Dennett, Davidson and the Reality of Belief

Richard Rorty likes to say that one of the most important philosophical lessons he has learned from Dewey is that if you take care of freedom, truth takes care of itself. That is, Rorty credits Dewey for offering the therapeutic recommendation that we stop thinking of a true belief as something that corresponds with something “out there” and begin thinking of it as the outcome of an open conversation. Rorty puts the point this way: “Instead of saying that free and open encounters track truth by permitting a mythical faculty called ‘reason’ to function unfettered, he says simply that we have no better criterion of truth than that it is what results from such en-counters.”⁹

While some critics argue that this attitude toward the epistemic status of truth amounts to negligence, he describes it as a healthy dose of philosophical superficiality—a form of deliberate light-mindedness which rightly focuses our attention on the priority of democracy over philosophy. This “air of light-minded aestheticism,” heinsists, “helps make the world's inhabitants more pragmatic, more tolerant, more liberal, more receptive to the appeal of instrumental rationality.”¹⁰ In other words, there are times when we should be willing to demote specifically philosophical concerns in favor of something that makes a measurable, empirical difference in the world.

This willingness to substitute a set of “merely philosophical” concerns with some that produce a practical difference is a family trait among pragmatists. For example, a simple way of introducing Daniel Dennett's own form of philosophical superficiality is to modify Rorty's motto just a bit and say that for him, if you attend to the relevant science, most philosophical worries take care of themselves. In his judgment, the philosophical interest in metaphysics, ontology and truth are actually “the last questions we need to answer—and their answers will not be very interesting or useful once we've got the science in place.”¹¹

Dennett's pragmatic irritation with traditional philosophic commerce is that it is often superfluous to the business of science. The basic reason goes something like this: even if we could explain why modern particle physics works, this would not enhance the predictive success of these conceptual instruments. Therefore, any epistemological worry about whether particle physics explains reality by corresponding with it or whether it is a tool for coping with reality is a classic case of a merely philosophical question whose answer fails to make a salient empirical difference. After all, the ability to effectively use a crowbar around the house doesn't require an explanatory theory about why crowbars work.

Guided by the pragmatist's touchstone of searching for “a difference in the world that makes a subsequent difference testable by standard empirical methods,” Dennett has constructed an enviable career by defending the provisional use of intentional idioms within the standard scientific scheme.¹²

His justification for strategically invoking folk psychology's vocabulary of beliefs and desires is that the practice of ascribing propositional attitudes to a human or natural artifact is a powerful tool for predicting, manipulating and explaining its behavior.¹³ Indeed, if there are no testable differences between the claim that X believes Y and that the behavior of X can be predicted by ascribing to it the belief Y, they practically mean the same thing. The more-or-less instrumentalist use of the intentional idiom is justified, Dennett thinks, because it quickly and reliably produces what engineers call a “spec sheet,” a sketch of the aptitudes that a physical system's design must accomplish. As he summarizes this point, once the intentional stance has generated a system's profile, it “then becomes an engineering task to design such a system, typically breaking it up into organized teams of subagents, smaller, more stupid homunculi, until finally all the homunculi have been discharged—replaced by machines.”¹⁴

On this explanatory model, the mental isn't really “reduced” to the physical because the intentional vocabulary is merely an abstract tool for simplifying our research. Drawing on Hans Reichenbach's distinction between illata and abstracta, Dennett proposes that beliefs “are best viewed as abstracta—more like centers of gravity or vectors than as individualizable concrete states of a mechanism.”¹⁵

People thus have beliefs the same way the earth has an equator—neither are really parts of the world's furniture. So, instead of outright inter-theoretical reduction, the assignment of establishing an intentional system's profile is replaced by the engineering assignment of designing a physical system complex enough to warrant the provisional use of intentional idioms. Without putting too fine a point on things, the folk psychological entities of beliefs and desires that we ascribe to an object are only temporary instruments of inquiry, well-placed rest stops on the way to an utterly mechanistic and scientific explanation of a complex system's capacities.¹⁶ The as-if convention of the intentional stance is important, then, because it paves the way for a vigorous cognitive science.

