Moral realism

Based on Wikipedia: Moral realism

In 2020, a rigorous study of the philosophical community revealed that 62.1% of respondents accept or lean toward the view that moral facts exist independently of human opinion. This is not a mere statistical curiosity; it represents a profound commitment to the idea that when we say "slavery is unjust" or "torture is wrong," we are not merely expressing our personal distaste or cultural conditioning, but reporting on the fabric of reality itself. This position, known as moral realism, stands as the bedrock for anyone arguing that AI alignment is not just a technical puzzle, but a moral imperative grounded in objective truth. If morality were merely a matter of subjective preference or social convention, the project of aligning artificial intelligence with human values would dissolve into a game of arbitrary preference matching. But if moral realism holds, then there is a correct answer to the question of how an AI should behave, and that answer exists whether any human or machine currently recognizes it.

Moral realism, also termed ethical realism, asserts that ethical sentences express propositions that refer to objective features of the world. These features are independent of subjective opinion. Consequently, some of these moral propositions can be true to the extent that they accurately report these features. This stance makes moral realism a non-nihilist form of ethical cognitivism. Cognitivism is the meta-ethical view that ethical sentences express propositions and can therefore be true or false. Realism adds an ontological orientation to this, standing in direct opposition to all forms of moral anti-realism and moral skepticism. The opposition is stark. Ethical subjectivism denies that moral propositions refer to objective facts, suggesting instead that morality is a reflection of individual or cultural sentiment. Error theory goes further, denying that any moral propositions are true at all, implying that our moral language is a systematic error. Non-cognitivism rejects the premise entirely, denying that moral sentences express propositions, viewing them instead as emotional exclamations or commands.

The philosophical lineage of this doctrine is deep and formidable. Most philosophers claim that moral realism dates at least to Plato as a philosophical doctrine, and it remains a fully defensible form of moral argument in the contemporary landscape. The two main subdivisions of this vast territory are ethical naturalism and ethical non-naturalism. Naturalism argues that moral properties are identical to, or reducible to, natural properties, while non-naturalism posits that moral properties are sui generis, irreducible to the natural world. Despite these internal divisions, the consensus among realists is that moral facts are real.

The strength of this position is evident in the demographics of the field. A 2009 survey involving 3,226 respondents from mostly English-speaking universities, including faculty members, PhDs, and graduate students, found that 56% accept or lean toward moral realism. The opposition accounted for 28% (anti-realism) and 16% (other). By 2020, the tide had shifted slightly, with the number of realists rising to 62.1%. This growing consensus is not limited to abstract theorists; it includes some of the most notable figures in modern philosophy. David Brink, John McDowell, Peter Railton, Geoffrey Sayre-McCord, Michael Smith, Terence Cuneo, and Russ Shafer-Landau are robust defenders of the view. The list extends to G. E. Moore, John Finnis, Richard Boyd, Nicholas Sturgeon, Thomas Nagel, Derek Parfit, and Peter Singer. Even the interpretation of historical figures has yielded realists; Norman Geras has argued that Karl Marx was, in fact, a moral realist, suggesting that his critique of capitalism was not just a materialist analysis but a judgment on objective moral wrongs.

To understand the architecture of moral realism, one must look at the specific theses that define its robust form. The literature has delineated moral realism into a minimal form, a moderate form, and a robust form. The robust model commits moral realists to three specific theses. First, the semantic thesis: The primary semantic role of moral predicates, such as "right" and "wrong," is to refer to moral properties like rightness and wrongness. Therefore, moral statements such as "honesty is good" and "slavery is unjust" purport to represent moral facts and express propositions that are true or false, or approximately true, largely false, and so on. Second, the alethic thesis: Some moral propositions are in fact true. Third, the metaphysical thesis: Moral propositions are true when actions and other objects of moral assessment have the relevant moral properties. These facts and properties are robust; their metaphysical status, whatever it is, is not relevantly different from that of ordinary non-moral facts and properties.

