May 22, 2018
This is post one of a sequence of posts I am writing on moral anti-realism and moral reasoning.
To start off this sequence, I want to give a short description of moral realism; I’ll be arguing against moral realism in later posts, and I want to clearly explain what it is I’m arguing against.
When I’m arguing against moral realism, I will deliberately set aside some moral realist views and focus on those forms of moral realism that I find most relevant – in the sense that the “relevant” versions, if correct, would be the most relevant to effective altruism and to people’s lives in general. I will call these versions of moral realism strong moral realism. Thus, I don’t claim that all versions of moral realism discussed in the academic literature are mistaken.
The goal of this introductory post is threefold:
Moral realism has been defined in different ways by different authors. I will start by discussing two different definitions, both of which are broad in that they allow for many different positions to count as ‘moral realism.’
The first definition is from Geoffrey Sayre-McCord’s Essays on Moral Realism (1988, p. 5). It is meant to serve as a definition for realism, not just about morality, but about any domain of claims under scrutiny.
Wherever it is found, [...] realism involves embracing just two theses: (i) the claims in question, when literally construed, are literally true or false (cognitivism), and (ii) some are literally true. Nothing more.
Sayre-McCord’s definition illustrates a confusing feature of the debate between moral realists and moral anti-realists: the discussion can happen simultaneously on two different levels. At the first level (I will call it the linguistic level), people disagree about the nature of first-order moral claims – about what competent speakers mean when they say things such as “Murder is wrong.” On the second, substantive level, people disagree about whether some moral claims (properly interpreted) are true, or whether all first-order moral claims are false. (Second-order moral claims, such as “all first-order moral claims are false,” can still be true even if all first-order moral claims are false.)
We can now see that different views regarding the linguistic level lead to different types of moral realism. Consider two moral realists. One thinks that moral claims such as “X is wrong” just mean the same thing as, e.g., “X reduces net happiness,” and she thinks that some of these claims are true. Another thinks that moral claims such as “X is wrong” refer to an irreducible property of wrongness, and he thinks that some of these claims are true. While these are both forms of moral realism, they are quite different.
The semantic definition allows for the situation that whether one endorses realism or not can depend solely on one’s views about language rather than one’s views about morality. Specifically, some versions of moral realism are grounded in idiosyncratic or minimalist accounts of what it means for a claim to be true (see pragmatism or minimalism about talk of truth). I won’t address these views further both because they are rarely explicitly defended and because any moral realism that is endorsed on merely semantic grounds is going to be inconsequential: whether we consider it true or not only has consequences for how we would speak (i.e., whether to call something moral realism or not), not for how we would act.
Some moral anti-realists hold that moral claims are not best interpreted as claims that can conceivably be true or false (that is, are not truth-apt). This position is called non-cognitivism (as opposed to cognitivism); it is the most radical form of anti-realism.
One non-cognitivist view is expressivism, which holds that moral claims are best interpreted as expressions of an evaluative attitude (veiled expressions of, e.g., the speaker’s approval or disapproval). According to the expressivist interpretation, when someone says “Murder is wrong,” the best interpretation of that statement is not something that can literally be true or false. Rather, the statement employs similar language to that used in truth-apt statements to express disapproval. “Murder is wrong,” that is to say, is just a non-cognitive expression of disapproval of murder; it can’t be true or false any more than it can be true or false to say “Ouch!” when you hit yourself with a hammer. “Murder is wrong” looks like a claim, but the appearance is deceptive.
Non-cognitivists have a great deal of explaining to do. They need to account for why moral discourse has the appearance of asserting truths, not just when we make emotionally loaded statements like “This is just wrong!”, but also in the context of carefully reasoned philosophical discussions. We readily use logic to analyze moral claims, or we talk about moral ‘beliefs’ and moral ‘knowledge’, and we have devoted an entire branch of philosophy (normative and applied ethics) to figuring out which moral claims are true.
Furthermore, it seems strange to deny that at least some people’s moral statements are truth-apt. After all, some people are ardent moral realists who make moral claims themselves, and at least some of them will tell us explicitly that they give zero credence to the non-cognitivist interpretation of their moral claims. Whether their intended usage of moral claims is the same as typical usage or not feels like a secondary question to me. Indeed, I find it surprising that such a substantial portion of metaethics is centered around conceptual analysis on the linguistic level, debating what people might mean when they say things like “X is wrong.” Given that moral realists exist and given that they believe that their interpretations of moral discourse are intelligible and important, it seems like we should be able to address their claims on their merits (or lack thereof), regardless of what else moral discourse may sometimes be about.
This makes debates about what moral claims mean in everyday language less relevant to my current project. An analogy: Suppose that I’m discussing theology with some philosophers. The philosophers are trying to interpret the Bible, making claims like, “It’s not true within the Biblical storyline that Noah was a woman but true that he was a man,” or, “Religious claims aren’t truth-apt because the authors of religious texts were telling parables to express thoughts on the human condition, they didn’t intend to say things that can be true or false in a literal sense.” And all I’m thinking is, “Why are we so focused on interpreting religious claims? Isn’t the major question here whether there are things such as God, or life after death!?” The question that is of utmost relevance to our lives is whether religion’s metaphysical claims, interpreted in a straightforward and realist fashion, are true or not. An analysis of other claims can come later.
