Moral Anti-Realism Sequence #4: Why the Moral Realism Wager Fails

by Lukas_Gloor11 min read14th Jun 202012 comments


Meta-EthicsMoral UncertaintyMoral Philosophy

This is the fourth post in my sequence on moral anti-realism; it works well as a standalone piece.


In my previous post, I argued that irreducible normativity may not be meaningful. One might hold the intuition that if our actions don’t matter in the irreducibly normative sense, they don’t matter at all. In this post, I’ll address the argument that as long as we believe there is a slight chance that irreducible normativity might be true, we should act as though it’s true.

This wager for moral realism has the same structure as a related argument described by Michael Huemer (2013). I will discuss Huemer’s argument and show how we can expand it into a wager for moral realism. Then, I’ll explain why I consider the resulting wager unconvincing.

Huemer’s “proof of moral realism”

In the paper “An Ontological Proof of Moral Realism,” Michael Huemer presents the following argument in support of moral realism:

Given that moral realism might be true, and given that we know some of

the things we ought to do if it is true, we have a reason to do those things. Furthermore, this reason is itself an objective moral reason. Thus, we have at least one objective moral reason.

The conclusion in the first sentence (“we have a reason to do those things”) derives from what Huemer calls the Probabilistic Reasons Principle:

The rough idea is that if some fact would (if you knew it) provide a reason for you to behave in a certain way, then your having some reason to believe that fact obtains also provides you with a reason to behave in the same way. Even a small epistemic probability of the fact’s obtaining provides you with a (perhaps very small) first person reason for action.

So, the argument is that if we start with non-zero credence in the existence of moral reasons, and if we have at least a vague sense about what those reasons would imply, then, via the Probabilistic Reasons Principle, we can conclude that we have one type of irreducible reason—and Huemer argues that this would be a moral reason—with certainty. Namely, we would then, with certainty, have a moral reason (possibly a very weak one) to act as though those other moral reasons apply.

A quick note on terminology

Huemer distinguishes between moral reasons (which, according to his usage, are necessarily other-regarding reasons) and prudential reasons (which are reasons related to one’s self-interest). For this article, I will adopt Huemer’s distinction. That said, I reject prudential reasons and moral reasons based on the same arguments: I cannot make sense of “reasons” (in the irreducibly normative, reasons externalist sense) as a concept.

Transforming Huemer’s argument into the moral realism wager

By framing his argument in the language of objective reasons for action, Huemer already takes some version of irreducible normativity for granted. His claimed contribution is merely about deriving a particular kind of irreducible reason (a moral one) from other irreducible reasons. If our initial credence in the existence of moral reasons was low, the newfound moral reason we obtain through Huemer’s argument would remain comparatively weak.

For example, suppose I believe in the existence of irreducible prudential reasons, and I think that those favor me buying a watch. And suppose I assign a 1% credence to the existence of moral reasons, of which I believe that they support giving the watch money to charity. Huemer’s argument now accomplishes the following:

Via the Probabilistic Reasons Principle, I can conclude that I hold at least one moral reason with certainty. Namely, I have reason to feel at least somewhat (reflected by my initial 1% credence) moved by the moral reason for donating to charity. Correspondingly, I should refine my initial (1%) credence in the existence of any moral reasons to the following, more nuanced set of credences:

  1. With certainty, I have at least one moral (meta-)reason: I should feel at least somewhat moved by my best guess about the content of the other, object-level moral reasons.
  2. I still maintain my original 1% credence in those other, object-level moral reasons (e.g., reasons in favor of making a charitable donation).

Note that I wrote “moral (meta-)reason” to highlight that we are dealing with an unusual type of moral reason. Huemer argues that deriving these “moral (meta-)reasons” from prudential reasons qualifies as proving moral realism. (I’m somewhat skeptical about this claim, but it has no bearing on why I ultimately reject the moral realism wager.)

Secondly, note that in the above example, the newfound moral reason for donating—despite being held with certainty—may not be strong enough to outweigh my prudential reasons for buying the watch. More generally, as long as our credence in object-level moral reasons was initially low, newly derived moral reasons won’t outweigh our strongest prudential reasons. This conclusion may seem underwhelming.

Huemer’s argument becomes much more interesting, however, if we start out disbelieving all irreducibly normative reasons, whether prudential or moral ones. Let’s say we start 99% convinced that irreducible normativity is altogether nonsensical. And let’s say we have a vague idea what irreducible normativity would imply for us, should it exist. (This assumption is questionable—I will come back to it in the next section.) Again, then we can conclude that we have at least one irreducibly normative (meta-)reason: We should act as though object-level irreducibly normative reasons apply.

