Cosmological Fine-Tuning
A Critique
1. A Fine Argument
I will state the argument from cosmological fine-tuning for a designer in the following way, roughly following Robin Collins,1 Luke Barnes,2 and others. The basic idea, on which I will elaborate shortly, is that life-permitting conditions are so unlikely on atheism, and not so unlikely on theism, that the fact that there are life-permitting conditions constitutes strong evidence in favor of theism over atheism.
I will make a few simplifying assumptions. First, I will start by considering designer hypotheses rather than theistic hypotheses, since the former are more neutral and less loaded. Second, I will compare only single-universe models. More precisely, I will be only considering models according to which the relevant physical parameters do not vary by spatial or temporal location. Henceforth, for simplicity,3 I will refer to this as the hypothesis that there is precisely one universe, but keep the above clarification in mind. Third, I will grant the basic fine-tuning assumption, that of all the ways the universe might relevantly be, the vast majority of this space is comprised by life-prohibiting universes.4 I will now give the basic Bayesian argument as follows:
The probability that there would be life-permitting conditions given that there is exactly one universe and it wasn’t designed is negligible.
The probability that there would be life-permitting conditions given that there is exactly one universe and it was designed is not negligible.
The prior probability that there is exactly one universe and it was designed is not negligible relative to the prior probability that there is exactly one universe and it wasn’t designed.
There are life-permitting conditions.
Hence, the posterior probability that there is exactly one universe and it was designed is greater than the probability that there is exactly one universe and it wasn’t designed.
Letting “D(x)” be the hypothesis the ontological posits of x are designed, U be the hypothesis that there is exactly one universe, and L be the hypothesis that there are life-permitting conditions, we could then formalize this argument as follows:
The upshot is that the fact that there are life-permitting conditions seems to warrant assigning a higher probability to the hypothesis that the universe is designed. Proponents of this sort of reasoning may suppose a stronger constraint in place of (3), enabling them to infer that the posterior on the given design hypothesis is much greater than the posterior on the no-design hypothesis conditional on the same datum. This will come up later, but for now, I will be considering the given version with the more modest constraint on the priors.
2. Looking a Gift Stalking-Horse in the Mouth
I want to first consider a style of reply to fine-tuning arguments, which we might call “stalking-horse” style replies. The basic idea can be seen in some of Graham Oppy’s writing,5 and has also been presented by Alex Malpass.6 Here’s the basic idea: whatever design hypothesis a proponent of fine-tuning provides, we can construct a non-design hypothesis which is at least as plausible antecedently and predicts life-permitting conditions just as well. If this is right, then it follows that, for any given design hypothesis, there is at least one non-design hypothesis which is not less probable given that there are life-permitting conditions (henceforth, L).
We can provide a rubric for generating such stalking-horses. Any given single-universe design hypothesis D will, for sake of simplicity, predict L with probability c.7 We can then construct non-design hypotheses in the following way:
N: there is a single universe which is as described by D except that it wasn’t designed, and L obtains with probability c.
We then only need the assumption that any D isn’t antecedently more probable than its corresponding N, and we can infer that for any given D, there is an N which is at least as plausible conditional on L. We can then conclude that L does not probabilistically favor design to non-design. I will now consider four objections to this strategy.
2.1 Objection 1: N cannot stipulate a probability in this way
We might think that a hypothesis cannot simply stipulate that some result obtain with probability c. Opponents of the Fine-Tuning Argument (hereafter, FTA) may attempt to make an additional posit (e.g., some necessary origin, natural disposition, etc.) which, on the theory, produces L with probability c. This is perhaps an acceptable move, but not a necessary one: there is nothing illegitimate in having a theory which makes various probabilistic predictions. For sake of brevity, I will not expand on this issue further here.
2.2 Objection 2: N cannot be defined in terms of D in this way
There are some design hypotheses for which we cannot easily talk about the universe absent said designer. This may be because the designer, on the given hypothesis, plays an important role in the internal aspects of the universe, or (at least) partly comprises the universe, or something else along these lines. In this case, we can modify N to accommodate this, to include only descriptions of the universe from D that are stated in a designer-neutral way. There are various ways this could be done to similar effect, although I will not explore it further here.