This is also how I look at the literature on moral realism: Some versions of moral realism would be rather inconsequential if true, while others would be vastly relevant to our lives. Whether the latter stem from a typical interpretation of moral discourse or not is not that important (although if no one took strong realism about morality seriously, that may indicate obvious flaws in the moral realism hypothesis). So given that some versions of moral realism would be vastly more relevant to people’s lives than others, I want to primarily focus on assessing whether strong forms of moral realism are true. Correspondingly, this means I won’t be satisfied with non-cognitivist accounts that brush aside the possibility that strong moral realism is intelligible and evaluable on its own merits, the meaning of ordinary moral discourse aside. (See Kahane, 2013, who argues that moral realism can be intelligible and defensible even if it doesn’t reflect ordinary moral discourse).
For the purposes of the present essay, I’ll provide a definition of moral realism that works independently of the debate about the proper linguistic interpretation of moral claims. Stephen Finlay’s (2007) definition of what he calls ‘ontological moral realism’ is just such a definition.
[Ontological moral realism is the claim that] moral claims describe and are made true by some moral facts involving moral entities (e.g., reasons, obligations), relations (e.g., justification), or properties (e.g., goodness, rightness, virtue). [...] This form of realism takes as its objects the truth-makers of moral claims, holding that they include moral properties such as value (e.g., the goodness of charity) and moral entities such as practical reasons and obligations (e.g., reasons not to tell lies, obligations to keep promises).
Ontological moral realism is correct if moral claims are sometimes true in virtue of correctly referring to a ‘moral reality’ consisting of the “truth-makers” for moral claims, the entities that make those claims true: moral entities, relations, properties, etc. These entities, further, must exist independently of anyone’s beliefs. What this moral reality would consist of is explicitly left open. Ontological realism is therefore compatible with views according to which moral facts, such as facts about value or disvalue, can be identified with natural facts (e.g. facts about pleasure or suffering), and with non-naturalist views where the moral ‘reality’ consists of something more abstract (e.g. facts about reasons for action postulated by reasons externalism). (If the difference between naturalism and non-naturalism seems too abstract now, I’ll address it further later.)
We can further distinguish three, increasingly ‘objective’, conceptions of moral reality. All versions of ontological moral realism have in common that they are objective in the sense of being about the existence of moral facts: Facts do not change depending on who we ask or how we look at something. Then there is a second sense of ‘objective’ that refers to whether those facts are about something that (in part) depends on one’s personal desires/goals/preferences, or whether the facts remain the same even if one’s own desires/goals/preferences are changed.
Subjectivism is the view that moral claims are made true or false with respect to facts about one’s own (i.e., subjective) desires/preferences/goals about the world.
Intersubjectivism is the view that moral claims are made true or false with respect to facts about both one’s own desires/goals/preferences and those of other people. For instance, according to some intersubjectivist views, morality is about rational actors pursuing their own ends while respecting an envisioned social contract.
Objectivism (not to be confused with Ayn Rand’s Objectivism, which describes a subjectivist morality based on self-interest) is the view that morality is the same for everyone and independent from one’s personal desires/preferences/goals. (Or that one’s own desires at most count as “one out of thousands” e.g. in preference utilitarianism.)
Arguably, it is only objectivism that captures the ways in which (at least some people’s) moral intuitions make morality out to be something all-encompassing that every person is bound to.
Nevertheless, there are subjectivist and intersubjectivist views that would be relevant to our lives if they were true. In this section, I will give two examples of views, one subjectivist and one intersubjectivist, that seem to me like correct and practically relevant ways of thinking about issues in moral philosophy, even though they would plausibly count as “not realist” according to at least the semantic definition of moral realism.
Subjectivism holds that moral value is determined by an agent’s personal desires. According to the subjectivist Michael Smith, for instance, what is good for an agent is what the agent would desire if they had perfect information and were perfectly rational. I’m including the following quote from Finlay (2007) because the position sounds like some of the views that have been discussed prominently on LessWrong:
[...] Smith bases each person’s normative requirements on his or her own desires, subject only to rational enhancement (full information and coherence). Moral claims can be true, he maintains, provided that all rational persons would converge on a common set of desires with a distinctly moral content (Moral Problem 173, 187–9). Richard Joyce, who largely accepts Smith’s subjectivist approach as an account of normativity, reasonably objects that this claim on behalf of morality is implausible. Rational selves’ desires are reached by correction from actual selves’ desires, and these starting points are too diverse to support the required kind of convergence (89–94).
As a claim about about how moral discourse is to be interpreted, subjectivism holds that “X is good” should be treated as shorthand for “X is good according to my desires” (Sayre-McCord, 1988, p. 18). This seems like a somewhat implausible interpretation of moral discourse to me – but people’s intuitions about what morality is about may differ. Under the assumption that moral discourse is usually objectivist, while objectivist moral realism is false, perhaps one could resort to subjectivist moral discourse as a way to salvage something useful from the debris. In that case, subjectivism would rather count as (a constructive proposal within) anti-realism rather than a version of realism.
In any case, I have a lot of thoughts on the merits of subjectivist accounts that specifically refer to what we would come to value after moral reflection under ideal conditions, but will reserve them for later parts of this sequence where I explore options within what I think of as moral anti-realism. So for our purposes here with respect to my upcoming posts on why I’m not a moral realist, I’ll treat subjectivism as one version of anti-realism. Not because this is the obvious way of categorizing it, but because I want to reserve the moral realism label for only the most consequential versions of moral realism.