Compared to the previous application of Huemer’s argument, this application here is importantly different. Since we are contrasting irreducibly normative reasons with reasons anti-realism, we cannot contrast our newfound (meta-)reason with potentially more strongly held reasons of the same, irreducibly normative kind. Instead, we have to compare the newfound (meta-)reason to the view that there are no irreducibly normative reasons at all.

In personal communication, several effective altruists have expressed the intuition that irreducibly normative reasons are always decisive. According to this intuition, what we do matters infinitely less, in some robust and all-things-considered sense, if there are no irreducibly normative reasons. If we accept this picture, then we have to either reject irreducible normativity with complete confidence or act as though it is true, even if our actual credence is low. This is the wager for moral realism based on irreducible normativity.

Side note: different interpretations of irreducible normativity

In my previous post “#3 Against Irreducible Normativity,” I outlined three ways to interpret the notion of irreducible normativity. They correspond to the following section headings in my previous post:

1. “Super-reasons

2. “Is (our knowledge of) irreducible normativity limited to self-evident principles?

3. “Is there a speaker-independent normative reality?

To follow the argument in this post here, there’s no need to read up on the above distinctions. I just want to briefly note that the moral realism wager is inconsistent with interpretations 2.[1] and 3.[2]

The intuition “our actions matter infinitely more if irreducible normativity applies” only applies to interpretation 1.—the interpretation of irreducible normativity that I find the most strange. I couldn’t make sense of this notion in my attempt to explain it. (Of course, since my inability to make sense of something doesn’t mean that the probability of it making sense is exactly zero, this is where the moral realism wager could come in.)

In the following section, I will provide counter-arguments against the moral realism wager. I wrote this side note to point out that those counter-arguments are only needed if we subscribe to a particular, already contested interpretation of irreducible normativity.


The moral realism wager seems questionable to me in several ways. I will start by describing two objections that I consider forceful, but not in themselves decisive. Then, in subsection 3., I’ll present what I think to be a decisive counter-argument.

1. Incoherence

To derive practical implications from the moral realism wager, we need an informed guess about the content irreducibly normative reasons would have, if they existed. As Brian Tomasik (2014) explains in this section of his essay “Why the Modesty Argument for Moral Realism Fails,” it’s controversial whether we can obtain this type of probabilistic knowledge. We may think that morality is more likely to be about donating to charity than buying luxury goods. However, if moral anti-realism is right, moral language cannot successfully refer to irreducibly normative facts. Moral anti-realists typically don’t think that moral realism is coherent but wrong. Instead, they find the concept inherently confused—like a square circle. Therefore, one might argue that all bets about the content of irreducibly normative reasons are off.

That said, as Tomasik also concedes in his essay, proponents of the moral realism wager can now resort to an argument from modesty. Many philosophers consider irreducible normativity to be meaningful. Those philosophers may advance informed guesses about the content of irreducible reasons. For instance, in the PhilPapers survey (Bourget & Chalmers, 2014), 56.4% of surveyed philosophers “accept[ed]” or “lean[ed toward” moral realism (note, however, that this also includes versions of moral realism not based on irreducible normativity). While we may consider irreducible normativity meaningless ourselves, perhaps this reflects a shortcoming on our part.

Via the argument from modesty, it seems plausible that even people who lean heavily toward moral anti-realism should—on peer-disagreement grounds—assign at least some probability to the hypothesis that irreducible normativity is a coherent concept. Still, I think there are a few reasons why it could be defensible to have this probability be very low. While moral realism is a common belief, the wager for it only applies to versions grounded in irreducible normativity. Most people whose reasoning I hold in high regard are skeptical of irreducible normativity. Moreover, my impression is that among this set of people, most people who place significant credence on irreducible normativity do so because of their epistemic modesty. To avoid double-counting, we should only update toward other people’s credences if they endorse irreducible normativity for direct reasons, i.e., if they confidently claim that they understand it. This seems rare.[3]

2. Infectiousness

The moral realism wager requires us to compare irreducibly normative reasons to the view that there are no objective reasons at all. MacAskill (2013) points out that this is a situation where intertheoretic comparisons are problematic. His argument has two parts. First, he argues that we should generally use an expected value framework for reasoning under intertheoretic uncertainty. Then, he highlights that moral anti-realism neither assigns value zero to all options nor assigns any well-specified but uniform value. Instead, according to moral anti-realism, actions simply don’t have any objective value. We can represent this as “according to anti-realism, the values of all possible actions are undefined.” Consequently, any non-zero credence in anti-realism “infects” the entire expected value framework for deciding under intertheoretic uncertainty, rendering the value of every conceivable action undefined.