In contrast to Dennett's scientific aspirations and confidence, however, the late Donald Davidson insisted that psychology could never “be made into part of physics or any other ‘natural’ science.”¹⁷

Why was Davidson a pessimist about the prospects of cognitive science where Dennett is an optimist? Part of the answer is that Davidson was convinced that cognition is normative in ways that cannot be reduced, circumvented, or replaced. In other words, the propositional contentfulness of human thought implicitly establishes norms that allow us to make evaluative judgments about what one should say or believe. In any given context, for example, a particular concept such as “mammal” can be used rightly or wrongly to categorize some element of the world. Additionally, the purely logical relations that exist between concepts—such as dog and mammal—establish norms that one ought rationally to observe (e.g., if one believes that an animal is a dog one should also believe that the same animal is a mammal). As he put it: According to Davidson, no matter how many times we break up an intelligent agent into organized teams of increasingly stupid homunculi there is a normative element that can never be “replaced by machines.” At the end of the day, the cognitive scientific enterprise will inevitably stub its toes on the problem of deontic ¹⁹commitments and entitlements.

There was, however, another point of contention between the two; one that can be traced back to their philosophically divergent readings of Quine's principle of indeterminacy of translation. ²⁰

According to Dennett, Quine's principle suggests two conclusions when it comes to the status of belief. First, two substantially different interpretive schemes could in principle yield equally predictive profiles of a complex system— despite attributing different beliefs to it. Second, in such a case, there would be no “deeper fact of the matter” to establish which scheme accurately describes the agent's real beliefs. As he puts it, these “rival theorists would not even agree on which parts of the world were pattern and which were noise, and yet nothing deeper would settle the issue.”²¹ Even though the possibility of two radically different but equally predictive interpretive stances is negligible in practice, Dennett argues, the principle and its far-reaching consequences survive.²²

Davidson, however, thought that Dennett made far too much of Quine's insight because the possibility of rival interpretive schemes in themselves do nothing to undermine the reality of beliefs. More to the point, there is no compelling reason to think that the interpretive pluralism that Quine anticipated describes anything more than local options for capturing the same basic reality. The fact that Americans call my golden retriever a dog while the French call her une chienne and the Germans call her eine Hunde does not (thankfully) put her very existence in jeopardy. As Davidson wrote: “We know there is no contradiction between saying that the temperature of the air is 32° Fahrenheit and saying that it is 0° Celsius; there is nothing in this ‘relativism’ to show that the properties being measured aren't real.” Thus, he concluded, when we apply Quine's principle of indeterminacy of translation to the status of intentional states, it “does not challenge the reality of the attitudes or meanings variously reported.”²³

In pedestrian terms, then, Davidson was more of a realist about belief than Dennett.²⁴

I've briefly reviewed this dispute over the status of belief because it nicely illustrates how sometimes a philosophical interest in epistemology or ontology isn't “merely philosophical.” That is to say, it has some impact on what we should do. In the case of Davidson and Dennett, what may sound like a relatively insignificant disagreement over the meaning of Quinean indeterminacy of translation ultimately results in rival judgments regarding the empirical prospects for a vital science of the mind. If one accepts Davidson's pessimistic evaluation of cognitive science, for example, the hours and effort required to master the vocabulary of cognitive science will look like an unwise or even pointless investment because the project can't make good on its claim to “mechanize” thought. Thus, because it leads academics to forge distinct intellectual trajectories by reading different books, writing different papers, and asking different questions, our philosophical reflections on the motivating assumptions of cognitive science appear to make a Jamesian practical difference.

Realism, Instrumentalism and Evolutionary Complexity

The disagreement between Davidson and Dennett over belief and the prospects of success for the cognitive sciences is, in some sense, a familiar question about the epistemic results of scientific theorizing. In this case, the question is: Do humans really have beliefs or is it is simply a powerful predictive strategy to treat humans as if they have beliefs? Couched in these terms, the opposition between scientific realism and scientific instrumentalism floats to the surface. For a scientific instrumentalist, although our scientific theories (ideally) allow us to cope with reality, we are never in a position to determine whether they “match up” with the world in such a way that we also arrive at an explanatory account of that reality. Science, from this perspective, is nothing more or less than a particularly powerful set of skills and strategies that equips humans to confront the world in profitable ways. For a scientific realist, however, our best scientific theories are not only empirically adequate solutions for coping with reality; they also provide an objectively true explanation of it. In other words, the realist insists that the best science reveals the world as it really is.

