sign in sign up
<<
Home , Cognitive neuroscience, Neuroscience

Is cognitive an oxymoron? 1 Forthcoming: Current Controversies in Philosophy of Cognitive , Eds. S. Cullen, S.J. Leslie & A. Lerner

Is an oxymoron?

Fiery Cushman

1. Introduction have a clear image in of just what “cognitive research” is. Cognitive theories Cognitive neuroscience research requires a attempt to describe the functional organization great deal of time and money, and it attracts a of the mind (Gallistel, 1990). The guiding great deal of . Consequently, there is a principle of the is that we subversive temptation to declare it useless. This can understand how the mind processes general grumble comprises many related gripes: . Specifically, we can characterize Cognitive neuroscience is poorly executed; its mental activity as a set of representations and costs do not justify its benefits; for a long time we computations over those representations. This all got along just fine without it, and I still do; it’s depends crucially upon an understanding of the just a fad; people are just wowed by pictures of of the mental processes under study ; etc. (Marr, 1982). Still, is the study of mental compu- Some classic examples of psychological tations that are performed by a neural machine. research questions formulated in this framework It is remarkable to suppose that studying the are visual object recognition, machine itself would not usefully inform production, and theory of mind. In the case of cognitive research. So which is it? Does visual object recognition, for instance, it is neuroscience illuminate cognitive research, or is assumed that the mind begins with a pixel-like it the moon that never rose? Is cognitive representation of the wavelengths and neuroscience an oxymoron? luminosities of light across the . It ends This essay dodges numerous gripes. It does with the capacity to categorize objects in the not address whether most cognitive neurosci- visual scene—e.g., categorizing one as a CAT and ence is well executed; whether its benefits another as HAT. This occurs as successive typically outweigh its costs; whether neurosci- computations are performed on the early ence could ever be necessary to develop a representation, transforming it over a series of psychological theory; or whether its celebrity representations. outstrips its merit. But it does conclude that This commitment to characterizing the neuroscience can important role in internal representations that mediate between cognitive research. The argument has three percept and action is precisely why behaviorists parts: A general account of the relationship found the cognitive revolution so revolting. between neuroscience and cognition, a specific is committed, of course, to merely of the contribution of neuroscience to characterizing the relationship between a the study of visual cognition, and a quantitative and a response. This is a kind of analysis of citation records in the journal psychological theory. Indeed, it aptly describes Cognition. the kinds of theories of applied in some other disciplines. For instance, several 1.1 What is cognitive research? key principles of microeconomics do not attempt to mirror the representations or computations Before asking whether cognitive neurosci- involved in choice, except insofar as they ence contributes to cognitive research, it helps to appropriately characterize the relationship