The infectiousness problem illustrates just how incompatible irreducible normativity is with moral anti-realism. In the next section, I will expand on this point and show that there isn’t a way to find a suitable method of comparison.

Another quick note on terminology

MacAskill’s paper is titled “The Infectiousness of Nihilism.” By Nihilism, he and other philosophers refer to moral anti-realism, or, more generally, anti-realism about reasons. I don’t like this choice of terminology because it smuggles in question-begging connotations. Those are exactly the type of connotations I want to argue against in the next section.

3. Begging the question

I believe that the moral realism wager fails because it begs the question. The wager only works if we stipulate that our actions matter infinitely more if irreducible normativity is true. There is no theory-neutral way to compare reasons anti-realism with reasons realism. Moreover, from the perspective of what we care about, I would be surprised if many people confidently endorsed the view “irreducible normativity always dominates.”

The situation is that we are uncertain between two frameworks: realism and anti-realism about reasons. Both of those are frameworks about which goals to pursue. Our situation is vexed because without a theory about which goals to pursue, we obviously cannot figure out which goals to pursue. Once we pick a theory, we are no longer neutral.

Huemer claims reasons imported via the Probabilistic Reasons Principle are irreducibly normative reasons. This point seems question-begging,[4] but I am willing to grant it. Even then, it remains question-begging whether the newfound reasons—which would only count weakly compared to more typical irreducibly normative reasons—should matter infinitely more than the merely subjective reasons behind anti-realism.

As I have argued in previous posts (especially in this section), anti-realism in no way means forfeiting all aspirations. On the anti-realist account, failing to act based on what one cares about can be as catastrophic as it gets. Subjective reasons for action aren’t just random whims. They include everything we care about—except, admittedly, for sentiments we can only express with normative terminology.

Even if someone’s life goals are entirely focused on “doing what’s right” or “doing the most good,” this doesn’t mean that they have to buy into the moral realism wager. For the wager to go through,[5] a person needs to explicitly think of these expressions in the irreducibly normative sense and stake all their caring capacity into that particular interpretation.

In practice, most people who want to “do the most good” presumably have in mind specific connotations that they’d consider a necessary part of that phrase’s meaning. If they somehow became convinced that the phrase “do the most good” referred to collecting pebbles, they would no longer identify that way.

Metaethical fanaticism: adhering to irreducible normativity as one’s only goal

I have argued that the moral realism wager fails for everyone who isn’t already committed to the view that irreducible normativity trumps everything. I believe that this captures the vast majority of people.

Admittedly, the intuition that one’s actions are meaningless without irreducible normativity is widespread. Most people may interpret this intuition in a tentative way that allows for the possibility that it's misguided. However, if someone is deeply attached to this intuition, it could function as a terminal value. On this view, "acting according to irreducible normativity" would constitute a personal life goal, which would stay in place even from an anti-realist way of looking at it. If this is the case, the moral realism wager goes through.

I have talked to two people in the effective altruism movement who claimed to endorse this position. I’m talking about a type of (stated) endorsement that goes well beyond “this position is worth exploring” or “this position is maybe true.” At least from the way those two people described it to me, they were willing to bet their life’s impact on this position. I will use the term metaethical fanaticism to refer to this stance.

As the label suggests, I’d caution against it. Before placing all caring capacity into a shaky philosophical assumption which may not even be meaningful, we should think carefully about the potential implications. In my next post, I will illustrate how metaethical fanaticism commits those who endorse it to potentially absurd consequences.


I’m grateful for helpful comments by David Althaus, Elijah Armstrong, Max Daniel, Sofia Davis-Fogel, Stefan Torges, and Johannes Treutlein.

My work on this post was funded by the Center on Long-Term Risk.


Bourget, D. and D. Chalmers. (2014). What do philosophers believe? Philosophical Studies, 170(3):465–500.

Huemer, M. (2013). An Ontological Proof of Moral Realism. Social Philosophy and Policy, 30(1–2):259–79.

MacAskill, W. (2013). The Infectiousness of Nihilism. Ethics, 123(3):508–520.

