The Evolutionary Argument Against Naturalism

The previous rung ran evolutionary biology against theism. This one runs it in the other direction, and the claim is more ambitious than anything met so far in this tier. It is not that naturalism is improbable, or that some evidence counts against it. It is that a naturalist who understands their own position has thereby acquired a reason to distrust every belief they hold, including that one.

If the argument works, naturalism cannot be rationally accepted. Not because it is false, but because accepting it undermines the standing of the acceptance. That is an unusual kind of failure, and the first task is to see clearly what would have to be true for it to occur.

Darwin's doubt

The seed of the argument is a remark of Darwin's, and Plantinga names it after him. Writing to William Graham on 3 July 1881, Darwin reported a doubt that recurred to him: whether the convictions of a mind developed from the minds of lower animals are worth anything, or are trustworthy at all. He put the point by asking whether anyone would trust the convictions of a monkey's mind.

Darwin did not develop this into an argument, and he did not draw a theistic conclusion from it. But the shape of the worry is exactly the shape of what follows. Our cognitive faculties have a history. That history was not aimed at truth. And if the process that produced an instrument was not aimed at accuracy, its accuracy stands in need of some further explanation.

The worry is easy to state and surprisingly hard to dismiss, because the obvious replies turn out to use the very faculties whose standing is at issue.

The claim

The argument's target is a conjunction, and both halves matter. Evolution alone is not the problem, and Plantinga has always insisted he has no quarrel with the biology: a theist can hold that God brought about the human mind through evolutionary means and that the process was aimed, however indirectly, at producing knowers. Naturalism alone is not obviously the problem either. What the argument attacks is the conjunction, on which the process that produced our faculties had no aim at all.

That is also why the conclusion is not simply that naturalism is false. If the argument works, the naturalist has a defeater, which is a reason to give up a belief rather than a demonstration that it is mistaken. The book title of the collection assembled around the argument, Naturalism Defeated?, edited by James Beilby (Cornell University Press, 2002), is chosen carefully: defeat, not refutation.

What is meant by reliability

The domain-relativity is worth flagging at the outset, because it does a great deal of work in the objections later. The question the argument poses is what probability to assign to the proposition that our faculties are reliable, given naturalism and unguided evolution. Plantinga's answer is that the probability is low, or else that it cannot be assessed at all, and either answer is enough for his purposes.

Why the probability might be low

The engine of the argument is a claim about what natural selection can see.

The reasoning behind the claim runs as follows. What selection acts on is behaviour, since behaviour is what bears on survival and reproduction. Beliefs matter to selection only insofar as they produce behaviour. So what selection favours is whatever internal states produce fitness-enhancing behaviour, and whether those states are also true is a further question that selection has no direct grip on.

Plantinga presses the point with cases. A creature that flees a predator because it believes the predator is dangerous behaves the same as one that flees because it believes the predator is a large friendly object best approached by running away, and which then has a further belief that the best way to approach something is to run away from it. The pair of false beliefs generates the same behaviour, and selection cannot distinguish the two creatures. Extend the point across a whole system of belief and it seems that many different assignments of content would produce identical behaviour, and only a small proportion of those assignments are mostly true.

The claim needs a further step to reach the whole belief system, and Plantinga supplies it by considering what naturalism says beliefs are. If naturalism is true, materialism about the mind is very probably true, and on materialism a belief is a physical structure in the nervous system. Such a structure has two sorts of property: neurophysiological ones, which are what cause the muscle movements, and content properties, which are what the belief is about. It is the first sort that does the causal work in producing behaviour. The content is carried along.

If that is right, then the content of our beliefs is invisible to selection, and the probability that the content is largely true is whatever it would be if content had been assigned at random. There is no reason to think that probability is high.

What "unguided" is doing

One clarification prevents most of the confusion the argument attracts, and it concerns a word in the setup.

Evolutionary theory, as a scientific theory, does not say that the process is unguided. It describes mechanisms: variation, inheritance, differential reproductive success. Whether the whole is superintended by anything is not a question the theory addresses, and could not be, given the methodological naturalism of the previous rung. The claim that evolution is unguided is a metaphysical addition, supplied by naturalism rather than by biology.

Plantinga agrees with all of this, and it is why the argument targets the conjunction. He is explicit that a theist can accept the entire biological account, including common descent and selection as the principal mechanism, and simply deny that the process was unaimed. His larger thesis in the 2011 book is precisely this: that the apparent conflict between science and theistic religion is superficial, and that the deep conflict is between science and naturalism, since it is naturalism that has trouble making sense of our being able to do science at all.

Whether that reversal succeeds is what the rest of this rung is about. But the setup has to be granted before it can be assessed, and critics who reply that evolution is well confirmed have not addressed anything the argument says.