1 Is cognitive neuroscience an oxymoron? 2 Forthcoming: Current Controversies in Philosophy of , Eds. S. Cullen, S.J. Leslie & A. Lerner between input and output of a human chooser of theory they seek to advance, this would beg over time. Such theories achieve descriptive the question of whether neuroscience methods adequacy, in the that they accurately (e.g., cognitive neuroscience) can inform describe and predict human behavior (Chomsky, cognitive theory. The answer would be 1965). Cognitive theories, in contrast, guaranteed to be “no” for a trivial reason: Any additionally achieve explanatory adequacy. That study using apparent “neuroscience” methods to is, they aim to characterize the set of internal answer cognitive questions would simply have representations and computations that provide been “cognitive” research all along, by definition. an explanation for the reliability of descriptive relations between input and output. Thus, if 1.3 What is “important”? neuroscience plays an important role in cognitive research, it must do so by constraining theories What does it mean to say that neuroscience of representation and computation. has played, or could play, an “important” role in the of cognitive theories? This 1.2 What is neuroscience? question is most interesting if it is reframed pragmatically: If we are interested in developing Neuroscience is the study of the nervous an accurate model of cognition, is it a useful system, including its physical and biological strategy to seek evidence from neuroscience properties. Like all cognitive theories, many methods? neuroscience theories are motivated by the Thus, neuroscience could be a useful source ultimate desire to explain the functional of evidence without being sufficient for the organization of the mind. Yet theories of development of cognitive theories. In other neuroscience can be distinguished from words, having a complete theory of cognition cognitive theories in part because they are stated may require us to know about more than just the over physical and biological terms (or explicit . A complete cognitive theory of visual abstractions from them) rather than over processing, for instance, might depend upon exclusively representational and computational information about typical scenes in natural terms. contexts (i.e., the things we see), as well as the Neuroscience comprises not only a set of functional role of vision in guiding human theories, but also a set of methods. A neurosci- interactions in the world. Whether cognitive ence method involves either manipulating or theories must be stated over terms external to measuring neural processes directly. In other the is a matter of significant words, it either involves an intervention upon philosophical controversy. But, even if we adopt neural processes in a manner that is not the position that a theory of mental representa- mediated by ordinary (e.g., neural tion and computation must be stated partially stimulation by electrical or magnetic forces, over terms external to the nervous system, the lesion by natural or artificial means, experi- could still play a useful role in mental control of gene expression, etc.), or else it developing those theories. involves measuring neural processes in a manner Similarly, neuroscience could be a useful that is not mediated by ordinary behavior (e.g., source of evidence in developing cognitive EEG, MRI, single unit recording, etc.). In theories without being a definitive source of contrast, in “behavioral” methods, the evidence. (Indeed, it’s hard to say whether any experimenter’s access to mental processes is form of scientific evidence is ever definitive). mediated by ordinary perceptual and behavioral Presumably neuroscience will constrain processes. cognitive theories via inference to the best It is not necessary that neuroscience meth- explanation. That is, certain cognitive theories ods are used exclusively to inform neuroscience will provide a better explanation for the theories, or that behavioral methods are used neuroscience evidence, others will provide a exclusively to inform cognitive theories. This worse explanation, and we can use this to inform point is crucial: If we defined methods in terms our assessment of the quality of the cognitive

2 Is cognitive neuroscience an oxymoron? 3 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner theories. In this method of reasoning the For instance, there are detailed theories of the neuroscience evidence can play an important cellular mechanisms that enable neuronal firing; role in adjusting our assessment of the however, these are unlikely to be sufficiently probability of cognitive theories being correct, related to and computa- even without definitively endorsing or defeating tion to play an important role in cognitive any particular theory. research. Even if every cognitive representation Neuroscience could also be a useful source of is encoded by the firing of , this does not evidence without being necessary for the mean that a theory of the mechanics of neural development of cognitive theories. This point is firing at the cellular level informs a theory of obvious enough. If I want to know what color cognitive representation. my socks are, looking at them would be a useful Consider a slightly different question, source of evidence, but it is certainly not a however: If a scholar wishes to understand the necessary source of evidence. I could buy a influence of Uncle Tom’s Cabin on American camera, hire a film crew to shoot a movie of my attitudes towards slavery, would it be useful for socks, show it to my wife and ask her what color her to read the book—i.e., to interact directly she sees. But even if it isn’t necessary for me to with it as a physical object? Presumably it look at my own socks to determine their color, it would, simply because this would provide could certainly be useful. relevant information about its contents, which Finally, it might be the case that neurosci- would in turn inform a theory of its influence. ence is a possible source of evidence without Reading the book might not be necessary. The being a practical one. For instance, the film- scholar could rely on indirect sources of crew approach is a possible method of attaining evidence about its content, and many of the evidence about the color of my socks, but it isn’t inferences she would wish to draw about its a very practical one. Although practicality is influence wouldn’t depend on knowing its important, I won’t address it much further. precise contents anyway. Nor, of course, would reading the book be sufficient to answer all the 2. A priori analysis scholar’s questions. But if the scholar had never read the book before, and if it sat on her shelf, The brain is the principal physical substrate surely she would be crazy not to pick it up. of human cognition. Doesn’t this guarantee that This analogy highlights the utility of distin- theories of brain function will be relevant to guishing between neuroscience theory and theories of cognition? In fact it does not, as a neuroscience methods. A general theory of simple analogy illustrates. Consider a scholar “books” as a physical object might be stated over who wishes to develop a theory of the influence terms irrelevant to the informational content of of Uncle Tom’s Cabin on American attitudes any given book—e.g., principles of chemical towards slavery. This novel was distributed in bonding, methods of manufacture, etc. Such printed books, and so the influence of its content theories would typically be useless to a scholar upon readers was fully mediated by physical interested in their informational content. Yet, objects. Would it be useful, then, for the scholar the “method” of interacting physically with to investigate properties of those objects, such as particular books (i.e., reading them) is obviously their material composition (paper and ink) or a very useful method of understanding their method of manufacture? Probably not: This informational content. Similarly, it might be the approach would fail because general information case that theories of neuroscience typically about the material composition of the book or its generalize over properties of neural systems that method of manufacture are not sufficiently are not particularly informative about mental related to the information contained in the book, organization at the level of representation and which is what shaped American attitudes computation. Yet, interacting directly with the towards slavery. physical substrate of a neural system might be a The same is true for many properties of very useful method of about the neural systems when applied to cognitive theory.

