New Draft Plantinga Paper
A work in progress – posted for comments – NOT TO BE COPIED OR REPRODUCED ELSEWHERE!
Plantinga’s Belief-Cum-Desire Argument Refuted
In the final chapter of Warrant and Proper Function , Plantinga argues that, if both:
(N) naturalism – the view that there are no supernatural beings
(E) evolution – current evolutionary doctrine
are true, then the probability that:
(R) our cognitive faculties are reliable and produce mostly true beliefs
must be either low or inscrutable.
Plantinga argues, further, that this argument furnishes anyone who accepts N&E with a undefeatable defeater for any belief produced by those faculties, including N&E itself. Hence, N&E has been shown to be self-defeating.
One part of this larger argument is what I call Plantinga’s belief-cum-desire argument. The belief-cum-desire argument is designed to show something more specific – that if the content of our beliefs does causally affect behaviour, and N&E, then the probability of R cannot be high.
Critics of Plantinga’s larger argument against N&E have usually tended to concede, whether or not for the sake of argument, that the conclusion of the belief-cum-desire argument is true, and have instead rejected Plantinga’s suggestion that his larger conclusion that N&E is self-defeating does not follow (or, if it does, much the same problem plagues theism). My aim here is different – I aim simply to refute the belief-cum-desire argument.
Plantinga’s belief-cum-desire argument
Suppose some hypothetical rational creatures a lot like us evolve on a planet a lot like Earth – they “hold beliefs, change beliefs, make inferences, and so on” . Suppose:
(C) causal efficacy – the content of beliefs causally affects behaviour
is true. What is the probability of R/N&E&C specified with respect to these creatures – what is the probability that their cognitive faculties are reliable?
The probability, says Plantinga, is not as high as you might initially be tempted to suppose. For it is not belief per se that is adaptive, but the behaviour it produces. And behaviour is caused by both belief and desire. But then, claims Plantinga, for any given adaptive action (action that enhances the creatures ability to survive and reproduce),
there will be many belief-desire combinations that could produce that action; and very many of those belief-desire combinations will be such that the belief involved is false.
Plantinga illustrates like so:
So suppose Paul is a prehistoric hominid; a hungry tiger approaches. Fleeing is perhaps the most appropriate behavior: I pointed out that this behavior could be produced by a large number of different belief-desire pairs. To quote myself: “Perhaps Paul very much likes the idea of being eaten, but when he sees a tiger, always runs off looking for a better prospect, because he thinks it unlikely that the tiger he sees will eat him. This will get his body parts in the right place so far as survival is concerned, without involving much by way of true belief. . . . . Or perhaps he thinks the tiger is a large, friendly, cuddly pussycat and wants to pet it; but he also believes that the best way to pet it is to run away from it. . . . or perhaps he thinks the tiger is a regularly recurring illusion, and, hoping to keep his weight down, has formed the resolution to run a mile at top speed whenever presented with such an illusion; or perhaps he thinks he is about to take part in a 1600 meter race, wants to win, and believes the appearance of the tiger is the starting signal; or perhaps . . . . Clearly there are any number of belief-cum-desire systems that equally fit a given bit of behavior.”
So adaptive behaviour can be produced by many belief-desire combinations and, “in many of these combinations, the beliefs are false” . We cannot, concludes Plantinga, estimate the probability of R on N&E&C as high. And of course, if we cannot estimate that probability as high for these hypothetical creatures, then we cannot estimate it as high in our own case either.
The above argument that the probability of R given N&E&C cannot be high has some superficial plausibility. Plantinga is surely correct that:
(i) it is behaviour that evolution selects for rather than beliefs per se.
He is also correct that:
(ii) for any piece of adaptive behaviour, there are many belief-desire combinations that might produce it, on many of which the belief or beliefs in question are false.
However, I will show that, appearances to the contrary, it does not follow from (i) and (ii) that we cannot reasonably estimate the probability of R on N&E&C as being high. Indeed, I shall go further, and sketch out some reasons for supposing that the probability of R given N&E&C must, in fact, be fairly high.
Refutation of Plantinga’s belief-cum-desire argument
Consider two possible scenarios:
(a) we have evolved certain false beliefs and certain desires that, in combination, result in adaptive behaviour
(b) we evolved certain unreliable belief-producing mechanisms and certain desires that, in combination, result in adaptive behaviour.
