This part of the book is where Carroll puts forward his epistemology, which he calls "poetic naturalism." He does a great job of articulating his case, and is very thorough and covers many topics while not sacrificing depth. In fact, there were many times where I thought to myself "but what about X?" and, sure enough, that would be the very next topic Carroll addressed. Also, Carroll's first chapter in this section is a great primer on Bayesian Reasoning for laymen -- he explains it beautifully with simple and common-sense examples.
Contrary to what many might think, I actually agreed with Carroll on many aspects of his epistemology -- he does make some correct predications. But obviously I do not agree with it all, since it is, after all, a form of naturalism.
That being said, let us dive into the material and see what Carroll has to say.
Bayesian reasoning, based on Bayes' Theorem, is foundational to abductive reasoning. It gives you, in principle, a very logical and matter-of- fact way of forming beliefs and updating them in light of new observations and information. It is a way of making inferences to the best explanation -- the very type of inferences science is interested in.
I have no qualms about Bayesian reasoning itself, only about when it can, and cannot, be applied. Bayesian reasoning is useful for (and really is) abductive reasoning, however, it is irrelevant to deductive reasoning. For if a deductive argument is sound, then the truth of the conclusion is guaranteed, whereas nothing guarantees an inference predicated on abductive reasoning. On the contrary, abductive reasoning deals with probabilities, and this should make sense since Bayes' Theorem is itself a theorem dealing in probability.
Why is it important to contrast Bayesian inference with deduction here? Well, because Carroll seems to pay little attention to deduction and focuses mainly on abduction (for reasons we'll survey below). This isn't a problem in itself, but it becomes one when one realizes that many arguments that attempt to demonstrate the falsity of naturalism are deductive in nature.
In any event, Carroll also claims the following:
Each of us comes equipped with a rich variety of beliefs, for or against all sorts of proportions. Bayes teaches us (1) never to assign perfect certainty to any such belief; (2) always be prepared to update our credences when new evidence comes along; and (3) how exactly such evidence alters the credences we assign. It's a road map for coming closer and closer to the truth. (p. 82-83)
On first glance this seems like a great point, and I would, in most cases agree. Bayesian reasoning is indeed a fantastic "road map for coming closer and closer to truth." However, contrary to Carroll's claims here, there are beliefs that are (to borrow a phrase from the presuppositionalist camp) preconditions of intelligibility. That is to say, there are beliefs and propositions that are necessary conditions for any sort of intelligible epistemology, and because of this it follows that such beliefs must be certain and cannot be "updated" or revised pending any new evidence (again, see below). One such proposition is the law of noncontradiction. This is a belief that we can be absolutely certain about, especially since any denial of it, even in principle, requires that it be true.
I briefly expounded PN in the last post, yet there wasn't much space devoted to it. That's because the current section is where Carroll largely unpacks this epistemology.
[O]ur best approach to describing the universe is not a single, unified story but an interconnected series of models appropriate at different levels. Each model has a domain in which it is applicable, and the ideas that appear as essential parts of each story have every right to be thought of as "real." Our task is to assemble an interlocking set of descriptions, based on some fundamental ideas, that fit together to form a stable planet of belief. (p. 4)
This brings us to the "poetic" part of poetic naturalism. While there is one world, there are many ways of talking about it. We refer to those ways as "models" or "theories" or "stories"; it doesn't matter. (p. 94)
[S]omething is "real" if it plays an essential role in some particular story of reality that, as far as we can tell, provides an accurate description of the world within its domain of applicability. (p. 111)Again, on first glance there isn't much to quibble with here. It is true that certain models and theories regarding parts of reality, with their respective vocabularies, only need to be utilized in their respective domains of inquiry. Talking about free will and volition on the macroscopic scale of human experience utilizes a different model, and different vocabulary, than the models utilized when inquiring into the nature of atoms, electrons, and other subatomic particles.