Another position located somewhere near the boundary between realist and anti-realist views is constructivism as a metaethical view, which holds that morality is about what rational actors would hypothetically agree to (under certain idealized conditions) with respect to how everyone, themselves included, should act. Different versions of constructivism give different accounts of how to think about this hypothetical agreement: Some are based on considerations about social contracts, others on Kantian universalization of one’s decision maxims (“acting as though one expects all other rational people to choose the same decision procedure”). A specification of the conditions under which hypothetical agreement is to be derived is called a constructive function (as explained by Shafer-Landau, 2003, pos. 201).
Metaethical constructivism is an intersubjectivist position. It is not objectivist because constructive functions merely constrain what follows from people’s desires, preferences, or goals – they do not introduce anything that we ought to do (or is morally good or bad) that goes beyond those desires.
While I am unconvinced by constructivism as a metaethical position (because that would commit us to the claim that moral discourse is necessarily all about hypothetical contracts rather than also e.g. unconditional altruism or care), I am sympathetic to constructivism being important on pragmatic or prudential grounds. Is there a single, uniquely compelling way to choose a constructive function? I think that the answer is not obviously no. I find it noteworthy that central aspects of (mostly Kantian) constructivism are mirrored in LessWrong-inspired discussions about the implications of non-causal decision theories. Perhaps these considerations could be thought of as plausible extensions of the concept “rationality as systematized winning,” such that, by getting their implications right, one could increase one’s all-things-considered degree of goal achievement. In any case, whether we want to call this moral realism or not, it is worth flagging constructivism as a moral view according to which morality is potentially action-relevant in a surprisingly non-trivial and yet rationally binding way.
One reason to perhaps not think of constructivism as moral realism is precisely because it seems to be more of an extension of what it means to be rational rather than what it means to be moral. At least according to some connotations of the word ‘moral,’ it is tied not only to notions of fairness or rational cooperation, but also to considerations of care or altruism. And while Kantianism or non-causal decision theories may imply that one should care substantially about the desires of other rational agents (in a perhaps power-weighted fashion), they do not imply anything about the content of one’s own desires, including whether to care about the well-being of sentient beings that are not (or insufficiently) rational.
If someone talks about moral realism without further elaboration, this person is probably talking about what I here call objectivist moral realism: The view that there are speaker-independent moral facts (ontological moral realism) that hold for each person independently of their personal desires/preferences/goals (objectivism). I also think that objectivism makes for the most straightforward linguistic interpretation of moral claims (although as I argued above, this should in itself not be our main criterion for selecting which positions to pay most attention to).
Error theory is the moral anti-realist counterpart to objectivist moral realism. (In theory it seems conceivable to me that one can be an error theorist believing that moral discourse is subjectivist or intersubjectivist in nature and is all false, but that would be unusual.) Unlike non-cognitivists, error theorists agree with realists that moral claims are best interpreted as saying things that can be true or false: as being truth-apt. Nevertheless, they deny the realist claim that first-order moral claims can be true. That is to say, error theorists hold that all first-order moral claims are, and must be, false.
There are broadly two types of objectivist realist positions: moral naturalism and moral non-naturalism. They differ with respect to what they take to be the nature of moral facts. In the same way it is true that a bachelor is an unmarried man, moral naturalists believe that that moral terms such as ‘morally good’ just refer to some natural property or natural properties, such as for instance pleasure or desire fulfillment. By contrast, moral non-naturalists think we will never be able to identify moral terms with natural properties, i.e., they believe that moral terms are basic and mean more than what can be expressed in non-moral language. Moral non-naturalists believe that, at best, we could only discover how non-naturalist moral facts map onto (or ‘supervene on’) natural facts. For instance, we could discover that situations where someone needlessly harms others always involve moral wrongness, but wrongness, thusly interpreted, is not synonymous with needlessly harming others.
How to distinguish between naturalist and non-naturalist positions is sometimes a subject of extensive debate. But perhaps the most salient difference between naturalism and non-naturalism is that the two positions tend to be susceptible to different types of challenges. In general, moral realism is backed by the ‘moral appearances’ (Finlay, 2007) – our realist intuitions about morality – and challenged by external pressures about how to reconcile realism about morality with what we know about the world and about the mind. Moral naturalism solves this challenge by making concessions (so one could argue) that weaken moral appearances, while moral non-naturalism stays maximally close to these moral appearances.
But staying so close to the moral appearances creates difficulties in reconciling non-naturalism with the rest of what we know about the world. These difficulties were famously summarized by John Mackie in his Argument from Queerness. Queerness is the charge that non-naturalist moral facts, should they exist, would be so different from all the other things we are used to in our conceptual repertoire that we had better think twice about incorporating them at all. The moral facts would be “queer” because, depending on the (usually non-naturalist) moral realist account in question, they may be causally redundant or impotent, be epistemically inaccessible, or have an (allegedly) mysterious connection to human motivation (see reasons externalism).
By contrast, moral naturalism (at least in most versions) evades these accommodation charges because naturalists believe that moral facts are simply natural facts, and that we could express moral claims in non-moral terminology without necessarily altering the meaning. However, moral naturalists are faced with a different conundrum, summarized in G.E. Moore’s Open Question Argument: Simply arguing that moral and non-moral terms are synonymous (e.g. that 'goodness' and 'desire satisfaction' or “happiness” are synonymous) is dubious. This is because it seems perfectly coherent to ask: 'Sure, this is an example of desire satisfaction, but is it good?' (Or: 'Sure, this is an example of happiness, but is it good?' Etc.) For illustration, note that one cannot ask these questions coherently about pairs of terms that are truly synonymous. For example, one cannot say: 'Sure, John is an unmarried man, but is he really a bachelor?' This question is not coherent because 'bachelor' and 'unmarried man' mean the same thing. Thus, moral naturalists have some explaining to do when they they hold that moral and non-moral terms are synonymous.