3 Is cognitive neuroscience an oxymoron? 4 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner representations and computations instantiated in work, which is an ideal case study of the one. potential for neuroscience to advance cognitive Of course, reading the brain is not as easy as research. reading a book. If your uncle Tim is acting oddly Feed-forward visual processing begins when and you want to understand why, it probably light hits a photoreceptor in the retina. These makes more sense to use behavioral methods photoreceptors relay through another layer of (i.e., observing the relationship between cells to eventually activate retinal ganglion cells, perceptual inputs and behavioral outputs) than which transmit information from the eye to the to crack open his head and take a close look at brain. In the years surrounding World War II, a his brain. In order for the “direct” investigation number of researchers performed single-unit of the neural system to usefully inform recordings of the evoked response of retinal , we must have theories capable of ganglion cells during the presentation of brief relating physical data to cognitive theories, and visual stimuli to the eye (Barlow, 1953; H. instruments capable of collecting that data. Hartline, 1940; H. K. Hartline, 1938; Kuffler, Given such theories and instruments, however, it 1953). This line of investigation lead to a pivotal is hard to see how neuroscience methods could characterization of a type of receptive field fail to be informative for cognitive research. common to many retinal ganglion cells: “center surround” (Figure 1). 3. A case study

A classic cognitive research problem is to understand the information processing that enables vision. Somehow, we transform a dynamic two-dimensional projection of light onto the retina in each eye into a three dimensional representation of space, a categorization of objects within that space, and a representation of the motion of those objects over time. In cognitive terms, the goal is to understand the of the information represented on the retina, and then to specify the Fig. 1 A of an illustration from Kuffler (1953) showing the receptive field of a retinal ganglion series of computations that transform this . In the diagrammed receptive field, the innermost information across a series of representations, portion responds preferentially to the onset of light, resulting finally in representations of three while the outermost portion responds preferentially to the offset of light. dimensional space, object identity, motion, etc. Such a cognitive theory can be neutral with respect to neural mechanism. For instance, An immediately subsequent stage of pro- consider the hypothesis that a key step during cessing was characterized in a classic series of object recognition is the discovery of object experiments by Hubel and Weisel (1962). This boundaries by computing local luminosity stage takes as input representations with a gradients. By itself, such a hypothesis is not center-surround format. By computing the committed to any particular neural mechanism; spatial relations among cells of this type, it indeed, it applies as naturally to visual derives a representation of local linear contrast processing on a laptop computer as it does to boundaries and their orientation (Figure 2). visual processing in a brain. Yet, over a period of Here, again, the experiments that established the about 20 years, actually did drive representational content of early visual pioneering advances in our understanding of the processing were single-unit recordings, in this earliest stages of visual processing. Now several case taken from the lateral geniculate nucleus decades past, we are in a good position to and cortical area V1. appreciate the historical significance of this