Tomasik, B. (2014). Why the Modesty Argument for Moral Realism Fails. reducing-suffering(.)org. <reducing-suffering(.)org/why-the-modesty-argument-for-moral-realism-fails/>.

  1. On the notion where our knowledge of irreducible normativity is forever limited to self-evident principles, the moral realism wager would remain irrelevant in practice. By definition, people will recognize self-evident principles either way, whether they try to act in accordance with irreducible normativity, or whether they just do what they most feel like doing. ↩︎

  2. We’re trying to answer whether being an anti-realist who’s wrong about realism is comparably bad to the converse case—being a realist wrong about anti-realism. A categorical moral realism wager only applies if the former is infinitely worse than the latter.
    In this endnote, I’ll adopt an interpretation of irreducible normativity according to which normative terms successfully describe a speaker-independent reality of principles that feel compelling to us. To assess what this means for the moral realism wager, we have to ask ourselves what sort of mistake each side would be making.
    To use a concrete example, let’s imagine someone with a strongly held but somewhat idiosyncratic moral belief. For instance, the belief that making new happy people is similarly important as preventing suffering. Let’s contrast two possibilities: 1. moral realism applies and the true moral theory captures this person’s intuition. 2. moral realism is false, but the person’s strongly held moral belief doesn’t contradict any self-evident normative principles, and the person arrived at their belief with sensible premises and sound reasoning.
    In possibility 1., the arguments for the belief in question were universally convincing, whereas in case 2., they were “only” convincing for some people, including the person in our example. Now, when I find myself having some opinion, it doesn't matter if all other people have the same opinion. As long as the opinion is genuine and holds up in light of peer disagreement, it's still my opinion. Therefore, it it doesn’t seem to me that there is a fundamental difference between the two types of being wrong ("being wrong as a realist vs. being wrong as an anti-realist").
    Take the analogy of culinary taste. I love cilantro, and I remember having read that for some people, it tastes like soap. If I now learned that the soap thing was a myth, and that cilantro is actually universally very tasty, that discovery wouldn’t update the value I get out of cilantro. I already liked it before, but me liking it wouldn't feel more significant just because I now think there's no room for disagreement.
    The situation is interestingly different when it comes to only weakly held moral beliefs. (“Weakly held” in terms of weak confidence, and/or in terms of only being weakly attached to a particular moral intuition.) If I find myself with the weakly held moral belief X, and it turns out that this belief is actually a central part of the true morality, this should strengthen my attachment to X. (In the cilantro analogy, this would correspond to thinking that one likes cilantro only weakly, but then finding out that every person likes it strongly.) While we might be tempted to turn these considerations into a conditional wager for moral realism (“acting as though moral realism provided that one's object-level moral beliefs are only weakly held”), there’s a dominant alternative principle that anti-realists can adopt. Instead of assuming that realism is true, which would incur costs in the case of being wrong (e.g., opportunity costs from spending time searching for the true moral theory), anti-realists can value further moral reflection. While this would look similar to “acting as though moral realism is true,” it's the more flexible option. Someone who values moral reflection doesn't have to make up their mind on whether realism (of the particular flavor under discussion in this footnote) is true, because they'd have a high chance of finding out if it was. A reflection procedure for anti-realists need not be not open-ended, but it can have specified conditions to assess the information value of further reflection steps, on an ongoing basis. (I plan to write more on valuing reflection in a future post.) ↩︎

  3. Derek Parfit, whose arguments I have addressed in previous posts, is the main exception that comes to my mind. On the outside-view, we may note that Parfit built his reputation primarily in the fields of normative ethics, rationality, and personal identity. His later writings on metaethics have not had a similarly discipline-defining impact so far. Admittedly, these things may take time. And on the inside-view, it is unclear to me whether Parfit’s conception of moral realism even supports the moral realism wager. It seems plausible to me that Parfit’s view comes closest to the third interpretation of irreducible normativity I laid out in my previous post. See endnote 2 for my thoughts on its relation to the moral realism wager. ↩︎

  4. In Huemer’s defense, his point makes sense in a context where we already postulate some externalist reasons. ↩︎

  5. When I speak of “the wager going through,” I mean that even the smallest non-zero probability placed on irreducible normativity would imply that one should discount all the other metaethical possibilities. Of course, one could also make a more gradualist argument based on the view that our actions may matter somewhat more if irreducible normativity applies. I don’t intend to argue against this possibility. (Whether or not this is rings true to someone would depend on their degree of attachment to the view that normativity is irreducible.) ↩︎