2.3 Objection 3: There are some D such that there is no N of this sort which is at least as plausible, antecedently
There are a few things to say here, although I will be brief. First, I’d be interested to see alleged cases of this sort. Standard design hypotheses, including those forwarded in the context of FTAs (e.g., by Swinburne, Barnes, Collins, etc.), do not seem to suffice, and I think the given rubric for generating stalking-horses will work for any D.
Second, I do not suppose that there is some objective fact about which priors are correct. Someone might have an array of priors which blocks the move here. I do not suppose that I can show that they are wrong, but I do think that the approach here is at least reasonable, and I might try to motivate it further by appealing to common probabilistic commitments and relevant heuristics.
2.4 Objection 4: This approach does not show why the FTA is unsound
If the stalking-horse strategy discussed is correct, then it may suffice to show that the conclusion of the FTA is incorrect, and so the argument is unsound. However, it’s not clear where said argument goes wrong, and we might want an explanation of that in order to be satisfied that it fails. I will explore this in the next section.
3 Getting Out of Tune
In this section, I will develop an argument against the given FTA, which should make it clear why the FTA fails. For reasons that will become apparent later, I first want to talk about what I will call “bare design” (BD) and “bare non-design” (BND) hypotheses. To be clear, I do not want to suggest that proponents of FTAs have to argue only for BD. The relevance of BD to more substantive design hypothesis will become evident in due course. To be explicit, then, BD and BND are, respectively:
Bare design (BD): there is exactly one universe, and it was designed8
Bare non-design (BND): there is exactly one universe, and it wasn’t designed
If we want, we could state these hypotheses as U∧D(U) and U∧¬D(U), respectively, but for simplicity, I will proceed with BD and BND. Next, I will assume two probabilistic constraints.
3.1 First constraint
The first constraint is that the prior on the bare design hypothesis does not exceed the prior on the bare non-design hypothesis, that is,
In other words, prior to considering the evidence, bare design is not more likely than bare non-design. I will consider objections to this constraint in §4.
3.2 Second constraint
The second constraint is that the likelihood that there would be life-permitting conditions (L) given BD does not exceed the likelihood that there would be life-permitting conditions given BND, that is,
In other words, BD does not more strongly predict life-permitting conditions than does BND. The basic motivation behind this is fairly simple: BD tells us nothing about what sort of designer there is, or what sort of universe is more or less likely to be designed. It merely says that there is a single universe and it was designed. We can think about this a bit more rigorously.
For our purposes, the possibilities we are concerned with are possible worlds at which exactly one universe exists, and that universe has physical parameters of the sort assumed by proponents of the FTA.9 The details are not important to my purposes here, since for sake of analysis, I am willing to grant the FTA proponent’s assumptions.
Consider the support typically adduced for the claim that L is very unlikely on BND. The vast majority of the possibility space such that ¬D(U) is populated by life-prohibiting universes, and because BND tells us nothing about which are more or less likely to be actual, we infer that it is very unlikely that L on BND. Granting the assumptions about the possibility space and the claims concerning the the physics, this appears to be a fair conclusion.
I want to say that the same is true, however, on BD. The vast majority of the possibility space such that D(U) is populated by life-prohibiting universes because we have have the same range of possible universes that are consistent with design, and by assumption, most of those are life-prohibiting. Additionally, because BD tells us nothing about which are more or less likely to be actual, we can equally infer that it is very unlikely that L on BD. This supports the second constraint, that P(L|BD) ≤ P(L|BND).
The key thing to note here is that the posterior on BD does not exceed the posterior on BND after each has been updated on L. This follows immediately from the two constraints and Bayes’ theorem:
The move to the second line follows from the two constraints.
3.3 More substantive design hypotheses
To my knowledge, nobody has forwarded a version of the FTA merely in support of the bare design (BD) hypothesis, and many may grant everything I’ve said thus far about bare design. Luke Barnes advocates for an omnipotent god which acts for reasons.10 Richard Swinburne advocates for a designer which is the god of traditional theism, which is essentially eternal, omnipotent, perfectly free, and perfectly good.11 Robin Collins advocates for a god which is omnipotent, omniscient, eternal, perfectly free creator of the universe whose existence does not depend on anything outside itself.12
The idea is that these hypotheses, or others like them, at least very modestly predict L, such that the FTA argument gets off the ground. What’s common between all of these more substantive hypotheses is that they posit BD plus some additional assumption(s). We can symbolize any hypothesis of this sort as BD∧A, where any additional assumptions are included in A. With this in mind, we are now equipped to state the counter-argument.