Traditionally, the dispute between instrumentalists and realists over science has been a global affair. That is, the point of contention typically has been whether science in general copes with or accurately explains reality. Yet, given that for most of the twentieth century whenever philosophers spoke of science they really meant physics, much of this conversation has been complicated by the emergence of biology as a paradigm science.²⁵

Nowadays, unless one is prepared to naively defend the logical positivist's doctrine of the unity of science, it is absolutely necessary to identify which scientific enterprise we are discussing. Over the past decade or so, Alexander Rosenberg has been cutting an intriguing local path through these traditional issues by asking incisive epistemological questions about the biological sciences.²⁶

Painted in broad strokes, Rosenberg's contention is that once we have climbed to the level of physical organization at which natural selection is a causal factor, the relevant phenomena have become maddeningly complex. Compared to the small, manageable classes of kinds and processes that physicists and chemists address, biologists are consistently stymied by the complex, disjunctive processes of evolution. For example, there is now a fairly strong consensus that the dream of a smooth inter-theoretic reduction between Mendelian genetics and molecular genetics is impossible. The difficulty is that the various theoretical entities of Mendelian genetics are “multiply realizable” by many different molecular arrangements. In other words, physically distinct molecular structures may generate identical or near-identical functional properties. A relatively straightforward example of physically distinct structures producing functionally identical properties is the amino acid valine, which is coded for by four different codons (GUU, GUC, GUA, GUG). The general lesson we should draw from this is that there are no clean one-to-one relationships between a natural kind in molecular genetics and a natural kind in Mendelian genetics. As a result, the bridge laws necessary for reducing one enterprise to another—that is, the principles that would allow us to correlate an entity in one vocabulary with an entity in another—will be riddled with disjunctive exceptions and ceteris paribus clauses; so riddled, in fact, that these prospective laws “will not pass muster as an interesting generalization or even the first approximation to one.”²⁷

The larger philosophical moral that Rosenberg would have us draw from this recognition is that, since natural selection operates on biological function and is blind to biological structure, there will be no nomological generalizations at the phenotypic level for biology to discover. From his vantage point, the success of the physical sciences is due in part to the fact that they tackle non-evolutionary processes such as thermodynamics and electromagnetism—processes where structural differences with equivalent function are the rare exception. Seen in this light, the scientific success of physics ultimately depends on the physical world's underlying simplicity.²⁸

However, given that natural selection generates an organic world in which functional equivalence through structural diversity is the rule, Rosenberg doesn't think we should expect biology to provide any exceptionless generalizations beyond the micro-level organization of organic molecules. In fact, he argues that since natural selection operates on multiply realizable traits in ever-changing environments, even such crucial concepts as evolutionary fitness must remain primitive, undefined terms.²⁹

One possible strategy for responding to this state of affairs would be to soften the requirements of what a law is, and simply treat any generalization with disjunctive conditionals as a non-strict law. Yet, Rosenberg flatly rejects this as an unworkable solution. The predicament is that while this strategy may solve the problem of what to do with exception-filled generalizations, it simultaneously weakens our ability to generate robust predictions by allowing a potentially endless list of variables to sneak in through the back door. The intimate relationship between limited generalizations and anemic predictive power becomes clearer when we think of what a “non-strict law” such as X behaves like Y, unless A, B, C, D, E and/or F is the case allows one to say about what kind of thing X is. Or, as Rosenberg makes the point: “Not only can these ‘laws’ not be absorbed by reduction to strict laws, but their weakness makes a permanent barrier to the sort of predictive improvement characteristic of scientific knowledge.”³⁰

So, if there are no nomological generalizations to be found above the macro-molecular level, what are we to make of the modern biological sciences? One might argue that if biology is as different from physics as Rosenberg suggests then it isn't a real science after all.³¹ Alternatively, one could embrace a kind of neo-vitalism and insist that the apparent irreducibility of biology demonstrates that life ultimately depends on non-physical forces or processes. Rosenberg wisely refuses both of these options, and by attending to the stochastic character of evolutionary theory develops something relatively novel: a realist argument for the instrumental nature of biology.³²

