Design and Fine-Tuning Arguments

Two quite different arguments travel under the name design, and the difference is about a century and a half wide.

The first points at living things and asks how anything so evidently built for a purpose could have arisen without a builder. It was the most persuasive argument in natural theology for two centuries and it was answered in 1859.

The second points at nothing in the world at all, but at the numbers in the laws that govern it, and observes that if several of them had been slightly different there would be no chemistry, no stars, and nothing complex enough to notice. That argument was not available until physics could say what would happen if the numbers were changed, which is to say until the 1970s, and it is the version now argued about.

The family

The name comes from telos, an end or goal. It marks a family rather than a single argument, and the members differ in what they point at, what they take the pointing to show, and what form of inference they use.

Two features separate this family from the cosmological arguments of the previous article. These arguments start from a particular feature of the world rather than from its bare existence, which means they can be defeated by a better explanation of that feature. And they aim at something more specific than a first cause: an intelligence, something with purposes, which is closer to what theism actually needs.

The argument from contrivance

William Paley's Natural Theology of 1802 gives it its most famous form. Crossing a heath, you find a watch. Unlike a stone, its parts are shaped and fitted together so as to produce a result, and no one who examined it would conclude it had always been there or had come about by the accumulation of accidents. The eye, Paley argues, shows the same marks to at least the same degree, and the inference from the eye to a maker is at least as good as the inference from the watch.

The argument is an analogy, and analogies are assessed by how similar the compared cases are in the relevant respects. This is the weakness Hume had already identified in the Dialogues Concerning Natural Religion, published in 1779, twenty-three years before Paley's book. The objections are still the standard ones.

Organisms are not much like machines. They grow, feed, repair themselves, and reproduce, and none of these is true of watches. The features that make organisms look designed sit alongside features that make them look like nothing anyone builds, and an analogy that selects the flattering half is not an argument.

We know that watches have makers because we have seen watches being made and have a large sample of artefacts and their origins. We have no sample of universes and their origins, and no way of forming one. An inference from a single case, with no comparison class, is not the kind of inference that supported the watch conclusion.

And the conclusion overshoots. Suppose the analogy holds. Artefacts are commonly made by teams, by beings of finite power who work by trial and error, whose earlier attempts fail, and who die. The evidence supports a designer with those features as readily as one with the attributes of classical theism, and the argument as stated cannot distinguish them.

Darwin's On the Origin of Species in 1859 did something the objections had not. It supplied a mechanism by which the appearance of contrivance arises without contrivance, through variation and differential survival, and it did so in the very cases the argument relied on. This is why the biological version is not merely disputed but abandoned by nearly everyone who works on this subject. The considerations Hume raised weaken an inference; a working alternative explanation removes its premise.

Two things worth keeping straight follow.

Darwin removed biology as a source of design premises. He did not touch the question of why there are laws under which variation and selection produce anything at all, and the modern version of the argument moved to exactly that ground.

And the form of the argument changed. Nobody now argues by analogy. The serious versions claim that design is the better explanation of some feature, or, in the probabilistic form set out in the beginner tier, that the feature is more to be expected if there is a designer than if there is not. That form is not vulnerable to the weakness of the analogy, and it acquires a different vulnerability, since it now requires an assessment of how expected the evidence is on the alternatives.

The argument from the constants

The term of art needs handling with care. In physics, saying a parameter is fine-tuned means that outcomes are highly sensitive to small changes in it. It does not mean that anyone tuned it, and the argument has to earn that, not assume it by borrowing the word.

The examples usually cited include the strength of gravity relative to electromagnetism, the strength of the strong nuclear force relative to electromagnetism, the masses of the up and down quarks, the strength of the weak force, and the cosmological constant, which is the most discussed case because the discrepancy between its observed value and the value naive theory suggests is enormous. To these are added conditions of the early universe: an energy density extraordinarily close to a critical value, a very specific amplitude of density fluctuations, and an initial entropy so low that its improbability on standard assumptions is difficult to write down.

Two honest qualifications belong here, and they cut in both directions.

These are claims within physics, they are disputed within physics, and their strength varies from case to case. Non-specialists on both sides of the theological argument state them with more confidence than the physics literature supports.

And the general phenomenon is not a theological invention. Physicists who have no interest in the design question take the sensitivity of the parameters seriously, and it is one of the motivations for the multiverse proposals discussed below. Someone who dismisses fine-tuning as a religious talking point has not looked at where the discussion came from.

The argument's structure is a likelihood comparison. Life-permitting values are extremely improbable if the values were set by nothing at all, and are not improbable if they were set by something that wanted life. So the values are evidence for design. Since it is a comparison, it needs both terms, and most of the debate is about whether either of them can be stated.

Observers and what they can observe

Brandon Carter introduced the term in 1974, in two versions. The weak version states the restriction just given: where and when we observe from is necessarily compatible with our existing. The strong version, which is more contentious and often misread, says that the universe must be such as to admit the creation of observers within it. Carter's own view was that the strong version highlights the same kind of observational bias as the weak one rather than asserting a cosmic requirement.