However, the pertinent questions that need to be posed regarding this epistemology are the following: (1) Carroll places a lot of weight on models being "useful," and this utility being the litmus test for labeling something as "real," but what determines whether a specific model is useful, and why is utility important for our epistemology in the first place? Also, why should the fact that it is useful to describe reality using different models at different levels dictate what the actual nature of reality is, in principle? (2) Are there other ways of talking about or modeling the world, besides PN?
Note that Carroll attempts to answer (almost) all of these questions. So, it's not as if he hasn't thought this through, and he probably wouldn't at all be fazed by any of them. However, that doesn't mean that his answers are convincing.
(1) is a question, or set of questions, that Carroll attempts to answer:
The question, however, is whether a particular way of talking about the world is useful. And usefulness is always relative to some purpose [...] "useful" means "providing an accurate model of some aspect of reality." (p. 143)It is important that our models are accurate representations of some aspect of reality, however, this is not the traditional meaning of usefulness. Something is not useful if it accurately reflects a part of reality, rather something is useful if it can serve a practical purpose. Now, it is true that models that accurately reflect reality tend to be more useful than models that do not, but this doesn't mean that the latter is defined in terms of the former. This might seem like a minor quibble, but one problem with Carroll's PN is that throughout the section he constantly equivocates between both of these definitions. Sometimes "useful" is used as "serving a practical purpose" and other times it's used as "provides an accurate model of a part of reality."
What's the problem though (as if equivocation is not bad enough)? The problem is that the two definitions utilized by Carroll are sometimes mutually exclusive -- that is, there are times when a "model" or "story" can be useful even though it fails to accurately reflect reality. Carroll actually demonstrates such with his own example: he highlights the phenomena of transgender individuals -- people whose gender identity is different than the gender they were assigned at birth. Carroll says that if someone born male identifies as a female, then if this identification is "useful and meaningful" (p. 142) then why not identify them as a female? The problem here is that no matter how "useful" it is to identify someone as a female, this doesn't entail the fact that they are, in reality, a female.
Carroll doesn't see this as problem since he believes that classifications like male, female, and even human, are "human inventions." (p. 142) He claims these categorizations are not illusions because they are useful and meaningful in the contexts that we apply them. But again, something being useful does not necessitate that it accurately reflects an aspect of reality, and therefore it is possible that something be useful, yet be illusory.
And even worse for Carroll, the claim that a classification like "human" is a human invention is actually self-defeating. For to say that something is a cognitive construct of the human mind, one needs a nature of "humanity" to objectively obtain. That is to say, for one to assert that something is a human invention necessitates that "human" be an objective referent, and not just a construct. So the proposition that the concept human is a human invention literally contradicts itself.
While Carroll doesn't comment on question (2) directly, he does make the following comment:
Poetic naturalism is at least consistent with its own standards: it tries to provide the most useful way of talking about the world we have. (p. 112)Carroll is correct that PN is prima facie consistent with itself, but what about when we delve a bit deeper and consider the implications of question (2)? That is, is PN the only, or even the most useful, model, or story, of reality? Well, surely it's not the only model of reality: there are a multitude of different ontologies. Ok. So then how do we determine which model is most useful? Well again, it depends on how Carroll is using the word useful.
If by "useful" Carroll means "serves a practical purpose," then it's not clear that PN is the most useful way of talking about the world. Moreover, how, on PN, do we even determine the most useful, or practical, way of talking about reality? It seems that we would need some idea regarding the fundamental nature of reality to be able to say model X is more useful than model Y. But then, is this "idea" about the fundamental nature of reality itself merely another "story" or "way of talking about the world" on some narrow domain? If so, then we're right back to the problem at hand wherein we need to determine how we know that this "model" or idea is "useful." What this means is that you can't have an ontology founded on nothing but "stories" and "ways of talking about the world." At some point you need a concrete ontology that describes how nature objectively operates, and this entails that you can't have an ontology that is purely "poetic", as Carroll uses the term here.