There are many different versions of moral naturalism and I will focus specifically on just two versions. We will notice that moral naturalist accounts often share strong similarities with subjectivism, even though they are objectivist views. This makes sense because a main challenge for naturalists is to show that they have singled out the right natural facts in their analysis of what is morally good or bad. And one promising way of establishing that one has singled out the right natural facts is by appealing to natural facts that are already of concern to each person individually.
The first example of an approach to moral naturalism is exemplified by Aristotelian virtue ethics as endorsed by Philippa Foot (1958) and Paul Bloomfield (as described in Finlay, 2007). Here we are less interested in this position itself, and more in the methodology or approach that motivates the position. Foot and Bloomfield both appeal to biology in determining which natural facts are moral facts; they note that certain things are conducive to the attainment of our biological ends – e.g. health, well-being, survival – and others aren’t; and conclude via conceptual analysis that ‘good’ (and other moral terms) refer to things conducive to the attainment of these ends.
For instance, in Moral Beliefs (1958), Philippa Foot argues against the idea that ‘good’ is a non-naturalist concept that solely expresses some kind of positive attitude. She points out that if we did use the term ‘good’ that way, nothing would prevent a “moral eccentric” to say that a good man is someone who randomly claps his hands simply because he (the eccentric) approves when people randomly clap their hands. Instead, drawing an analogy to the concept ‘injury,’ Foot advocates for a specific, substantive understanding of the term ‘good’ informed by conceptual analysis:
[It] may seem that the only way to make a necessary connexion between 'injury' and the things that are to be avoided, is to say that it is only used in an "action-guiding sense" when applied to something the speaker intends to avoid. But we should look carefully at the crucial move in that argument, and query the suggestion that someone might happen not to want anything for which he would need the use of hands or eyes. Hands and eyes, like ears and legs, play a part in so many operations that a man could only be said not to need them if he had no wants at all. That such people exist, in asylums, is not to the present purpose at all; the proper use of his limbs is something a man has reason to want if he wants anything. [...]
It will be noticed that this account of the action-guiding force of 'injury' links it with reasons for acting rather than with actually doing something. Just like our concept for ‘injury’ has both speaker-independent and "action-guiding" features, Foot argues that the same holds for the cardinal virtues prudence, temperance, and courage, and perhaps also of justice. Her theory of moral discourse is that it is discourse about virtues: character traits and dispositions that are beneficial for one’s natural ends. Interestingly enough, Foot notes that according to her view, justice only constitutes a virtue if it benefits the person who is being just. After all, she equates goodness with what is conducive to one’s natural ends, not with e.g. any notion of altruism or of benefiting others. Given that it is unclear whether justice is even a virtue on Foot’s account (as she herself points out), anyone with the intuition that moral discourse is also about considerations of justice – or simply someone who personally values justice – might then question whether Foot’s account really captures what is ‘good,’ and whether we should not rather want to seek justice for its own sake.
For contrast, we will now consider another naturalist position: that of Peter Railton as outlined in his paper Moral Realism. While Foot’s account based on conceptual analysis of what we mean by ‘good’ may fail to be convincing for people with different intuitions, Railton instead tries to establish the meaning of ‘good’ analytically: What is good for a person is what they would desire if they had full information about all the relevant facts and moral arguments. The advantage of this approach is obvious: Such a reduction of the term ‘good’ is personally relevant to us by definition. Railton makes two claims:
With regards to (1), Railton’s position resembles that of the subjectivist Michael Smith: What is good for us is the fulfilment of the desires we’d have if we were fully informed about our situation. What makes Railton’s position different from subjectivism is only that he further holds (2) that there is such a thing as objectivist morality, concerning what is non-morally good for everyone. He writes:
[M]oral resolutions are thought to be determined by criteria of choice that are non-indexical and in some sense comprehensive. This has led a number of philosophers to seek to capture the special character of moral evaluation by identifying a moral point of view that is impartial, but equally concerned with all those potentially affected.
Fixing the content of morality as something “impartial” is what allows Railton to have objectivist moral facts about what is good (for everyone) even though his conception of what is valuable for any given person only depends on their personal desires. Interestingly enough, Railton’s moral realism does not come with any rationally binding recommendations for how to act. His theory has an axiological component (axiology being the study of what is valuable), postulating objective value. But his theory has no deontic component (he does not believe in objective moral obligations).
Railton uses the slogan “rationality does go relative when it goes instrumental, but epistemology need not follow.“ In other words: While there is no direct reason to act morally for any one individual because rationality – procedurally interpreted – only concerns itself with drawing proper inferences from one’s pre-existing desires, there are nevertheless facts about what would make society good or bad from a perspective of maximizing desire fulfillment for all individuals. Whether to act morally is optional and up to one’s own desires, but reasoning about morality, on a purely epistemic level, can be done with an objective foundation.
This may feel like a disappointing conclusion from a paper titled Moral Realism. Railton himself notes that people might object that his view “may not make morality serious enough.” Having said that, personally I am happy to call positions that postulate only an axiology (and no moral obligations) moral realism – provided that there really is one uniquely correct, compelling axiology, as opposed to many different ways of determining what is “good for everyone” depending on different specifications of what constitutes (moral or non-moral) goodness.