Perhaps, on N&E&C, (a) is not so unlikely, for the reasons Plantinga cites. Suppose I have an innate belief that tigers are cuddly and that best way to pet a tiger is to run away from it. If I am also equipped with an innate desire to pet tigers, this results in adaptive behaviour.
But what about (b)? How likely is it on N&E&C that our belief-producing mechanisms are unreliable? Consider the question: what particular set of desires would a species need to evolve in order for the beliefs generated by such an unreliable mechanism to result in generally adaptive behaviour? We will look at some examples, beginning with the cognitive faculty of reason.
Example one: fallacy of affirming the consequent
Consider the fallacy of affirming the consequent (FAC). The FAC is an unreliable form of inference. It sometimes produces true conclusions, but often false.
Suppose evolution hard-wires a species of hominid H to be highly prone to the FAC. Suppose a member of this species, H1, concludes using the FAC that jumping out of planes is not safe. Another member of the same species, H2, concludes using the FAC that jumping off tall buildings is safe. They might reason like so:
If jumping out of planes is not safe, jumping out of balloons is not safe
Jumping out of balloons is not safe
Jumping out of planes is not safe
If jumping out of planes is safe, then jumping out of planes wearing a parachute is safe
Jumping out of planes wearing a parachute is safe
Jumping out of planes is safe
If evolution hard-wires a desire into species H to make H2’s resulting belief that jumping out of planes is safe adaptive – e.g. a powerful desire to commit suicide – that same hard-wired desire will result in the likely death of H1.
What set of desires must evolution instil in species H to render adaptive the potentially-mal-adaptive consequences of applying the FAC? There is none. But then evolution cannot render this unreliable form of inference adaptive by instilling a particular set of desires in H.
The FAC sometimes produces false beliefs, but sometimes true. Is that the reason why evolution cannot render the FAC adaptive? Could a method of inference that consistently produced false conclusions from true premises be made adaptive by pairing it with an appropriate set of desires? No, as I explain below.
Example two: Counter-induction
Suppose that two hominids appear: hominid A and hominid B. A reasons inductively, B counter-inductively, like so:
A observes that whenever other hominids eat, they usually continue to live, and when they stop eating, they die. He concludes that if he eats, he’ll probably continue to live, and if he stops eating, he’ll die.
B observes that whenever other hominids eat, they usually continue to live, and when they stop eating, they die. He concludes that if he eats, he’ll probably die, and if he stops eating, he’ll continue to live.
When A applies his form of reasoning to true premises, he is likely to end up with a true belief. B on the other hand, is likely to end up with a false belief. His counter-inductive method of reasoning consistently produces false beliefs.
However, if evolution equips A with a desires to live, and B with a desires to die, B’s false belief is just as adaptive as A’s true belief. So while counter-induction has delivered a false belief, it has not delivered behaviour that is mal-adaptive. In fact, it has delivered behaviour just as adaptive as that delivered by induction.
So far, it seems that Plantinga is correct: given evolution equips A and B with the right desires, the behaviour produced by the two belief-forming mechanisms is equally adaptive.
But now suppose A and B engage in further reasoning, applying their respective methods of inference like so:
A observes that other hominids that forage and hunt get food to eat, and those who don’t get none. A concludes that if he hunts and gathers, he’ll get food to eat, and if he doesn’t he’ll get none.
B observes that other hominids that forage and hunt get food to eat, and those that don’t get none. B concludes that if he doesn’t hunts and gather, he’ll get food to eat, and if he does, he’ll get none.
Now, A’s reasoning helps him survive. Given his desire to live, these two inferences together will lead him to hunt and gather. That’s adaptive behaviour.
The problem is, given the desire required to get B’s first counter-inductive inference to produce adaptive behaviour, B’s second counter-inductive inference is now likely to produce mal-adaptive behaviour. Given B’s desire to die, plus his false belief that eating will kill him, his second counter-inductively generated conclusion will no doubt lead him not to go hunting and gathering. That’s not adaptive behaviour. B will probably starve to death.
Plantinga is correct that for any piece of adaptive behaviour, there are many belief-desire combinations that might produce it, on many of which the belief or beliefs in question are false. But it does not follow that the probability of R given N&E&C cannot reasonably be estimated as high. The members of a species equipped with unreliable belief-forming mechanisms such as the FAC or counter-induction will end up with all sorts of combinations of false beliefs the mal-adaptive consequences of which cannot be neutralized by evolution hard-wiring that species with some particular set of desires.