Moreover, if by "useful" Carroll instead means "providing an accurate model of some aspect of reality," then we run into the same problems. For, again, in order to know that a model is an accurate model of reality, we need to know something objective about the behavior and nature of reality that is not just another "way of talking about it" -- otherwise we go round and round in a circle again.
Carroll then promulgates an emergentist ontology to make sense of different models of the world at different levels:
One pivotal word enables [the] reconciliation between all the different stories: emergence [...] A property of a system is "emergent" if it is not part of a detailed "fundamental" description of the system, but it becomes useful or even inevitable when we look at the system more broadly. (p. 94)I do agree with Carroll that there are "emergent" phenomena. There are many such examples one could give of such phenomena, and Carroll gives a good explication using his own examples. But how does emergentism help PN? Carroll explains:
[E]mergence is about different theories speaking different languages, but offering compatible descriptions of the same underlying phenomena in their respective domain of applicability. (p. 100)So emergence is a way to continue talking about the world at different levels, wherein the higher levels usually "emerge" from the lower levels. This is all well and good, and so far I have no bone to pick with Carroll here, in principle. Where I might begin to disagree is if Carroll, later on in the book, tries to explain away something like consciousness as merely a "model" or "way of talking about" human experience that only applies at higher levels. (I haven't read the whole book yet, so if he does argue this then we can cross that bridge when we get there. )
Rationalism vs. empiricism
Since Carroll's epistemology is heavy on science and Bayesian reasoning, it should come as no surprise that he advocates a form of empiricism. That is, he believes that we can only know things about reality by looking at it and observing it. He contrasts this with rationalism wherein it is believed that knowledge comes intuitively without any need for empirical observation.
Carroll surveys the case usually given in justification of rationalism wherein the proponents appeal to the domains of mathematics and logic. Carroll argues against such an appeal by claiming that in both math and logic we simply derive consequences from differing axioms. This is, in a sense, correct, at least for math. For example, Euclidean consequences follow from Euclidean axioms, and non-Euclidean consequences follow from non-Euclidean axioms. But we need to actually look at the world to be able to determine whether it can be described by Euclidean or non-Euclidean geometry. No problem there.
However, there are those, like myself, who claim that there are axioms and necessary truths that must hold in any possible world at all. One example of this, as was already mentioned above, is the law of noncontradiction. This is not an axiom that might only hold in certain worlds, but might possibly fail to hold in others -- as in the case of Euclidean geometry. Rather, it is a necessary feature of having any kind of a reality at all. There is no axiom of non-contradiction and axiom of non-noncontradiction; there is only the former.
There are other propositions that are also claimed to be axiomatic and necessary -- e.g. law of identity, PSR, principle of causality etc. -- but I need not defend them presently. The point is that there are indeed things that we can know a priori, that is, without having to actually look at the world to justify them.
Carroll considers this point (briefly), and claims the following:
If we were thinking deductively, like a mathematician or logician, we would say that no collection of particular facts suffices to derive a general principle, since the very next fact might contradict the principle. (p. 135)This seems to be quite absurd. For according to Carroll's reasoning, for instance, we cannot say the law of identity is an immutable principle, because we might one day come across something which is not identical to itself. This seems silly, and it should, because we already know intuitively that identity is a precondition of intelligibility of any thing that will ever exist. And so it goes with all other necessary truths. Now, there may not be many necessary truths with which we can know a priori, but there are a few, and they are a very crucial aspect of one's metaphysic.
All of this is important because it goes against Carroll's PN, as well as his form of empiricism. For if there are propositions that are necessary truths, then these truths aren't merely "models", "stories", or "ways of talking about reality." Rather, they are immutable principles that hold throughout any domain and any level, and cannot be shoved aside by the thought that one day we could discover them false. This is like saying that one day we might find someone who is successfully married and a bachelor.