Ever since Peirce first made the case in “The Fixation of Belief,” philosophers have viewed Darwinian natural selection as a statistical theory of evolutionary change akin to the second law of thermodynamics.³³

When asked why our best theory of the biological world is probabilistic, the most common response has been that this theoretical feature accurately reflects the indeterministic nature of the evolutionary process itself.³⁴ Yet, Rosenberg maintains that this move naively confuses indeterminism as a property of the world (i.e., ontological probabilism) with statistical probability as a property of our theories (i.e., epistemic probabi-lism).³⁵ On the one hand, the world could be fundamentally deterministic and yet our best theories about this world might be unavoidably statistical, e.g., the second law of thermodynamics. In this case, the inability to capture the determined nature of reality would tell us more about ourselves than the world. On the other hand, the natural world could be essentially indeterministic but our best theories could be deterministic, e.g., Newtonian mechanics. In this case, our very best theories would yield the kind of predictive control that is the hallmark of science but nevertheless miss some features of the natural world. As Rosenberg summarizes the point: “Whether the world is indeterministic or not does not settle the matter of whether a true and correct theory about the world must be statistical or not … even if the world is indeterministic, the probabilities in our best theories about the world need not themselves reflect this indeterminism but may have another source.”³⁶ What could be the source of evolutionary theory's stochastic makeup if it isn't the world?

To a first approximation, Rosenberg's answer is that this feature of biological theorizing reveals the epistemic limitations imposed by our evolved cognitive abilities. If we were Laplacean demons, cognitive agents with unlimited computational powers, we would have no need for a statistical evolutionary theory that depends on probabilities and statistical generalizations—we could actually track the physical processes that we now only treat as probabilistic “tendencies.” On this account, evolutionary theory is stochastic because—once we have arrived at the levels of organization where natural selection is a causal factor—we can no longer keep track of all the variables in a way that yields nomological generalizations. Rephrased in a way that emphasizes the kind of epistemic probability that evolutionary theory and the second law of thermodynamics share, the technical lesson goes something like this: Since the state-space of evolutionary change is indefinitely large, the ratio of physical states of increased fitness to the demographically possible physical states cannot be computed (by us) because the total number of points in evolutionary state-space is simply too large. Thus, biology can only be an instrumental science because it is a science for limited cognitive agents like us. The “generalizations we frame and the theories we employ,” he writes, “can reflect only approximations useful for various purposes, models to be applied in some cases, but not in others, heuristic devices whose limits are known to us through experience but which we can correct and improve in generality only at the expense of their usefulness to us.”³⁷

Biological theories are therefore, in Rosenberg's estimation, “irretrievably subjective” inasmuch as they reflect the limitations, needs and interests of finite cognitive agents theorizing about—and interacting with—the natural world. If he is correct about this, while we may be realists about the theoretical achievements of physics and chemistry, we really have no choice but to be instrumentalists about biology.

Instrumental Biology and the Prospects for a Robust Cognitive Science of Religion

What does the case for biology as an unavoidably instrumentalist science have to do with the academic study of religion? More pointedly, what possible difference does the stochastic character of modern evolutionary theory make for the academic study of religion? If Rosenberg is correct, quite a lot.

According to Rosenberg, psychology is an inescapably instrumental science for one simple reason. Inasmuch as it studies how complex human cognitive systems process information, it must be viewed as a branch of biology. After all, evolution is responsible for creating these systems in the first place. However, he predicts that in general psychology will fail to achieve the instrumental and predictive success of, say, population biology because the various phenomena of human intentionality cannot be systemically reduced to the physiology of the brain. One reason for this irreducibility is that we run into another version of the problem of squaring Mendelian and molecular genetics. As Hilary Putnam first argued in the 1960s, the “mental” entities of the cognitive vocabulary are multiply realizable at the neuroscientific level by many different brain states. In other words, physically distinct brain states may generate identical or near-identical cognitive properties. This suggests to Rosenberg that, as in the case of genetics, the goal of finding bridge laws that will allow us to reduce one enterprise (psychology) to another (neuroscience), and thereby articulate nomological generalizations about human cognition, is a fantasy. The “prospects for uncovering general theories about behavior,” he writes, “will be at least as remote as biology's prospects for uncovering general theories about the physiology upon which behavior supervenes.”³⁸

Worries about “reduction” and multiple realizability are fairly standard topics in the philosophy of cognitive science. Despite the philosophical intuition that, taken together, these two philosophical dilemmas would spell the end of the cognitive agenda, it turns out that they can be dealt with in a variety of ways.³⁹

As a case in point, if one is prepared to adopt Dennett's urbane instrumentalism, many of these concerns can be pushed quietly off stage. What makes Rosenberg an interesting conversation partner in this context is that he helpfully brings into focus issues about the cognitive program that haven't drawn as much attention.