3.4 The counter-argument
P(BD) ≤ P(BND) (first constraint)
P(L|BD) ≤ P(L|BND) (second constraint)
P(BD∧A|L) ≤ P(BD|L) (theorem of probability)
P(BD|L) ≤ P(BND|L) (from 1, 2, Bayes’ theorem)
P(BD∧A|L) ≤ P(BND|L) (from 3, 4, transitivity of ‘≤’)
L
P⁺(BD∧A) ≤ P⁺(BND) (from 5, 6)
The conclusion is that the posterior probability of any design hypothesis, not just BD, will not exceed the posterior probability of BND after updating on L. Accordingly, L itself provides no probabilistic reason to prefer any design hypothesis to bare non-design.
This is my point, in a nutshell. Note that the only substantive premises in my argument, apart from the fact that L, are my two constraints discussed in §3.1 and §3.2. In §4, I will consider a range of possible objections. I will conclude in §5.
4 The Fine Print
4.1 Objection 1: The first constraint is false
There are a few different reasons why someone might have this attitude. First, someone might just antecedently find design, or something which entails design, highly probable. I might say more regarding why I have the priors that I have, but I don’t suppose that disagreements on priors here invariably can be resolved. I have little else to say against a person with priors of that sort except to say that they are not mine, and if the argument fails to move an open-minded agnostic merely because they do not have a design-favoring prior, then the FTA seems rather weak.
Second, someone might provide additional evidence in support of a higher prior on design. While I do not share the view that other evidence favors design to non-design, this approach involves a basic mistake. For the purpose of the FTA, we are considering priors independent of any such evidence which might count for or against it - apart, of course, from L.
4.2 Objection 2: Bare design does more strongly predict L
Again, there are a few different reasons why someone might have this attitude. I will consider the three I find most plausible. First, someone might think that not all design hypotheses are created equal, and that relatively simpler design hypotheses will be antecedently more likely, and so more likely given bare design. If some such hypotheses at least modestly predict L, then we have reason to think that there is at least some bias toward L given bare design. Second, someone might think that is it is built in to bare design in someway that L would be at least moderately more likely. Third, someone might think that there is other data about designers given which bare design will more strongly predict L. I will consider these responses in turn.
On the first approach, for such a response to be successful, the given L-favoring design hypotheses have to be simpler in this way, and this sort of simplicity has to warrant a higher prior. Although I will not explore this to the depth it certainly deserves, I do not, in general, think that simplicity has any direct bearing on priors. The only sense of simplicity that I take to directly bear on priors has to do with how committal the hypothesis is, how tight a probability distribution it places across the background possibility space.13 Moreover, on many senses of simplicity, any given design hypothesis will be less simple than its non-design counterpart (see section 2 on stalking-horses). Accordingly, if the relative simplicity of these design hypotheses biases bare design toward L, then the relative simplicity of their non-design counterparts will at least equally bias bare non-design toward L. I will not draw out this point here, though, since I ultimately do not think that simplicity considerations, apart from the one described, bear on my critique.
On the second approach, we might think that (bare) design entails intent, or acting for reasons, or something else, and this would at least modestly bias bare design toward L. I have two replies. First, assuming that such things are built into or are consequences of bare design, I fail to see why that would bias bare design toward L. After all, a designer might just as well have intent or have reason to bring about any among the range of possible universes. Second, even granting that these consequences would bias bare design toward L, my case could be modified slightly to avoid this worry. Replace my talk of “design” with a more neutral sense of the term, or something like “caused”, and my critique will avoid this worry.
On the third approach, we might infer inductively from general observations regarding designers that a cosmic designer would be at least modestly likely to share features of other designers, which would at least modestly bias bare design toward L. For example, designers tend to make structures that function in interesting ways. While we might worry about the reliability of an induction of this sort, let’s just grant it. Even if it is right, it would not undermine the second constraint, but would be other data on which we could update our confidence in various designer theses. Since it doesn’t undermine either of my constraints, it is not relevant to my critique. However, could the proponent of FTAs just incorporate this data to make a more successful argument? I don’t think so. In fact, it might even be some evidence against design, by narrowing the range of plausible BD∧L hypotheses at least slightly more than it does BND∧L hypotheses. There is a lot to say on this point, but for sake of space, I will not explore it further here.