Branching on the mind

The step from selection acting on behaviour to a low probability for reliability needs the intermediate claim about content, and it is worth laying out the alternatives that claim is meant to eliminate.

Suppose beliefs have no causal role in behaviour at all. Then their content is entirely unconstrained by anything selection has acted on, and there is no reason at all to think it mostly true.

Suppose instead that beliefs do cause behaviour, but by virtue of their physical properties rather than their content, which is the semantic version of the same idea. The result is the same. What the physical structure causes is fixed by its physical characteristics, and the content is a further fact about it that selection has not tested.

Suppose finally that content is itself causally efficacious. Even then, two further things have to hold before reliability follows. The content-driven behaviour has to be adaptive, and the adaptive content has to be true. Neither is automatic: a system whose contents drive behaviour could be driven towards behaviour that reduces fitness, and content that drives fitness-enhancing behaviour need not be accurate, as the predator example was meant to show.

Plantinga's conclusion is that only the last of these branches yields reliability, that it is one option among several, and that naturalism gives no reason to expect it rather than the others. The objection developed below denies that the branches are genuine alternatives, on the grounds that what content is rules out the first two by definition rather than by good fortune.

The argument

Each premise has been attacked, and the shape of the disagreement is different in each case.

Two kinds of defeat

The distinction matters because the evolutionary argument is offered as an undercutting defeater and is often criticised as though it were a rebutting one. Plantinga is not arguing that our faculties are unreliable. He is arguing that the naturalist's grounds for thinking them reliable are removed, and that this is enough. A person in the trick lighting who goes on believing the wall is red is not believing something they have evidence against; they are believing something they no longer have any reason to hold.

The parallel with the verification principle is worth holding on to, because it shows both the force and the limit of this kind of criticism. The verificationist could not exempt their own principle from the standard it set. The naturalist, on this argument, cannot exempt their belief in naturalism from the doubt that naturalism generates. In each case the failure is structural rather than empirical, which is what makes it interesting, and in each case the natural response is to look for a way of restricting the scope so that the position exempts itself, which is what the objections below attempt.

Notice also that theism is not touched by the parallel argument. On theism, human beings have faculties designed by a being who knows everything and has some interest in their knowing things, so the probability of R given theism is high. That is not an argument that theism is true. It is a claim that theism is not in the same predicament, and it is what makes the argument comparative in the sense the previous two rungs established.

The main objection

The strongest reply denies the second premise by denying that content is idle.

On the standard naturalistic accounts of mental content, the content of a belief is not an extra property riding on top of the neurophysiology. It is constituted by the functional or causal role of the state, including what typically causes it and what it typically leads the organism to do. On such an account, a belief that there is a predator nearby just is a state that is reliably caused by nearby predators and reliably leads to avoidance. If that is what content is, then a state's having true content and its producing appropriate behaviour are not independent, and the fanciful assignments of content in Plantinga's examples are not genuine possibilities but descriptions that fail to attach to any real state.

The reply also has a simpler version that does not require any theory of content. Systems that track features of the environment accurately are, in general, better at producing behaviour appropriate to that environment than systems that do not, and the cheapest way to build a system that behaves appropriately across a wide range of unpredictable situations is to build one that represents the situation correctly. A creature with elaborately compensating false beliefs would need a separate compensation for every situation, which is more complicated and more fragile than getting it right.

Plantinga's response is that this presupposes exactly what is at issue. Accounts of content on which content is fixed by causal or functional role are themselves naturalistic theories, arrived at by the faculties under suspicion, and appealing to them to reassure oneself about those faculties is circular. And the simpler version, he argues, understates how much behaviour can be produced by systems that are not representational at all: a great deal of successful animal behaviour is produced by mechanisms that no one is tempted to call believing.

Whether the charge of circularity sticks is the deepest question in the debate. If it does, it appears to apply to any attempt to establish the reliability of one's faculties from within, including the theist's: the argument that God would have given us reliable faculties is itself a piece of reasoning conducted with those faculties.

The probabilistic objection

A second line of attack targets the numbers, and the most influential statement of it is Branden Fitelson and Elliott Sober's "Plantinga's Probability Arguments Against Evolutionary Naturalism" in the Pacific Philosophical Quarterly in 1998.

Their central complaint is that the argument requires a probability that nobody is in a position to assign. The proposition in question concerns the probability of general reliability, given naturalism and evolution, assessed from a standpoint that includes no information about ourselves, and it is not clear what could ground any number here, low or high. They also press the point that a similar sceptical manoeuvre can be run against almost any position that offers an account of its own origins, so that the argument, if successful, would prove far more than intended.

Plantinga's fallback is the inscrutability option. If no probability can be assigned, he argues, that is itself sufficient: someone who recognises that they have no way of assessing whether their faculties are reliable is in no better position than someone who has assessed the probability as low.