The problem I have with Railton’s position specifically is that it is in several respects underdefined. For instance, it could turn out to be very difficult to formalize a uniquely compelling notion of “desires” or “desires given full information” that captures all our intuitions about when desire satisfaction is or is not valuable. Furthermore, Railton’s account cannot easily be extended to taking a stance about population ethics and does not specify a precise notion of what it means to take an “impartial perspective."Railton considers this to be an advantage:
By itself, the equation of moral rightness with rationality from a social point of view is not terribly restrictive, for, depending upon what one takes rationality to be, this equation could be made by a utilitarian, a Kantian, or even a non-cognitivist. That is as it should be, for if it is to capture what is distinctive about moral norms, it should be compatible with the broadest possible range of recognized moral theories.
However, not committing to any specific perspective calls into question whether there even is, in theory, a correct answer. If there are many different and roughly equally plausible interpretations of “impartial perspective” or “desire fulfillment” (or more generally: of well-being defined as “that which is good for a person”), then the question, “Which of these different accounts is correct?” may not have an answer.
What Railton shows is that there is at least one (vague) view about what to call “good for everyone” that is plausible or defensible. And we know there are some views about this that are obviously false. This is already a lot to show, as it counters a position of extreme moral skepticism saying that there is no sense at all in which we can reason objectively about morality.
Nevertheless, Railton’s position is not quite what I would be inclined to call strong moral realism. It leaves too much open for interpretation because we can focus on widely different criteria when trying to systematize what doing good for others comes down to. What I am interested in is whether there is more to it: Can we show that there is a view that is not only defensible, but uniquely correct? In the last section, I will describe what conditions a Railton-like view would have to meet for me to count it as strong moral realism.
Whatever we think of non-naturalist moral realism, it is certainly ambitious. Finlay calls it the “normative face of moral realism” because it is committed to the existence of irreducibly normative moral facts. What is attractive about non-naturalism is that the other versions of moral realism appear to be somewhat watered down by contrast. Non-naturalism is arguably best able to capture the urgency attached to the sentiment that some things really are right or wrong.
Support for moral non-naturalism has been growing lately. Finlay (2007) writes:
Although long considered an absurd Platonism, [non-naturalism] today enjoys a renaissance and boasts many and distinguished champions. [...] Besides Scanlon and Shafer-Landau, contemporary philosophers who defend non-naturalism (although not all under that label) include Thomas Nagel, Derek Parfit, Jonathan Dancy, Joseph Raz, Jean Hampton, Philip Stratton-Lake, Colin McGinn, Terence Cuneo, David Enoch, Michael Huemer, and William Fitzpatrick.
The non-naturalist position about moral facts is often inspired by Moore’s Open Question Argument. Shafer-Landau, for instance, believes that “Moore was correct in thinking that we could always intelligibly question the propriety of any candidate naturalistic reduction [of moral terms]” (Shafer-Landau, 2003, pos. 738). This leaves two options: Either we accept that there is no speaker-independent normativity, or we regard it as a separate realm not reducible to physical facts. Shafer-Landau and other non-naturalist philosophers have opted to go for the latter (although what this means exactly can differ from account to account, and sometimes the difference between non-naturalism and naturalism is subtle).
Next to the strong intuition that moral naturalism is inadequate to deal with the moral appearances, both Parfit and Shafer-Landau also defend moral realism by drawing analogies between morality and other domains about which (allegedly convincing) realist interpretations have been forwarded. For instance, Shafer-Landau points out “partners in crime” within the philosophy of logic/mathematics, and philosophy itself (Shafer-Landau, 2007, pos. 646):
[...] my kind of realism must seek out partners in crime. I would point to correct logical standards or physical laws (assuming a realist construal of such things), and claim that there isn't anything that makes such things true – they simply are true.
And also from the philosophy of mind (Shafer-Landau, 2007, pos. 949):
The sort of non-naturalism that I find appealing is one that bears a very close structural parallel to certain non-reductionist theories in the philosophy of mind. According to these latter views, mental properties are not identical to physical ones; mental facts are not physical facts; but mental properties are realized by instantiations of physical properties. At least in worlds relevantly close to ours, there would be no mental life without the physical stuff that constitutes it.
We can however ask whether these partners in crime really function analogously, and whether realist accounts of them are even correct. As Hallvard Lillehammer notes in his review of Shafer-Landau’s Moral Realism: A Defense, “Perhaps the most interesting thing about these alleged companions in guilt is that none of them are obviously innocent.”
Another question is how we could come to know anything about the normative realm, since it is separate from everything else (cf. the Benacerraf-Field problem in the philosophy of mathematics).
Finally, a third challenge is that, assuming for the moment that we grant the existence of non-naturalist moral facts, moral skeptics can challenge whether these facts are really action-relevant for them. In response, Shafer-Landau advocates moral rationalism, the view that “moral obligations are or entail reasons for action.” On this account, moral beliefs are on their own capable of motivating someone, but may not always be decisive for motivation.
I am most interested in accounts of moral realism which, should they prove to be correct, will be highly relevant to people’s lives and life projects: either directly because they are inherently compelling (they provide ‘real’ reasons to act), or because they are compelling at least for those people interested in pursuing goals motivated by altruism ("doing good for others"). With this in mind, I will now describe two different ways in which I could be convinced of strong moral realism.
Drawing from the above discussion, I would call myself a moral realist if I could be convinced that there is One Compelling Axiology in the form of a more developed version of Railton’s position. Such a view, as I envisage it, would combine a specific, complete theory about what is objectively in someone’s own interest, or is good or bad for them, with a specific, complete theory of what it means to do good for others from a kind of “impartial perspective.” (Which beings qualify as morally relevant “others” is also something the One Compelling Axiology would have to tell us; as is whether to only count people who exist currently or will exist regardless of our actions or whether to also intrinsically count the creation of new beings.)