4.3 Objection 3: This response still doesn’t explain where the original FTA goes wrong
Someone might worry, as with the stalking-horse response, that this response doesn’t explain where the FTA goes wrong. The answer is fairly simple, though: if the design hypothesis is BD (or some other hypothesis which doesn’t predict L more strongly than BD), then (2) (regarding the conditional probability of L on that hypothesis) is false. If the design hypothesis is a more substantive design hypothesis that more strongly predicts L, as it typically is, then (3) (the constraint on the priors) is false. The latter can be shown fairly trivially. The easiest way to see this is that on such a hypothesis, (1), (2), and (4) are true, while the conclusion (5) is false (per my counter-argument). Hence, because the argument is valid, (3) is false, that is, the constraint on the priors is false. This can be shown in other ways, but I will not include that here.
4.4 Objection 4: L provides a non-probabilistic reason to prefer design
The proponent of FTAs might grant the probabilistic point, but nevertheless think that there is some other reason to prefer certain design hypotheses. For example, perhaps certain design hypotheses better explain a world with fine-tuning, or otherwise are more theoretically virtuous than bare non-design or other non-design hypotheses given this data, and we should prefer them on these grounds.
While I do not agree with the suggestion here, it is irrelevant to the current discussion. My critique is against the sort of Bayesian FTA I described, and this response essentially involves granting my critique that said arguments fail. Whatever other argument for a designer someone wants to make, I will not consider them here.
4.5 Objection 5: This response proves too much
Consider the following example, which has been discussed by Luke Barnes.14 Suppose we were to observe that the stars in night sky where perfectly arranged so as to spell out the verses of the Gospel of John, for days at a time. Clearly, this would serve as decent evidence for theism, or at least for some sort of powerful agent that can control the stars. Here’s the general concern for a response to the FTA: if our response to the FTA would commit us to denying this sort of conclusion, then the response is in error.
For my purpose, then, there are two things to consider. First, does my response commit us to denying the aforementioned sort of conclusion? Second, is denying that conclusion actually problematic? On the first question, narrowly considered, I think my response actually does generate this result. We could frame a FTA-style argument in exactly the same sort of way, and I would give the same analysis. However, given other data, this observation would serve as decent evidence for theism (or at least some sort of powerful agent that can control the stars). For example, we have very good reason to think that linguistic displays are the result of language users, especially when they correlate strongly with other such displays known to be produced by language-users. Given that sort of data as background, the observation of the stars would be evidence in the way expected, and this is all compatible with my critique. On the second question, then, denying the conclusion is not problematic in this context. Nevertheless, the implied FTA-style argument would be unsound, and the expanded version which would be sound draws on additional data of a sort which is not obviously present in the FT case.
4.6 Objection 6: I’ve ignored multiverse and no universe possibilities
There were a few reasons for this. First, it made the analysis relatively simple. Second, it makes it clear that the critique doesn’t rely on any sort of multiverse objection to the FTA.15
Nevertheless, someone might complain that my analysis is incomplete, and possibly unfair to the FTA or to its opponents. For my purposes, I do not think that adding in multiverses to the possibility space will undermine my critique in any way, but I will save discussion of this for another post. I take it that the possibility of no universes doesn’t make a significant difference here, but that my critique could be trivially be extended to account for that in the same way regardless.
4.7 Objection 7: Is it just “chance”, then?
Someone might worry that my critique suggests that, on the denial of design, the fact that the universe is life-permitting is left to chance.16 However, we might think that that’s a very implausible view.
First, it’s not clear in what way this would undermine or contradict my critique. As such, it does not seem relevant. Second, however, I think this sort of critique indicates a bit of a confusion regarding the probabilities that are relevant here. There’s a difference between a conditional probability (i.e., how probable something is on a hypothesis) and a chancy process. On BND, the fact that L obtains isn’t (or at least needn’t be) the result of “chance”, per se, except in the sense that the world could have been many other ways, and the world could have been many other ways and satisfied BND. No “cosmic dice” were rolled, so to speak; this just is (possibly) the universe there is. Third, design involves “chances” in precisely the same way, in that the world could have been many other ways, and many other ways and satisfied BD. The fact that there is a designer of some sort, L obtains, and so on, is a matter of “chance” in precisely the same respect. To be clear, it is compatible with BND (and BD) that the fact that L is the result of some chancy process, but it’s not a consequence of the hypothesis.