In a recent essay, Pascal Boyer argues that the cognitive science of religion depends on a probabilistic model of religious thought. For example, he notes that: “The fact that a given religious concept is easily acquired and recalled, salient and inferentially rich, explains why, all else being equal, there is some likelihood that such a concept will be imagined and, once imagined, successfully transmitted in a particular cultural milieu” (emphasis mine).⁴⁰

In other words, the theoretical scope of the cognitive platform thus far has been to describe in broadly probabilistic terms the mental states and processes found in most people, most of the time, in most religious contexts. Perhaps the best example of the cognitive bid to articulate a generally reliable prediction would be the so-called hypothesis of cognitive optimality, which contends that the eye-catching qualities of counter-intuitive religious concepts (e.g., agents with no bodies, etc.) are balanced out and constrained by the inferentially rich potential of representations that preserve our intuitions (e.g., agents have thoughts and desires). Boyer and others believe that the basic profile of human cognition makes the window of culturally successful religious concepts relatively narrow, and that as a result the constellation of beliefs and practices that occupy the cognitively optimum position will (more or less) exhaust the range of ethnographically relevant religious phenomena.⁴¹ Yet, as Harvey Whitehouse's work on the bifurcating “modes of religiosity” nicely demonstrates, there is nothing hard and fast about this cognitive claim. Whenever we scratch the surface of the anthropological record and look a little more closely, the religious systems that we discover are rarely the minimalist affairs of modestly counter-intuitive, easily acquired concepts that the cognitive optimum hypothesis forecasts. In fact, as Whitehouse observes, there even appears to be a general “tendency for religious traditions to eschew certain cognitively optimal concepts and to peddle more complex bodies of knowledge in their stead.”⁴²

Rosenberg would have us pay particularly close attention to the role of probabilistic language in articulating and defending the hypothesis of cognitive optimality. Even if all else is equal there is still only some likelihood that a given religious concept will exist and be transmitted. Given the relationship between limited generalizations and anemic predictive power that Rosenberg identifies, it is nearly impossible to miss just how statistically tentative Boyer's model appears to be. In a world where ceteris are hardly ever paribus, this could mean serious trouble for the enterprise. If this is the best that the cognitive science of religion has to offer, the scientific aspiration to explain religion may stumble over the inability to produce interesting generalizations. The implication of this for the budding cognitive science of religion is as obvious as it is stark. Given the persistent lack of universality, the necessity of ceteris paribus clauses, and the long list of exceptions and counterexamples that often obtain in psychological models, the cognitive scientific aspirations to explain religion—full stop—could be epistemically quixotic. On its best day, it would be a vital instrumental science that mirrors biology by offering us a powerful set of statistically reliable heuristic devices for coping with other cognitive agents; unfortunately, these devices would yield surprisingly limited generalizations and meager predictive power. However, Rosenberg believes that on most days this sort of project can't help but be far less robust than its strictly biological cousin. The fundamental reason for this is that at this level of inquiry we are dealing with phenomena that are more complicated by several orders of magnitude. On this scorecard, the cognitive science of religion would be an unavoidably instrumental science: useful for some purposes and not others, a set of theoretical models that are effectively applied in some cases, but not all.

Here's the issue. With physics, although textbook generalizations of basic mechanics remain silent about complicating forces or processes, it is nevertheless possible to specify the smallnumber offorces that “interfere to prevent other things from being equal, and thus excuse the predictive failures and explanatory imprecision” that textbook generalizations tend to produce. However, the ability to cash out ceteris paribus clauses in the biological sciences is hamstrung because, rather than limiting interfering forces, the operation of natural selection tends to amplify such complicating variables. As Rosenberg observes:

If we import this observation to the case at hand, we soon find ourselves asking: Is it possible for cognitive theorists of religion to specify, in sufficiently precise terms, a finite set of forces that prevents all else from being equal? That is, can one precisely identify the social, cognitive or neurological forces that complicate religious cognition in ways that mean there is still only somelikelihood that a given concept will exist and be transmitted in a given cultural context? This is a strictly empirical question, and as such it remains to be seen whether this epistemic responsibility can be discharged. Nevertheless, we should take note of the tentativeness of Boyer's probabilistic model of religious cognition.