As a loose (and untestable, at least not with current-day technology) criterion for what makes this form of realism true, I stipulate that I would count something as the One Compelling Axiology if all philosophers or philosophically-inclined reasoners, after having engaged in philosophical reflection under ideal conditions, would deem the search for the One Compelling Axiology to be a sufficiently precise, non-ambiguous undertaking for them to have made up their minds rather than “rejected the question,” and if these people would all come to largely the same conclusions. If the result was near-unanimous agreement about a highly specific view, I would count this as strong moral realism being true.
Note that this proposal makes no claims about the linguistic level: I’m not saying that ordinary moral discourse let’s us define morality as convergence in people’s moral views after philosophical reflection under ideal conditions. (This would be a circular definition.) Instead, I am focusing on the aspect that such convergence would be practically relevant: If maximally well-equipped and well-informed people would all come to the same conclusion of what it means to “do what is good for others” no matter the idiosyncrasies they started out with, then – assuming I find the prospect of doing what is good for others appealing – I have no reason to assume that my current thoughts on this are better than the current thoughts of someone who holds intuitions I find radically counterintuitive. This would be important to know!
I place no constraints on the possible outcomes of moral convergence, whether what is deemed good or bad for a person or a sentient being involves experiences, desire satisfaction, an objective list of things (e.g. friendship, love, exploration, etc.), or something we haven’t yet considered. The important point is that it needs to be a notion of well-being or “good for someone” that is widely compelling, not just as one defensible way of how to use the words “good for someone,” but as a compelling account of what is best for a person (or for a sentient being). A successful proposal has to give precise answers to questions such as which beings (or computations) matter morally (and how much?), or what the correct stance is on population ethics or aggregation in an infinite universe. For all these questions, the position would have to yield compelling arguments for why to take exactly one particular view as opposed to other plausible views. (This may sound overly demanding, but note that ideal conditions for philosophical reflection means having access to everything one can coherently ask for, including e.g. a well-intentioned, superintelligent oracle AI.)
The main challenge for strong moral realism in the form of a One Compelling Axiology is overcoming philosophical disagreement between sophisticated reasoners. There are several theories of well-being, and several notions of impartiality, that are internally coherent and highly intuitively appealing with respect to at least some people’s intuitions. These theories are mutually contradictory.
The other way I could become convinced of moral realism in the sense that I mean it is if, inspired by moral non-naturalism, I became convinced that irreducible normativity is a meaningful and somehow rationally binding (on some conception of rationality I currently find strange to envision). This would roughly correspond to moral non-naturalism being true. (Some people distinguish moral reasons from prudential reasons, whereas I tend to use the adjective 'moral' in a broader sense that relates to all one’s goals, both altruistic and non-altruistic, and generally to that which matters in one’s life. Since irreducible normativity also covers egoistic goals, it is broader than narrow-sense morality.) The challenges I see for a convincing account of irreducible normativity are threefold:
My next post will focus specifically on irreducible normativity, where I will explain in much more detail what I (and more importantly, others) understand under the concept of irreducible normativity.
 A note on terminology: Some people in my online network, particularly on LessWrong, seem to use the term ‘metaethics’ somewhat differently from standard usage. That is, they use ‘metaethics’ to refer to what I would call calling ‘normative ethics’ (or perhaps the best description would be “figuring out what humans value through philosophy and cognitive science”). Within academic philosophy, metaethics is the study of moral claims: what moral claims do or don’t assert and whether these assertions are sometimes true. The questions of whether e.g. utilitarianism is true, or whether human values are complex or not, are less likely to come up in a discussion about metaethics. Of course, metaethics is indirectly very relevant to all these questions and informs, for instance, whether inquiries into finding the ‘right’ human values or the ‘right’ version of consequentialism are well-posed questions or not. And it seems plausible to me that, according to some metaethical views, “figuring out what humans value through philosophy and cognitive science” is indeed how we should be doing normative ethics.↩
 According to pragmatism, a brand of philosophy that emphasizes the practical nature of ethics/life/everything and thereby – so one might argue – blurs the distinction between what is the case and what is practically useful, moral claims are ‘true’ not when they describe speaker-independent moral facts, rules or values, but when they result from “correct processes for solving practical problems” (Finlay, 2007).↩
 Quoting from the SEP Moral realism entry:
Yet, with the development of (what has come to be called) minimalism about talk of truth and fact, it might seem that this characterization makes being a moral realist easier than it should be. As minimalism would have it, saying that some claim is true is just a way of (re-)asserting the claim and carries no commitment beyond that expressed by the original claim. Thus, if one is willing to claim that “murdering innocent children for fun is wrong” one can comfortably claim as well that that “murdering innocent children for fun is wrong is true” without thereby taking on any additional metaphysical baggage.↩