In fact there are two difficulties here.
First, there is the problem of novel beliefs. An advantage of procedural reasoning is that it allows for creatures able to problem solve and adapt, within their own lifetimes, to a changing environment and novel situations. An adaptive inferential mechanism is likely to applied in new ways. But then evolution cannot then anticipate what desires will be required to render adaptive the innumerable potentially mal-adaptive conclusions likely to be drawn. If B draws the first counter-inductive conclusion, his desire to die renders his conclusion adaptive. But if B happens to go on to draw that second conclusion using the same unreliable form of inference, that very same desire now renders the conclusion mal-adaptive.
The second problem is – not only can evolution not anticipate which desires creatures will need to render the conclusions of an unreliable mechanism adaptive – when it comes to unreliable forms of inference, there just is no set of desires that will render the mechanism adaptive. A set of desires that renders one set of conclusions adaptive will render another set of conclusions generated by the same mechanism mal-adaptive.
On the other hand, evolution can make reliable forms of inference adaptive in a straightforward way, by equipping the species in question with desires for those things that enhance its ability to survive and reproduce. In which case, the probability that reliable forms of inference will evolve, as opposed to an unreliable forms of inference, looks to be high.
Other cognitive faculties
The considerations sketched out above suggest that N&E&C should lead us to estimate the probability that our cognitive faculty of procedural reasoning is reliable as fairly high. But of course, procedural reason alone furnishes us with little, if any, knowledge. Other cognitive faculties – mostly notable perception and memory – must also come into play.
How reasonable is it, given N&E&C, to suppose that these other faculties are reliable? If there is no good reason to suppose they are reliable, then there’s no good reason to suppose our various faculties working in conjunction constitute a reliable belief-forming system. My car may have a reliable carburettor, but if other parts are unreliable, the car as a whole remains unreliable.
So let’s now look at the cognitive faculties of memory and perception. Has Plantinga shown that, given N&E&C, the probability that these other faculties are reliable connot be high?
Suppose hominid species H is equipped with an unreliable memory. Hominid H1 has at time t1 true beliefs A and B. But, because H1’s memory is unreliable, she later believes the falsehoods not-A and not-B. Is there a desire or set of desires with which evolution might also equip species H that will render adaptive the behaviour produced by these two resulting false beliefs? Very probably. If A is the belief that if you eat you will live and B the belief that if you don’t eat you will die, these beliefs will result in adaptive action if H1 desires to die. However, because H1 previously believed A and B, she would previously have not eaten, which is mal-adaptive behaviour. There is no set of desires that will make both the input and output beliefs of this unreliable faculty result in adaptive behaviour. But then unguided evolution cannot equip species H with a set of desires that will make the input and output beliefs of this unreliable memory faculty generally adaptive. Evolution can, on the other hand, equip a species with a set of desires that will make the input and output beliefs of a reliable faculty generally adaptive. It appears, then, that N&E&C will therefore strongly favour a reliable memory faculty over an unreliable faculty.
How likely is it, on N&E&C, that evolution would produce a species with a reliable perceptual-mechanism-cum-desire combination, rather than an unreliable-perceptual-mechanism-cum-desire combination?
Fairly likely, I suspect. There are two categories of unreliable perceptual mechanisms:
(1) Unreliable mechanisms producing mostly false beliefs.
(2) Unreliable mechanisms that produce significant proportion of, but not mostly, false beliefs
Let’s begin by considering perceptual or quasi-perceptual mechanisms of type (1). Such mechanisms fall, in turn, into two categories:
(1a) Unreliable mechanisms producing mostly false beliefs but in a systematic, predictable way.
(1b) Unreliable mechanisms producing mostly false beliefs in a random, unpredictable way
An example of (1a) would be a perceptual or quasi-perceptual mechanism that, whenever the subject is presented with a tiger, produces the belief there is a rabbit present. There is consistency to the error. An example of (1b) would be a perceptual or quasi-perceptual mechanism that, when the subject is presented with a tiger, may the first time produce the belief there is nothing present, the next time the belief a rabbit present, the next time the belief there is a chair present, and the time after that the belief there’s a side of beef present, etc., but rarely if ever the belief that there is a tiger present. While we can predict that the subject will make an error about there being a tiger in front of them, it is not possible, even given knowledge of the erroneous beliefs previously produced when a tiger was present, to predict what erroneous belief will now be produced on this occasion.