Even if cognitive theorists were able to replace ceteris paribus clauses with specific variables that prevent other things from being equal, it appears as though our best (current) cognitive theory tells us that there is still only some likelihood that a given concept will emerge when all things are equal. If this remains the case, not only would the empirical generalizations of the cognitive science of religion not approximate the timeless truths of scientific laws, but whatever empirical generalizations we do offer would be so conditioned by qualifications, exceptions and disjuncts that—by the time we're through—these generalizations may not be very general at all. The flip side of this coin would be that as we pull away from the catalog of local qualifications and exceptions, the generalizations we're left with may lack any bite. This outcome would confirm Clifford Geertz's theoretical intuitions about what this kind of “big-picture” anthropological program can accomplish. As he sees it, the problem with these sorts of ambitious research programs isn't that there are no universals or broadly applicable generalizations regarding human culture. Rather, the problem is that these features—if empirical research demonstrates that they actually exist—aren't very interesting, important or illuminating once the sun has set. All purported anthropological universals, he argues, “are so general as to be without intellectual force or interest, are large banalities lacking either circumstantiality or surprise, precision or revelation, and thus are of precious little use.”⁴⁴

Examples of boring and pointless universals include People everywhere have views about the differences between the sexes and All societies have systems of social ranking. Will the cognitive program leave us only with the soggy empirical generalization People everywhere believe in non-obvious, extra-human beings because they have human brains? Only time will tell. In the meantime, two additional observations are also worth making.

First, going back to Dan Sperber's influential work on the cultural epidemiology of representations, cognitive theorists often employ a model of inter-generational transmission that is patterned on the mechanism of natural selection.⁴⁵

Accordingto Sperber's model, the inter-generational cultural reproduction of representations is unlike the conservative process of biological replication because it introduces astounding degrees of novelty every step of the way. Human communication is a noisy, low-fidelity medium, and one consequence of this is that every act of cultural transmission threatens the stability and coherence of the very information that is being reproduced. From this perspective, when it comes to human culture the explanatory challenge is providing a causal account of why some representations become widespread and long-lasting while others are nothing more than epiphenomenal blips on the cultural radar screen. To a first approximation, the proposed answer to this puzzle is that some representations interact with our cognitive abilities and dispositions in ways that make them more resistant to decomposition through transmission than others. For example, in his work on the “doctrinal” and “imagistic” modes of religiosity, Whitehouse adopts a Sperber-inspired “selectionist” stance towards cultural representations and treats the two modalities of religion as stable strategies for the inter- and intra-generational transmission of religious concepts, norms, and practices. The basic intuition is that the long-term stability of these two strategies—specifically, their ability to resist the continuous threat of distortion and erosion through transmission—ultimately rests on the ways in which they exploit the biases of human episodic and semantic memory. The working hypothesis is that those features of religious traditions that activate either episodic or semantic memory will be positively selected for because of the way they fit the native computational profile of human cognition.

The invocation of natural selection as a causal mechanism in cultural transmission raises an obvious question, however. If it is the case that there are always more religious cultural representations that can possibly make it into the next generation, does anything like an “arms race” ever emerge between religious concepts in a given setting? That is, does one local variation in a religious concept enhance its cultural and cognitive “fitness” in such a way that it creates a design problem for all other religious concepts? If this is so, and if the theoretical “space” of adaptive “moves” and “counter-moves” for cultural representations is very large, it could be the case that the ceteris paribus clauses of a cognitive theory of religion could remain forever open-ended. This possibility explains the strategic importance of the cognitive scientific claim that the computational profile of the human mind imposes surprisingly strict constraints on the kinds of religious concepts that are possible.⁴⁶

Without these constraints, the enterprise would be left to wrestle with the ad infinitum of variations and responses that natural selection tends to generate.