 This criticism applies to many instances where philosophers do conceptual analysis. See also Luke Muehlhauser’s post on conceptual analysis and metaethics, or section 6 of this paper by David Chalmers. In short, the problems with using conceptual analysis to establish normative conclusions are threefold: Firstly, there may be no uniquely typical set of intuitions about the ‘correct’ usage of moral terminology. Secondly, ordinary usage may often be underspecified, because most people are not rigorously trained moral philosophers. Thirdly, even if the vast majority of people did use moral terminology a certain way, this would not necessarily mean that they would be using it the most useful or most ‘right’ way (provided the moral realist premise that there is a uniquely right way). As an antidote to approaches anchored in the tradition of conceptual analysis, Chalmers makes the following proposal (“X” refers to concepts such as ‘knowledge,’ ‘moral,’ and ‘science’ that are difficult to define):’
On the picture I favor, instead of asking “What is X”, one should focus on the roles one wants X to play, and see what can play that role. The roles in question here may in principle be properties of all sorts: so one focuses on the properties one wants X to have, and figures out what has those properties. But very frequently, they will be causal roles, normative roles, and especially explanatory roles.↩
 See also Kahane (2013) for the same point argued for at length.↩
 It is important to note that Finlay uses the adjective ‘ontological’ in a weak sense. Derek Parfit (2011) defended a metaethical view he called Non-Ontological Cognitivism: Both mathematical talk and moral talk can be objectively true, but there are no mathematical or moral entities. Note that, according to Finlay’s typology, this would still count as ontological moral realism because Finlay’s definition liberally counts externalist reasons as (abstract) ‘moral entities.’↩
 Constructivism as a metaethical view is different from constructivism as a position in normative ethics. Tim Scanlon (2012), for instance, is a constructivist as regards normative ethics; however, his metaethical position is objectivist moral non-naturalism. Scanlon believes, like Parfit and Shafer-Landau, that there are irreducibly normative reasons about what people ought to do. He further believes that constructivism as an approach to normative ethics helps us determine which reasons are correct. But the question to be answered in the end is which reasons are really correct, rather than which reasons are correctly the output of a well-specified constructive function.↩
 Especially the “kingdom of ends” formulation of Kant’s categorical imperative suggests this interpretation. What follows is my own translation from German (Kant, 1986). Note that in producing this translation, I had to make several substantial judgment calls.
The idea that every rational being is compelled to regard itself as an arbiter of universalizable norms, in order to evaluate itself and its actions from this perspective, leads us to a related and extraordinarily fruitful concept, namely that of a realm of ends. Under a realm of ends, I understand the systematic connection between rational beings through collectively shared norms. Because ends are determined by the universal validity of these norms, it follows that, if one abstracts from the personal differences between rational beings and from the content of their personal ends, then we can think up a systematically connected whole that encompasses all ends (including both the rational beings as ends in themselves and the ends that any rational being may set for itself), i.e., a realm of ends, which we can conceive according to the aforementioned principles.
To be clear, I am not saying that Kantianism is best interpreted as making claims that are related to current discussions of non-causal decision theories. I think there are several aspects of Kantianism that go against this interpretation. I am only saying that there are interesting parallels, and that, if one wants to, one could make a case for such an interpretation (or extension) of Kantianism.↩
 Peter Carruthers, for instance, has argued on contractualist grounds against animals' having rights (Carruthers, 1992). There are, however, some constructivists who endorse animals having rights, most notably Christine Korsgaard (2012).↩
 Alternatively, someone may have in mind an even more restrictive definition of moral realism – what is sometimes called ‘robust moral realism’ – that only refers to a subtype of objectivism: moral non-naturalism. This definition focuses on whether there are facts that are irreducibly normative. (See the subsection on moral non-naturalism, as well as posts 2 through 4 in this sequence.)↩
 Note that some non-naturalists, such as Shafer-Landau, think that moral properties are realizable by many different ‘constellations’ of natural properties. Moral pluralism as a normative view is arguably more attractive for non-naturalists than it is for naturalists because naturalists seem to be committed to a one-to-one relationship between goodness and some other natural property (Shafer-Landau, 2003, pos. 1215).↩
 As an analogy: Displaying a particular image on a computer screen is not synonymous with displaying a specific configuration of pixels on the computer screen, because the image – at least when viewed subjectively at a macroscopic level with human-level vision – is realizable via many different pixel configurations. The picture “supervenes” on the pixel configurations. The analogy is imperfect because an image really is nothing more than the sum of its pixels, and a picture that is slightly but percievablly different from another picture is just that, another picture. With non-naturalist morality, going from a constellation of physical facts that does not form a moral category to a constellation that does form a moral category must make for a sharp boundary somehow. However, because we cannot articulate, with reference to physical facts alone, what this sharp boundary signifies, it seems strange or “queer” to think that such a boundary even exists in a meaningful and action-relevant sense.↩
 The SEP entry on moral non-naturalism reads: “There may be as much philosophical controversy about how to distinguish naturalism from non-naturalism as there is about which view is correct. [...] Perhaps the most vexing problem for any general characterization of non-naturalism is the bewildering array of ways in which the distinction between natural and non-natural properties has been drawn.”↩
 An exception is Cornell Realism, a version of naturalism that holds that moral facts, although natural, cannot be reduced to other natural facts. Cornell realists claim that their position avoids the Open Question Argument that threatens other versions of naturalism. (See also footnote 16.)↩
 One might think that the Open Question Argument leaves open the option of moral facts and natural facts merely being coextensional, i.e., that they refer to the same thing but via different routes. However, some philosophers (and myself) believe that if two concepts are coextensional in every possible world, that just means that they are synonymous.↩