4 Is cognitive neuroscience an oxymoron? 5 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner

These experiments had a transformational problem famously addressed in Lettvin, impact on the field because they provided a case Maturana, McCulloch, & Pitts, 1959). study of the mapping between neural function During this formative period of the study of and cognitive organization (Gordon, 2004; visual processing, then, neuroscience methods Wurtz, 2009). Experimental evidence drove breakthroughs in cognitive theory. Hubel characterized the representational content of and Wiesel’s experiments were examples of three successive stages of visual processing, neuroscience methods because their dependent specifically in terms of the activation of measure was a physiological signals produced by photoreceptors and the receptive fields of the brain, rather than behavior. Yet, these same individual neurons across two populations. And, experiments were designed to build a cognitive a biologically plausible model showed how theory of vision. Their basic strategy, which was connections between cells and their individual to characterize the response profile of neurons response properties could instantiate the (i.e., their receptive fields) is reflected across a computation of the later representations from vast subsequent literature, and encompasses not earlier ones. Perhaps most importantly, it was just single unit recording, but also EEG, PET, immediately obvious that this element of visual MRI, etc. (DiCarlo, Zoccolan, & Rust, 2012). processing constituted useful progress towards Indeed, an immediate impact of Hubel and the functional purpose of object recognition. Weisel’s research was to prompt new directions Specifically, a retinal map of luminosity was in behavioral and computational research on being transformed into elements useful for the early visual processing. The receptive fields of detection of object edges in natural scenes some LGN and V1 neurons characterized by (which often involve local high-contrast Hubel and Weisel resemble a local sinusoidal boundaries). Later cognitive theories posited contrast gradient. These are distributed at that this process of edge detection plays a crucial variable orientations and over variable spatial role in object identification and the construction frequencies (i.e., variable “widths” of the sine of a three dimensional representation of space wave). It was soon noted that a population of (Marr, 1982). neurons with receptive fields of this type would be capable of representing a visual scene by approximating a Fourier decomposition of the spatial frequency of luminosities. Eventually this line of inquiry established a correspondence between the receptive fields of cells and Gabor wavelets, a mathematical object that enables an efficient compression of the information contained in natural scenes. In fact,

the basic strategy of decomposing natural scenes Fig. 2 A reproduction of an illustration from Hubel & into summaries of local spatial frequency is Weisel (1962) showing how the activation levels of shared with common image compression several center surround cells could be summed in order to produce the receptive field of a complex cell. algorithms for computers, such as the JPEG format. Notably, in this case the neural evidence As this computational interpretation of the preceded the development of a cognitive theory. neuroscience was developed, a number of clever Researchers had not specified a hypothesis of behavioral paradigms were devised to visual processing that included representations demonstrate their (e.g. Campbell & of center surround receptive fields. To the Robson, 1968). Thus, in all likelihood, the same contrary, similar experiments performed on the cognitive theory could have been developed frog’s retinal ganglion identified representational without constraint or inspiration from contents that long confounded any attempt to neuroscientific evidence. Even if a suitably relate to the functional demands of vision (a precise cognitive theory of visual processing could be developed without evidence from