Can unguided evolution make an unreliable mechanism of type (1b) produce adaptive behaviour by combining it with an appropriate set of desires? It is hard to see how. If there is a tiger present and the mechanism makes me believe there is a rabbit present, my mistaken belief can still result in adaptive behaviour if evolution has given me a desire to run away from rabbits. But if the erroneous beliefs are being generated in a random way, there will be no particular desire or set of desires with which evolution might equip me that will make the random false beliefs generated by this mechanism adaptive.
What about a mechanism of type (1a)? Does the pattern to the errors produced by the mechanism mean that evolution can render the mechanism adaptive by combining it with an appropriate set of desires?
That suggestion might seem plausible when we consider a very simple example of adaptive behaviour, such as running away from tigers. If the mechanism systematically produces the belief that a rabbit is present whenever a tiger is present, all evolution need do is instil in these subjects a powerful desire to run away from rabbits.
But the suggestion becomes far less plausible when we consider more we consider sophisticated patterns of adaptive behaviour.
Suppose, for example, that to reach food you need to survive, you need to engage in some team activity with other members of your species – e.g. negotiating some tricky terrain that includes a narrow ledge and a poisonous snake. Someone has to distract the snake while someone else crawls carefully along the edge and leaps over the snake at the exact moment it is distracted.
Now try to imagine a perceptual mechanism of type (1a) that produces false mostly beliefs about your surroundings, but beliefs that, when paired with certain desires with which evolution has pre-equipped your species, will result in the required adaptive behaviour from you and your team mates.
You must not believe there is a snake and a ledge and some food and some team mates with whom you must co-operate. And nor must your team mates. You, and they, must have mostly false beliefs about your environment, but beliefs that, nevertheless, when paired with desires with which evolution has collectively furnished you, lead you to act in tandem with your other team members to retrieve and eat the food.
In fact, setting aside the challenge of imagining such a mechanism, it is a difficult enough challenge to construct just a set of mostly false beliefs and hard-wired desires that would result in the complex sequence of actions required. Perhaps it is not impossible. Perhaps your (1a) type mechanism causes you to believe that instead of food at the end of the ledge, there’s a little man who will give you a tickle stick if you walk carefully along a white line, jumping in the air after 15 seconds, and then reach down and take the stick. Perhaps you believe that eating the tickle stick is the best way to get tickled. If we pair this false belief with a desire to be tickled, your resulting sequence of actions might yet be adaptive. You might successfully negotiate the narrow ledge, leap over that snake (though who is going to distract it?) and then eat the food.
However, even if we can come up with a mostly-false-perceptual-belief-cum-desire combination that would, in this situation, result in adaptive action, it is still more difficult still to come up with a belief-forming mechanism of type (1a), which, paired with an appropriate set of desires, will result in sophisticated patterns of adaptive behaviour generally of the sort of which we are capable. If the next time the food lies beyond a precipice that can only be negotiated if you and your team place a tree trunk across the gap, then the false belief “There’s a little man who will give you a tickle stick if you walk carefully forward along the white line, jumping in the air after 15 seconds…” combined with that powerful desire to be tickled will send you and your team mates straight over the cliff. That is not adaptive behaviour.
It is not yet clear that there is any set of desires that, when combined with an unreliable perceptual mechanism of type (1a), will generally produce sophisticated patterns of adaptive behaviour of the kind we actually exhibit.
Let’s now turn to perceptual mechanisms of the second sort:
(2) Unreliable mechanisms that produce a significant proportion of, but not mostly, false beliefs
Such, as it were, hit and miss (as opposed to consistently miss) mechanisms may also be of two kinds: those that producing false beliefs in a random way – there being no pattern to the errors, and those in which the errors are, in certain respects, systematic.
We have already seen in the case of mechanisms of type (1b) that a mechanism producing erroneous beliefs in a random way is not a mechanism that evolution might pair off with a particular set of desires such that adaptive behaviour will result.
But what of a hit and miss mechanism in which there is a pattern to the misses? An example would be a mechanism that was reliable with respect to the shape of objects but systematically unreliable with respect to position. Equipped with such a mechanism, a creature might believe, correctly, that there is a square object in its vicinity, but it will be mistaken about where that object is located. Could evolution pair off such an unreliable mechanism with a set of desires that will make the false beliefs produced result in generally adaptive behaviour?