5 A Fine Tuning Fork
What is the error in the FTA, then? As discussed in §4.3, the argument either involves a false conditional probability, or more likely, a false constraint on the priors. Either the given design hypothesis is too weak to more strongly predict L, or it is too strong so as to have a prior that suits the argument. This conclusion follows, as explained, from the constraints I presented and briefly defended in §4. Accordingly, if this is right, then the FTA under consideration is unsound, and provides no support for design over non-design.
Suppose that Alice is an open-minded agnostic, who is approximately neutral between the views that the universe is roughly as it is described by our best science and it was designed (perhaps by a designer which wants life), and that the universe is roughly as it is described by our best science and it was not designed, and has negligible credence in (incompatible) alternatives. In light of my remarks here, how might Alice respond to the FTA? The first thing to note is that, plugging these hypotheses into the FTA, the argument turns out to be trivially unsound. After all, by her (not so unusual) lights, there is a non-design alternative which adequately stalks the design hypothesis in question.
If Alice is asked to consider some substantive design hypothesis and compare if with bare non-design, ignoring her other evidence, she may remark, in line with my critique, that while that hypothesis does more strongly predict L, it has a lower prior such that it is antecedently not more likely than BND. Additionally, none of this is in tension with her agnosticism described in the preceding paragraph.
Of course, Alice (or anyone) might have other reasons to prefer, or may just be disposed to accept, design hypotheses over their non-design counterparts. But if she accepts my critique here, the fact that the universe is fine-tuned for life will be (at least) probabilistically irrelevant.
Collins, Robin (2009). The Teleological Argument: An Exploration of the Fine‐Tuning of the Universe. In William Lane Craig & J. P. Moreland (eds.), The Blackwell Companion to Natural Theology. Oxford, UK: Wiley‐Blackwell. pp. 202–281.
Barnes, Luke A. (2019). A Reasonable Little Question: A Formulation of the Fine-Tuning Argument. Ergo: An Open Access Journal of Philosophy 6.
I take it that this simplification is favorable to the proponent of fine-tuning arguments. I will discuss the relevance of multiverse hypotheses in a latter post. See also footnote 15.
Those critical of fine-tuning arguments might not grant this assumption, perhaps because of concerns regarding the physics, or more general concerns regarding the construction of the possibility space. While there may be concerns here with merit, I will not explore them here.
Oppy, Graham (2013). The Best Argument Against God. New York: Palgrave-Macmillan.
For example, see Capturing Christianity’s video, Luke Barnes and Alex Malpass Discuss The Fine-Tuning Argument for God, starting at about 49:40.
This is an idealization, as the predictive consequences of hypotheses often do not easily admit of precise expectations like this. We can allow that c be a rough degree of expectation, but for simplicity, I will think of it as a value between 0 and 1.
We could, if we want, even dispense with talk of “design”. The relevant minimal pro-FTA commitment is that there’s some additional entity the existence of which predicts, to some degree, L. This alteration can be made, and the rest of my discussion here will still apply, mutatis mutandis.
Those who might balk at possibility talk may not grant this approach. I probably share some of their concerns. For the sake of discussion, we can treat this as an idealization. I will discuss modality in future posts.
Barnes, Luke A. (2019). A Reasonable Little Question: A Formulation of the Fine-Tuning Argument. Ergo: An Open Access Journal of Philosophy 6.
Swinburne, Richard (2003). The argument to God from fine-tuning reassessed. In Neil A. Manson (ed.), God and Design: The Teleological Argument and Modern Science. Routledge. pp. 80--105.
Collins, Robin (2009). The Teleological Argument: An Exploration of the Fine‐Tuning of the Universe. In William Lane Craig & J. P. Moreland (eds.), The Blackwell Companion to Natural Theology. Oxford, UK: Wiley‐Blackwell. pp. 202–281.
Again, I will save elaborating and motivating this approach for another time.
I will update this if I find where.
For more on the multiverse objection, see e.g. Collins 2005, Holder 2006, Draper et al. 2007, Palonen 2008, Mawson 2011, Wilkinson 2013, Saward 2013, Barnes 2017, and Boyce & Swenson 2024.
Hey just a heads up i sent you a friend request on discord (my discord username is "thestuffofstars") because i would like to have a chat about your anti FTA argument, i think it's pretty interesting