5 Is cognitive neuroscience an oxymoron? 6 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner neuroscience, however, it is easy to grasp how neuroscience will, of course, tend to cite that pivotal the role of neuroscience evidence was in very neuroscience. Thus, we might ask whether this case. A characterization of the receptive cognitive research that cites neuroscience is fields of individual neurons across multiples subsequently more cited. This evidence would stages of representation enabled Hubel and be consistent with the hypothesis that cognitive Weisel to posit an account of the representations research that is influenced by neuroscience is of and computations contained in early visual a higher quality (or at any rate more influential), processing that had not really been entertained although it is also consistent with some in any serious way by prior theorists. In other alternatives. words, they made a big step, relatively fast, and Of course, it is crucial that this hypothesis is there is little indication that the step would soon tested on papers that are unambiguously devoted have occurred without reliance on neuroscience. to the development of cognitive theory. It is Above, I introduced the conceptual distinc- possible, for instance, that neuroscience research tion between neuroscience theory and is more cited that cognitive research, but this is neuroscience methods. The key contribution of not the question. Our goal is not to establish Hubel and Weisel’s research to a cognitive whether neuroscience is influential in general. theory of visual processing appears to have been Rather, the goal is to assess whether neurosci- their use of neuroscience methods, not their ence improves cognitive research. Thus, we development of a general theory of neuroscience. want to consider only articles that are cognitive In other words, the information they obtained research, asking whether those that cite about neural responses did not influence neuroscience (i.e., are influenced by it) are cognitive theory by way of some general theory subsequently more cited (i.e., are of a higher of the nervous system—i.e., a theory that quality, or more influential). generalized over physical rather than We assessed this question in a sample of informational properties of the brain. Hubel and articles drawn from the journal Cognition. This Weisel instead moved directly from data about journal publishes articles almost exclusively on the neural system, collected by neuroscience cognition. It publishes almost no research methods, to theory of information processing. employing neuroscience methods (e.g. fMRI), Although their theory of information processing and the few articles that do employ neuroscience could also be construed as a “neuroscience methods are designed to address theories of theory”—after all, it described the implementa- cognition. Indeed, among the sample of articles tion of a algorithm in terms of the firing of and that we targeted, a search on the Thomson connections between neurons—it was no less a Reuters Web of Science did not reveal any that theory of cognition. By marrying an algorithmic included “fMRI” in its title or abstract. The specification of cognitive processing to a audience of Cognition is primarily , biological plausible neural implementation and it is among the more influential and widely (Marr, 1982) it was, without irony or contradic- read journals specific to the field of cognitive tion, an instance of true “cognitive neurosci- psychology. ence”. This analysis focuses on articles published during the calendar years 2008, 2009 and 2010. 4. Quantitative analysis This affords a large sample of published articles (N = 540) for which we could obtain adequate If neuroscience is important to the develop- citation information from the Web of Science. ment of cognitive theories, then cognitive We chose this range of dates because it seemed research that is directly informed by neurosci- late enough for the published articles to have ence will tend to be of a superior quality. How potentially benefitted from the explosive growth can we test this hypothesis by quantitative of cognitive neuroscience in the preceding analysis? A common proxy for the quality of decade, but early enough to have amassed a scientific research is its citation index. And, meaningful number of citations representative of cognitive research that is “influenced” by their quality.

6 Is cognitive neuroscience an oxymoron? 7 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner

For each article we recorded the number of sample. This pattern of correlations may Web of Science indexed citations (hereafter obscure the predicted effect of neuroscience “citations”) through August 2015 (the date at references on subsequent citations, and so it is which this manuscript was initially prepared); important to statistically control for the effect of this is our measure of the article’s quality and time. Third, articles that contained more influence. We used an automated text extraction references overall were more likely to contain at algorithm to identify the references cited in each least one reference to neuroscience, and also to article PDF (hereafter, “references”). This receive more citations (e.g., perhaps because of algorithm identified 21,159 references, including variable rates of citation across subdisciplines, many to the same sources. We then categorized reciprocity among authors, a correlation references targeting “neuroscience” (versus not) between an exhaustive literature review and by assessing whether each referenced article was article quality, etc.). This pattern of correlations published in any of 62 neuroscience journals, or may artificially inflate the predicted effect; again, whether its title contained word stems it is important to statistically control for it. associated with neuroscience (e.g. “Neur”, Thus, we performed a regression that “Brain”, etc.). predicted the natural logarithm of the number of The sample of articles we analyzed were citations received by each article in our sample. cited an average of 31 times each. Out of 540 We included three predictors: a dichotomous total articles, 311 (58%) referenced at least one variable that coded whether each article neuroscience article, while 229 (42%) referenced contained any references to neuroscience, a none. Articles that contained at least one variable coding the year of publication, and the neuroscience reference received an average of 34 total number of references contained in each citations, while those that contained no article. Again, we found a statistically significant neuroscience references received an average of effect indicating that articles that referenced 26 citations, which is a statistically significant neuroscience were more likely to receive difference t(538) = -2.88, p < .005. In other subsequent citations, p < .005, after controlling words, articles that referenced neuroscience for the other two predictors. tended to receive about eight more citations on Although we find clear evidence in this average than articles that did not, an increase of model that referencing any neuroscience is 31% in the frequency of citation. A comparison correlated with a higher citation index, we do of the most cited to least cited articles helps to not find evidence that referencing more illustrate this trend. Among the 20 most cited neuroscience helps. Specifically, after articles (mean = 148 citations), 17 contained at controlling for the presence versus absence of least one neuroscience reference. In contrast, any neuroscience reference, there is no further among the least cited articles (mean = 3 statistical relationship with the specific number citations), only 9 contained at least one of neuroscience references (p > .50) or the neuroscience reference. proportion of neuroscience references out of all This analysis can be improved in a few references (p > .50). On the one hand, this simple ways. First, citations are not distributed provides some evidence against the hypothesis normally across this sample of articles; rather, that articles referencing neuroscience are cited the distribution is highly right skewed, violating more only when they actually contribute to the an assumption of the parametric statistical tests neuroscience literature (i.e., report neuroscien- we employed to analyze our data. A log tific results, address neuroscientific theories, transformation of citation counts produces an etc.) Were this the case, presumably an article adequately normally distributed dependent that cites many neuroscience findings (and thus variable. Second, as might be expected, articles more likely contributes to the neuroscience published earlier (e.g. 2008) tended to receive literature) would receive more citations than an more citations than articles published later (e.g. article that cites only one or two neuroscience 2010), while the likelihood of neuroscience findings (and thus likely does not contribute to references also increased over this range in our the neuroscience literature). On the other hand,