The objection built on it is direct. Given that we exist and are making observations, we could not have observed anything other than life-permitting values. So the observation is guaranteed whether or not there is a designer. A piece of evidence that both hypotheses predict with certainty discriminates between them not at all, and the fine-tuning argument therefore has no force. Elliott Sober has pressed this version, arguing that once observation selection is taken into account both likelihoods go to one.

The standard answer is John Leslie's firing squad, from Universes in 1989. Fifty trained marksmen fire at you and you are unharmed. You could not have been observing anything else, since a hit would have ended your observing. Yet it would be obtuse to conclude that no explanation is called for and that you simply had to find yourself alive. The reasonable response is to suspect that they missed deliberately. So the presence of an observation-selection effect does not by itself neutralise a surprising observation.

The counter is that the analogy imports a background the cosmic case lacks. In the firing squad you know that marksmen exist, that they rarely miss, and that people sometimes arrange for them to miss, and it is that background that supports the inference. Matthew Kotzen has argued that the selection bias should be treated as evidence rather than as background information, which changes the calculation, and there is no agreement in the literature on how the case should be handled.

What can be settled is a smaller point that gets confused with the large one. The weak anthropic principle is true, and it is not an explanation. Saying that we observe life-permitting values because we could not have observed others explains why our observation came out as it did, given that the values are what they are. It does not explain why the values are what they are, and treating it as though it did is a mistake made frequently and by people who should know better.

Many universes

If the ensemble is large enough and the variation wide enough, some members have life-permitting values, and any observer is necessarily in one of those. The improbability of any particular universe being life-permitting is then no more surprising than the fact that some lottery ticket wins.

An objection often made against this is that it is invented to escape the argument, and that answer is not available. Multiverse proposals arise from independent physics, from eternal inflation and from the large number of possible vacuum states in string theory, and they were not proposed to answer natural theology. Whether that physics is correct is another matter, and it is far from established. But the hypothesis is not ad hoc in the way the complaint suggests.

Three objections have real weight.

The inverse gambler's fallacy. Ian Hacking identified the pattern in 1987 and Roger White applied it to this debate in 2000. Suppose you walk into a casino, watch one roll of the dice, and see a double six. Inferring that many rolls must have preceded yours is a fallacy: earlier rolls have no effect on this one. Applied here, the existence of other universes does nothing to make this universe more likely to be life-permitting, since the values here were not drawn from a pool affected by what happened elsewhere.

The reply, developed after Hacking's paper, is that the casino analogy misdescribes the case. You could have walked in and seen any roll, whereas we could not have found ourselves in any universe but a life-permitting one. Our sample is not a random draw and the comparison fails.

That reply should be noticed carefully, because it puts the two sides in a curious position. It is exactly the observation-selection reasoning that the fine-tuning defender rejects in the previous section, now deployed by the multiverse defender. Anyone who accepts observation selection to rescue the multiverse from the inverse gambler's fallacy has accepted the premise of Sober's objection to the design argument, and anyone who rejects it to save the design argument has given the fallacy charge back its force. The two disputes are the same dispute, and positions that treat them separately are unstable.

The measure problem. Multiverse models that are infinite need a way of saying what is typical among infinitely many observers, and the mathematics of regularising those infinities is unsolved. Without a measure, a multiverse model does not predict that we should observe what we observe, or anything else. One prominent cosmologist has called this the deepest crisis in the subject. This does not refute the hypothesis, and it does undercut confident use of it as an explanation of anything.

It does not answer everything. A multiverse explains the values of the constants by making all of them occur somewhere. It does not explain why there is an ensemble, why the mechanism generating it has the character it has, or why the laws that govern the variation are what they are. The demand can be raised again one level up, which is not a knockdown, and does mean the hypothesis relocates the question rather than closing it.

The problem of assigning the probability

An objection applies to the argument's own first step, and it is technical without being obscure.

To say that life-permitting values are improbable, you need a probability distribution over the possible values. If the range of possible values for a constant is unbounded, there is no uniform distribution over it, and the ratio of a narrow interval to an infinite range is not a defined quantity. The claim that the life-permitting window is a vanishing fraction of the possibilities then has no content as stated.

Defenders respond by restricting the range on physical grounds, on the basis that values beyond some bound are excluded by theory. Critics respond that the restriction is chosen and that the result depends on the choice, which is the same complaint made in the previous article about priors selected to produce a conclusion.

The point cuts in an interesting direction. It is usually pressed against the theist, and it is equally awkward for the confident atheist who says the values are unremarkable, since that claim also requires a distribution nobody has. The defensible position for either side is that the probabilities here are not well defined, which is a reason for less confidence all round rather than a victory for anyone.

Next

The two arguments so far both start from something observed: that there is a universe, and that it has the character it has. The next article takes up the argument that needs no observation at all, which claims that the concept of God is sufficient to establish that God exists, and which has been declared dead and revived more often than anything else in this subject.