Are there potentially many such perceptual or quasi-perceptual mechanisms that while systematically producing many false beliefs, will still result in generally adaptive behaviour given evolution pairs the mechanism with an appropriate set of desires? And, if so, is there a significant probability, on N&E&C, that we have evolved such an unreliable mechanism, rather than a reliable mechanism?
Here is a sketch of two reasons why we may not be able to answer yes to both these questions.
First, we have seen that it is difficult to envisage type (1a) mechanisms that, given N&E, will result not just generally adaptive bits of behaviour, but sophisticated sequences of team activity of the sort required to retrieve the food from that snake-inhabited narrow ledge. I cannot see that it is significantly easier to envisage a type (2) mechanisms of that sort. Try, for example, to envisage a type (2) mechanism producing mostly correct beliefs about the shape of objects but systematically incorrect beliefs about their location that will result in such successful sequences of team activity – I have tried, and failed. If someone claims there are many such potential mechanisms, the onus is on them to provide a list of examples to illustrate the point. I am unable.
Second, even if there are many such potential mechanisms, is there a significant probability, given N&E&C, that we have evolved such a mechanism rather than a reliable mechanism? Possibly not. Consider, again, a mechanism that is reliable about the shape of objects but systematically unreliable about their position. The most obvious way such a mechanism might evolve is in two stages: first evolving a mechanism that is reliable about both the shape and position of objects, and then engineering a mechanism that systematically reassigns positions to objects, but in such a way that, given the desires with which the species is also equipped, still results in adaptive behaviour. But why would that second level of engineering evolve given the reliable first level is already producing adaptive behaviour? What would be the pay off, for evolution, of now adding a sophisticated location-reassignment mechanism and changing the desires so that adaptive behaviour still results? If there is unlikely to be such a pay off, it is unlikely such a systematic-error-producing mechanism would evolve. Evolution will stick with the reliable mechanism.
In fact, even if N&E&C had equipped us with an unreliable perceptual faculty or faculties of type (2), it would still not follow that many of our beliefs are false. We have seen reasons to suppose that N&E&C will favour reliable as opposed to unreliable faculties of memory and procedural reasoning. If a species also possesses perceptual faculties that are partly reliable and partly, but systematically, unreliable, there arises the possibility that members of this species will be able to figure out that they are, to some extent, being systematically misled by those faculties. In which case, they may adjust their beliefs accordingly. Their beliefs would now reliably reflect reality, despite the fact that they possessed unreliable perceptual faculties. If R is the reliability of our cognitive faculties acting in tandem, the probability of R might still be high, even if it was more probable than not that we possess unreliable perceptual faculties of type (2).
Regarding his hypothetical creatures evolving on another planet, Plantinga claims that, for any given piece of adaptive behaviour,
there are many belief-desire combinations that will lead to the adaptive action; in many of these combinations, the beliefs are false. Without further knowledge of these creatures, therefore, we could hardly estimate the probability of R on N&E and this final possibility [C] as high.
The word “therefore” in this passage is not justified. While it is true that there are many belief-cum-desire combinations that will lead to any given piece of adaptive action, it does not follow that we cannot reasonably estimate the probability of R/N&E&C to be high. This is because, when we turn from beliefs to belief-producing cognitive mechanisms of the sort with which we are actually equipped (e.g. reason, perception, memory), it is no longer clear that there are many (indeed, any) unreliable versions of such mechanisms that, by virtue of unguided evolution pairing them with certain hard-wired desires, will nevertheless result in sophisticated patterns of adaptive behaviour of the sort we actually exhibit.
Further, we have seen that, even if there are many such unreliable but adaptive versions of our cognitive faculties, (i) it doesn’t follow, on N&E&C, that such an unreliable version is at least as likely to have evolved as a reliable version (for it may be that, for the most likely candidates, the evolutionary route to the unreliable version is likely to be via the reliable version), and (ii) even if it is probable, on N&E&C, that we have evolved an unreliable version of such a mechanism (perception, say), it still does not follow that our cognitive mechanisms, acting in tandem, are unlikely to be reliable (for the errors introduced by the most likely candidate unreliable mechanisms may still be correctable).
So Plantinga’s belief-cum-desire argument fails. Indeed, I have sketched out some reasons for thinking that while, for any given piece of adaptive action, there are many belief-cum-desire combinations that will produce it, the probability, on N&E&C, that our cognitive faculties, operating in tandem, are reliable, is pretty high (though I certainly do not claim to have established that here).