflict, and irreversibility. To ask whether an artificial system is aligned is therefore already to have displaced the ethical question it claims to address. This critique does not depend on speculative autonomous agents, as it applies already to decision-support systems, optimization engines, and hybrid arrangements where judgment is formally retained with humans-in-the-loop, but structurally sidelined.

The problem is not that alignment techniques are immature, but that they attempt to formalize something whose substance lies precisely in what resists formalization. Judgment lives in situation because we do. It requires immersion, not representation. AI systems operate entirely within representational worlds, however rich, where situations are reduced to inputs, outputs, and reward structures. What appears as moral behavior is the reproduction of patterns that resemble ethical action, abstracted from the conditions that make such action ethically intelligible. In practice, once systems demonstrate reliability at scale, human judgment is redefined as an exception-handling function, invoked only when alignment fails, rather than as an ongoing ethical activity. No increase in model capacity alters this basic structure, and neither does adversarial frameworks or agent assemblies with “judges”. A system can be aligned with objectives, norms, or preferences, but it cannot inhabit the space in which those objectives acquire their moral weight. And when systems are grown rather than engineered, their goals or desires are not something we get to choose, and not something we can predict.

The conclusion is that there can be no meaningful ethical alignment of AI in the sense the term is usually used. Ethics is not a property of behavior alone, nor a matter of convergence toward stable value functions. Alignment displaces the exposure to the real world by design, by producing systems that act correctly relative to given specifications while progressively narrowing the space in which those specifications can be questioned. What gets automated is not only action, but the question of whether the action should exist. The thing it claims to align against is not sitting there as an object, waiting to be encoded, but lives in the very judgment that alignment tends to remove. It is not so much that ethical AI is impossible, but that ethical alignment of AI misunderstands what ethics really is.

What emerges across these domains is a transformation in what counts as moral seriousness, accountability and responsibility. Alignment offers a way of acting in complex, accelerated systems without having to inhabit the burdens of judgment, uncertainty, and responsibility that ethical life has historically entailed. In doing so, it gradually reshapes our expectations of what ethics is for, recasting it as a problem of configuration rather than a practice of discernment. The result is a moral landscape in which actions can remain continuously justifiable while their meaning grows increasingly opaque.

That supposedly aligned systems will one day act against our values is one possible danger. However, the deeper danger is that we will lose the capacity to recognize when values themselves have become misaligned with the situations they are meant to address. As alignment comes to stand in for judgment, the space in which ends can be questioned, revised, or refused continues to atrophy and shrink. Ethics persists, but in a diminished form, stripped of the very activity that once gave it life. Ethical alignment does not fail because machines cannot share our values. It fails because ethics does not reside in values at all, but in judgment exercised under conditions that cannot be specified in advance. A society that entrusts its moral orientation to alignment may find itself impeccably correct and unable to decide what, if anything, remains worth deliberating and judging at all.

We are always immersed, never independent, never free to view ethics from nowhere. That is why judgment, and by extension wisdom, has the character it has, as it is not a method applied to life, it is life trying to see what life is doing. On the other hand, alignment treats ethics as though it were a diagram we could finalize, then inhabit, as though the world would hold still long enough for the diagram to become law. But the world never held still, and no ethics was ever built on firm footing.

References

Aristotle [350 BCE] (2009). Nicomachean Ethics. Oxford University Press.

Campbell, D. T. (1979). Assessing the Impact of Planned Social Change. Evaluation and Program Planning. 2 (1): 67–90.

Jonas, H. (1984). The Imperative of Responsibility. University of Chicago Press.

MacIntyre, A. (1981). After Virtue. University of Notre Dame Press.

Rosa, H. (2020). The Uncontrollability of the World. Polity Press.

Vallor, S. (2016). Technology and the Virtues. Oxford University Press.

Wittgenstein, L. (1953). Philosophical Investigations. Blackwell.