The minimal model, by contrast, leaves off the metaphysical thesis, treating it as a matter of contention among moral realists rather than a dividing line between realists and anti-realists. This distinction is not merely academic hair-splitting. It is significant because acceptance or rejection of the metaphysical thesis is taken by those employing the robust model as the key difference between moral realism and moral anti-realism. The classification of certain logically possible, if eccentric, views turns on which model one accepts. For instance, a view that rejects the semantic and alethic theses but accepts the metaphysical thesis might be called "realist non-cognitivism" by a proponent of the robust model, while a proponent of the minimal model might place such a view alongside other, more traditional, forms of non-cognitivism.

The disagreement extends to the classification of moral subjectivism. In the robust model, subjectivism—the view that moral facts are not mind-independent but that moral statements may still be true—is often treated as a form of anti-realism. The historical association of subjectivism with moral anti-realism explains why the robust model has been dominant, even if only implicitly, in both traditional and contemporary philosophical literature. However, the minimal sense of realism opens the door for figures like R. M. Hare. In his later works, Hare could be considered a realist in the minimalist sense because he is committed to the objectivity of value judgments, even though he denies that moral statements express propositions with truth-values per se. Similarly, moral constructivists like John Rawls and Christine Korsgaard may be realists in this minimalist sense; Korsgaard describes her own position as "procedural realism."

The implications of these philosophical stances extend into the realm of science and evolution. Some readings of evolutionary science, such as those by Charles Darwin and James Mark Baldwin, suggest that if an ethics is associated with survival strategies and natural selection, such behavior may be associated with a moderate position of moral realism. This view equates to an ethics of survival, suggesting that moral facts might be grounded in the biological imperatives of the species. This bridges the gap between the natural world and the normative world, offering a potential pathway for a naturalistic moral realism.

It is crucial to distinguish moral realism from moral objectivism, though they are closely related. Moral objectivism is the view that what is right or wrong does not depend on what anyone thinks is right or wrong. Moral objectivism allows for moral codes to be compared to each other through a set of universal facts. Nicholas Rescher argues that moral codes cannot derive from one's personal moral compass. A prime example is Immanuel Kant's categorical imperative: "Act only according to that maxim [i.e., rule] whereby you can at the same time will that it become a universal law." John Stuart Mill proposed utilitarianism, which asserts that in any situation, the right thing to do is whatever is likely to produce the most happiness overall.

According to the ethical objectivist, the truth or falsehood of typical moral judgments does not depend upon any person's or group of persons' beliefs or feelings. This view holds that moral propositions are analogous to propositions about chemistry, biology, or history, inasmuch as they are true despite what anyone believes, hopes, wishes, or feels. When they fail to describe this mind-independent moral reality, they are false—no matter what anyone believes, hopes, wishes, or feels. There are many versions of ethical objectivism, including various religious views of morality, Platonistic intuitionism, Kantianism, utilitarianism, and certain forms of ethical egoism and contractualism. However, Platonists define ethical objectivism even more narrowly, requiring the existence of intrinsic value. Consequently, they reject the idea that contractualists or egoists could be ethical objectivists. Objectivism places primacy on the origin of the frame of reference and considers any arbitrary frame of reference a form of ethical subjectivism by a transitive property, even when the frame incidentally coincides with reality.

The practical utility of moral realism becomes apparent when considering the mechanics of logic and disagreement. Moral realism allows the ordinary rules of logic, such as modus ponens, to be applied straightforwardly to moral statements. We can say that a moral belief is false, unjustified, or contradictory in the same way we would about a factual belief. This is a significant advantage over theories like expressivism, which struggle with the Frege–Geach problem. The Frege–Geach problem highlights the difficulty non-cognitivist theories face in explaining how moral statements function within complex logical arguments, such as conditionals. If "murder is wrong" is merely an expression of emotion, how can it be the antecedent of a conditional statement like "If murder is wrong, then getting your little brother to murder is wrong"?

Another advantage of moral realism is its capacity to resolve moral disagreements. If two moral beliefs contradict one another, realism asserts that they cannot both be right. Therefore, everyone involved ought to be seeking out the right answer to resolve the disagreement. This creates a framework for moral progress and rational debate. Contrary theories of meta-ethics have trouble even formulating the statement "this moral belief is wrong," and so they cannot resolve disagreements in this way. If morality is subjective, a disagreement is merely a clash of preferences, like one person preferring vanilla and another chocolate. But if morality is realist, a disagreement is a clash of truth claims, where one party is objectively mistaken and the goal is to discover the truth.