 Many moral realists reject the procedural account of rationality (which corresponds to the way ‘rationality’ is used on LessWrong: cognitive skills that help to achieve whatever goals one already endorses) in favor of what they call substantive rationality. On the substantive account, being rational may for instance entail having the right dispositions to apprehend or be motivated by externalist reasons for action. The belief in substantive rationality therefore tends to go together with reasons externalism. (And one may argue that this is circular: it defines substantive rationality with respect to externalist reasons, and externalist reasons with respect to substantive rationality.)↩
 As a third option, we may accept that normativity is nothing over and above the physical, but that it cannot be defined in terms of the physical easily, any more than a stock market can be defined in terms of the physical. This position describes non-reductionist moral naturalism, a view that is usually associated with Cornell Realism (cf. footnote 13), but not limited to it (see, Sinhababu, 2018). However, just like I think there is nothing of relevance that depends on whether we call ourselves “realists” about the stock market or not, I fail to see how such a position would be relevant to our lives if it were true. (I suppose it would be relevant insofar as it may come with metaphysical baggage of rejecting reductionism in general, which might change how we approach philosophical questions.)↩
 It need not be exactly like Railton’s position. Railton endorses some notion of desire fulfillment as what is good for a person. Another version of moral realism, one that I would consider to be “strong moral realism” in the One Compelling Axiology sense, is moral realism based on the idea that experiences can be intrinsically morally valuable or disvaluable. Proponents of such views believe that phenomenological introspection can tell us about pleasure’s (moral) goodness or pain’s (moral) badness (see e.g.: Hewitt, 2008 & Sinhababu, 2010). I will discuss moral realism based on phenomenological introspection in my seventh post in this sequence.↩
 By “ideal conditions”, I am envisioning a scenario that is perfectly suited for making progress on questions of philosophy. Imagine a setup that covers everyone’s needs and also provides access to all of the following:
Furthermore, there would be some mechanism in place to gently break up epistemically unhealthy group dynamics (for example, if charismatic people’s influence on others’ opinions was disproportionate). Alternatively, the journey could also be undertaken in solitude. In general, we could imagine a mechanism in place to prevent anything that radically alters the intuitions and goals of our would-be philosophers in ways that are not intended. Needless to say, there is no uniquely correct notion of “ideal conditions for philosophical reflection,” and if different plausible setups lead to radically different results, that would just be an additional way in which moral realism via One Compelling Axiology could fail.↩
 Someone may object that it doesn’t matter to them what the vast majority of people would conclude after philosophical reflection, because they have their own intuitions about what it means to do good for others, and because moral convergence is not necessarily the same as moral truth. I think this is a legitimate argument in a situation where people are pursuing different questions: If some people associate morality with words like ‘excitement,’ whereas other people associate it more with ‘seriousness,’ maybe that just means they are envisioning different things and are answering different questions when trying to systematize their moral intuitions. However, in the One Compelling Axiology scenario, I stipulate that there is agreement about what the question is, and that people who are relevantly similar to oneself with respect to how they approach existential questions also convergence on the same answer as everyone else. In that case, it would be weird to consider this fact irrelevant to one’s personal thinking about what it means to do good for others.↩
 Note that just because there may not be a One Compelling Axiology does not mean that we should not expect ideal conditions for philosophical reflection to be useful, or that we should think that there is no difference between obviously silly views about what matters and views that are plausible or defensible. I think of the difference between well-done and poorly-done moral reasoning as a continuum with different peaks, representing different questions being asked. Rejecting One Compelling Axiology only means that we need to put in more legwork upfront in order to decide what types of questions we want to answer, but it does not mean that everything related to moral-philosophical practice is useless.↩
Many people helped me with this post, but I want to specifically thank Simon Knutsson for important advice on earlier drafts that greatly improved the direction I went for with this post, and Elijah Armstrong for on-point reading suggestions and for tirelessly suggesting improvements to the ways I explain things.
Chalmers, D. (2011). Verbal Disputes. Philosophical Review 120(4):515-566.
Finlay, S. (2007). Four Faces of Moral Realism, Philosophy Compass 2(6):820-849.
Foot, P. (1958). Moral Beliefs. Meeting of the Aristotelian Society.
Hewitt, S. (2008). Normative Qualia and Robust Moral Realism. PhD thesis. New York University.
Kahane, G. (2013). Must Metaethical Realism Make a Semantic Claim? Journal of Moral Philosophy, 10(2):148-178.
Kant, I. (1986(1785)). Grundlegung zur Metaphysik der Sitten. Stuttgart: Reclam.
Korsgaard, C. (2012). A Kantian Case for Animal Rights. In: Animal Law – Tier und Recht: Developments and Perspectives in the 21st Century, ed. M. Michel, D. Kühne & J. Hänni. Dike: Zürich.
Lillehammer, H. (2004). Moral Realism: A Defense, Notre Dame Philosophical Reviews. https://ndpr.nd.edu/news/moral-realism-a-defense/.
Parfit, D. (2011). On What Matters, Volume II. Oxford: Oxford University Press.
Railton, P. (1986). Moral Realism. The Philosophical Review, 95(2):163-207.
Sayre-McCord, G. (ed.) (1988). Essays on Moral Realism. Ithaca: Cornell University Press.
Scanlon, T. (2012). The Appeal and Limits of Constructivism. In Constructivism in Practical Philosophy, ed. J. Lenman & Y. Shemmer, 226-242. Oxford: Oxford University Press.
Shafer-Landau, R. (2003). Moral Realism: A Defense [Kindle version]. Oxford: Oxford University Press. Retrieved from Amazon.com.
Sinhababu, N. (2010). The Epistemic Argument From Hedonism. Unpublished. (Retrieved: May 2018).
Sinhababu, N. (2018). Ethical Reductionism. Journal of Ethics and Social Philosophy, 13(1):32-52.