That fallback is contested in turn, and the dispute is about the theory of defeat rather than about evolution. On some accounts, a defeater requires a positive reason to doubt, and the mere absence of a supporting reason is not one; on others, the absence is enough. The objection and the response therefore pass each other unless a theory of defeat is settled first, which is one reason the literature has not converged.

The tu quoque

A third objection grants a great deal and asks whether the theist is better off.

Everyone, theist and naturalist alike, forms the belief that their faculties are reliable using those faculties. Everyone therefore faces a circularity if asked to justify that belief from scratch. The theist's escape route is that on theism the faculties were designed to be reliable, but that proposition is itself believed on the basis of the faculties in question, so the escape is only available to someone already entitled to trust them.

The theist's response is that this misunderstands what the argument claims. It does not claim that theism provides a non-circular justification of reliability. It claims that theism, unlike naturalism, does not generate a defeater. Everyone starts by trusting their faculties, because there is no alternative; the question is whether anything one comes to believe later undermines that starting trust, and Plantinga's claim is that naturalism does and theism does not.

That response is coherent, and it clarifies what is at stake: the argument is about which positions are stable under reflection on their own implications, not about which can be established from a neutral standpoint that nobody occupies.

Local reliability

The most deflationary objection concedes the argument for a restricted domain and denies that it generalises, and it uses the domain-relativity noted earlier.

Selection plausibly does favour accuracy about the immediate physical environment, since misjudging where a branch is has direct fitness consequences. What it gives no reason to expect is accuracy about matters remote from ancestral survival: theoretical physics, higher mathematics, and metaphysics included. So the sceptical conclusion should be local rather than global.

This is a real limitation, and it may be worse for both parties than either wants. A locally restricted conclusion does not defeat the naturalist's ordinary beliefs, so the dramatic self-defeat result disappears. But the beliefs it does put in question are precisely the abstract ones, and the belief in naturalism is an abstract one, as is the belief in theism, and as is the evolutionary argument itself. The restricted version generates a milder scepticism with a wider reach into philosophy, which is uncomfortable for everyone conducting the debate.

Where it sits in the larger project

The argument is the last piece of a structure whose earlier pieces appeared at rung 10, and seeing the whole makes its purpose clearer than treating it in isolation does.

The account of warrant developed there defines it in terms of faculties functioning properly, in an environment they suit, according to a design plan aimed at truth. That account was constructed to answer the objection that religious belief without argument is irrational. But the notion of proper function it uses has an obvious question attached: what makes it the case that a faculty has a design plan, and that the plan is aimed at truth, if nobody designed it? Plantinga's answer has always been that naturalism has no good account of this, and that the notion is at home in a theistic picture in a way it is not in a naturalistic one.

The evolutionary argument is the offensive version of that defensive claim. Rather than showing that theistic belief escapes an epistemic objection, it argues that naturalistic belief does not. The pattern is deliberate: the naturalist has been asking the theist to justify their faculties, and the reply is that the demand recoils.

Reading the argument this way also explains why the exchange has been so difficult to settle. Both sides are operating with different theories of when a belief is in good standing, and each theory was developed partly with an eye to the outcome of this dispute. That is not a charge of bad faith; it is a description of how philosophical positions in a mature debate tend to be shaped.

What the argument establishes

The argument has not persuaded many naturalists, and it would be misleading to present it as an unanswered result. What is fair to say is more limited and more interesting.

It has clarified what a naturalist owes. A naturalist who wants to avoid the conclusion has to hold some account on which mental content is not idle, and that commits them to a substantive position in the philosophy of mind rather than leaving the matter open. The argument thus converts a vague worry into a specific bill.

It has established a symmetry that is easy to miss. The intermediate tier's rung on naturalistic explanations of religious belief runs a debunking argument in the other direction: religious belief has a causal history in agency detection and social cognition that does not run through the truth of the belief. The evolutionary argument is the mirror image, and the criteria for when such arguments succeed have to be the same in both cases. Anyone who accepts one and dismisses the other owes an account of the difference.

And it has not established that naturalism is irrational, because the second premise remains genuinely disputed and the theory of defeat it relies on is not settled. The honest summary is that the argument identifies a real tension in the naturalist position, that the tension can probably be relieved by an account of content that most naturalists already hold for independent reasons, and that whether the relief is question-begging depends on a general question about circularity that touches every position equally.

Next

The tier now returns to evil, with the version that most of the recent literature has been organised around. Rowe's argument infers from cases of suffering in which no justifying good can be identified to the probable existence of suffering for which there is none, and the reply known as CORNEA questions whether that inference is ever entitled. The next rung sets out the inference, the restriction, and how theodicies enter the calculation as competing likelihoods.