Peter Railton's moral realism is often associated with a naturalist approach. He argues that moral facts can be reduced to non-moral facts and that our moral claims aim to describe an objective reality. In his well-known paper "Moral Realism" (1986), Railton advocates for a form of moral realism that is naturalistic and scientifically accessible. He suggests that moral facts can be understood in terms of the naturalistic concept of an individual's good. He employs a hypothetical observer's standpoint to explain moral judgments. This approach attempts to ground the normative in the descriptive without falling into the naturalistic fallacy, by positing that the "good" is what a fully informed, rational agent would desire.

The relevance of these abstract distinctions to the concrete problem of AI alignment cannot be overstated. If we accept the anti-realist view that moral values are merely human projections, then an AI that does not share our specific cultural or biological projections has no "error" to correct. It is simply operating on a different set of arbitrary preferences. The problem of alignment becomes intractable because there is no external standard against which to measure the AI's behavior. However, if moral realism is true, then there is a correct moral state of affairs that the AI should be aligned with, regardless of whether the AI was trained on human data or evolved in a vacuum. The AI's task is not to mimic human whim, but to approximate objective moral truth.

This shift in perspective transforms the nature of the challenge. It moves the conversation from "how do we program the AI to do what we want?" to "how do we ensure the AI understands what is right?" It implies that there are moral facts that the AI, in principle, could discover and adhere to, just as it discovers facts about physics or mathematics. The robust model of realism, with its commitment to the metaphysical independence of moral facts, provides the strongest foundation for this view. It suggests that the moral landscape is as real as the physical landscape, and that navigating it requires the same rigor, logic, and commitment to truth.

The debate is not settled. The minimal model and the robust model continue to clash over the classification of various ethical systems. The question of how to categorize views that accept metaphysical theses but deny semantic ones remains a point of contention. Yet, the sheer number of philosophers who lean toward realism, and the logical coherence it offers for resolving moral disagreement, suggests that the position is far from a relic of the past. It is a living, breathing framework that continues to shape how we think about value, truth, and the nature of reality itself.

For those concerned with the future of intelligence, the stakes are high. If moral realism is false, then the alignment problem is a philosophical dead end, a quest to program a machine with a ghost. But if moral realism is true, then the project is one of the most important intellectual endeavors of our time. It is the search for the objective moral facts that will guide the actions of entities far more powerful than ourselves. The work of thinkers like Railton, Parfit, and Singer provides the tools to navigate this search, grounding our hopes for a safe future in the hard bedrock of objective truth. The distinction between a world where morality is a matter of opinion and a world where it is a matter of fact is the distinction between a future of arbitrary power and a future of reasoned justice.

The survey data from 2020 showing a 62.1% acceptance rate is more than a poll; it is a signal. It indicates that the intellectual community, the very people tasked with building the frameworks for our future, overwhelmingly believes that truth in ethics is possible. This belief is the prerequisite for any serious attempt to align advanced AI. Without it, we are merely shouting into the void, hoping our echoes sound like commands. With it, we are charting a course toward a reality where the right thing to do is not just what we say it is, but what it actually is. The philosophical work of the last two millennia, from Plato to the present day, has built a case for this reality. The task now is to ensure that our creations can see it, too.

The robustness of the realist position lies in its ability to withstand the complexities of human disagreement and the challenges of scientific reductionism. It does not deny the difficulty of moral judgment; rather, it explains why that difficulty exists. We struggle because we are trying to perceive a reality that is not immediately obvious, just as we struggle to see atoms or understand the curvature of spacetime. The fact that we struggle does not mean the atoms or the spacetime do not exist. Similarly, the fact that we disagree on moral issues does not mean moral facts do not exist. It simply means that the moral landscape is complex and requires rigorous inquiry to navigate.

In the end, the choice between realism and anti-realism is a choice about the nature of the universe we inhabit. Do we live in a universe where value is a projection of our minds, a fleeting shadow of our biological and cultural history? Or do we live in a universe where value is woven into the fabric of existence, as real and objective as gravity? The majority of contemporary philosophers, the survey data suggests, believe the latter. They believe that when we say "this is wrong," we are pointing to something real. And if they are right, then the future of our species, and the intelligence we create, depends on our ability to see it clearly.