Second, Rosenberg's case for instrumental biology on the basis of evolutionary theory's stochastic character prompts us to ask about the source of the cognitive program's probabilism. Are the social, cognitive or neurological processes that generate religious thought fundamentally indeterministic? Or, is the cognitive theory of religion stochastic because of our own cognitive and computational limitations? More to the point, is the cognitive theory of religion statistical for ontological or epistemic reasons? For the sake of argument, let's grant that the processes of human cognition which are relevant for explaining the cognitive features and psychological origins of religious systems are in fact indeterministic. Unfortunately, if this is the case, it doesn't solve the broader philosophical problem. As Rosenberg points out, even if the processes that produce religious cognition turn out to indeterministic to one degree or another—another empirical question that remains to be answered—it could still be the case that our computational limitations are the most significant source for the statistical nature of cognitive theorizing. Here, then, is another case where the answer to a philosophical question makes a real-world difference. If it turns out that the cognitive science of religion is probabilistic for epistemic rea-sons—because it makes otherwise intractable problems manageable for us —this would impose clear limits on what we can expect the cognitive project to achieve in the years to come.

If we are epistemic probabilists about religious cognition, it may be the case that a comprehensive, genuinely explanatory theory of religion is simply out of our reach. Mind you, if this is so, it would not because such a theory is prima facie impossible, or because religion represents a sui generis fact about the world, or because religion is an irreducibly “spiritual” phenomenon. In fact, a philosophical naturalist should reject these conclusions out of hand. Rather, an explanatory theory of religion would be beyond our grasp because the disjunctive empirical complexities that such a theory must track—be they social, cognitive, neurological or all of the above—would far exceed our restricted computational powers. A Laplacean theorist of religion with unlimited cognitive powers could simply dispose of these stochastic measures and actually track the fundamental processes that we now can only treat as probabilistic “tendencies.” Unfortunately, on Rosenberg's account, we can't move the ball very far down field without these methodological tools. Ironically, on this scorecard the cognitive theory of religion tells us something real and important about the human mind—just not what it originally set out to discuss.

If the connection between theoretical probablism and theoretical instrumentalism that I've outlined in this essay has any merit, the cognitive research program—de-spite its laudable ambitions and theoretical rigor—could be destined to the same fate as Lévi-Strauss' bold attempt to construct a science of culture and cognition. In other words, I think the critics of the cognitive program, its advocates and those who—like myself—are broadly sympathetic to the cause for fundamentally Darwinian reasons have to ask: Are we dealing with an abstract tool box that does not cut nature at its joints, but is nevertheless useful to us insofar as it supplies what any pragmatically justified vocabulary in the human and social sciences supplies? If so, rather than explaining religion, a successful cognitive science of religion would ultimately yield what any instrumental science provides: heuristically valuable analytic tools, useful rule-of-thumb generalizations, intriguing laundry lists of exceptions and counterexamples, profoundly stochastic models, and lots of local details about what people say and do. As a case in point, it strikes me that at the very least the cognitive program's account of the god-concept offers anthropologists and historians of religion a particularly elegant heuristic tool for their work. From the cognitive scientific standpoint, gods and spirits are noteworthy because representations of these beings both preserve some of our assumptions about personhood (e.g., they want or desire certain things from us) and violate others (e.g., they don't die). For anyone interested in the cross-cultural comparison of religious traditions, this analytic lens seems to provide a wonderfully cogent and economical grid for charting and organizing the various ways in which gods and spirits ignore many of our basic biological and physiological assumptions about what it means to be a person. If Jonathan Z. Smith is right, and the necessary prerequisite for any viable “science of religion” is a workable system of classification, the sort of taxonomy of gods and spirits that this interpretive strategy provides could mean that we are well on our way.⁴⁷

Yet, given that if all else is equal there is only some likelihood that the god-concept will be imagined and transmitted in a particular setting or in general, what this strategy can't provide—and what Rosenberg's reflections on instrumental biology force us to look for—are nomological generalizations about this particular corner of human activity.

At the end of the day, then, Rosenberg's work in the philosophy of biology forces us all to confront a stark possibility about the ambitious research agenda of the cognitive science of religion. If we can't be realists about its findings, at some point we may have no other choice but to become realistic about its prospects.