7 Is cognitive neuroscience an oxymoron? 8 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner if neuroscience improves cognitive research, it is driven by features specific to different surprising that we do not find evidence that subdisciplines. more neuroscience is associated with greater Another explanation for the observed improvement. pattern of results is that references to Although the evidence described above is neuroscience impart a false aura of quality. In consistent with the possibility that neuroscience other words, people may cite articles that improves cognitive research, there are several reference neuroscience because they have the important alternative explanations for the impression that the research or theory is more observed effect. One possibility is that it may be cutting-edge, valid, interdisciplinary, plausible, explained by variability across sub-disciplines of etc., but in fact those articles may not differ in cognitive research. For instance, research into any of these respects. This explanation is may infrequently refer to difficult to rule out because of our use of citation neuroscience (e.g., because little neuroscience is as a proxy for quality. It may be amplified by our performed on infants and children) and also reliance on a sample of recently published receive fewer citations (e.g., because the pace of articles; perhaps the “early returns” of citations research is relatively slow), compared with other are driven relatively less by actual research subdisciplines. This would produce a spurious quality than by heuristic approximations of it. confound between neuroscience references and The foundations of this argument are more than subsequent citations. hypothetical: Perhaps ironically, the sixth-most- We conducted a quick and rough test of this cited article in our sample is titled, “Seeing is account, testing for the presence of the basic believing: The effect of brain images on “neuroscience advantage” already identified, but judgments of scientific reasoning” (McCabe & now within four separate categories of research. Castel, 2008). It shows that college students are We identified articles belonging to each research more likely to accept a scientific argument as categories by searching for keywords or word valid if it is accompanied by an uninformative stems in the titles of articles contained in our fMRI activation map. (On the other hand, other sample. For instance, twelve articles contained research provides some evidence that scientific the word “moral” in the title, and thus experts are immune to this effect (Weisberg, presumably focused on moral cognition. Eight of Keil, Goodstein, Rawson, & Gray, 2008)). these articles referenced neuroscience, and these A final alternative explanation for this effect received an average of 92 citations each; four did is that articles referencing neuroscience are not reference neuroscience, and these received more likely to be cited by the neurosciences, but an average of 13 citations each. Among 25 not by other branches of cognitive science. Why articles with the word “face” in the title, those might this be? An unlikely possibility is that referencing neuroscience received more articles in the journal Cognition that reference citations (mean = 29) than those that did not (25). neuroscience are being cited by neuroscientists Among 60 articles with the word stems “ling” or not because of their contributions to cognitive “lang” in the title, those referencing neurosci- theory, but rather because of contributions to ence again received more citations on average non-cognitive theories of neuroscience—e.g., a (42 vs. 22). The same trend emerged among 58 model of network dynamics, or articles with the word stem “child” or “infant” in receptor types, etc. This possibility is unlikely the title (35 vs. 32). These are the only categories because Cognition just doesn’t publish articles we analyzed; clearly, further and more with this kind of content. (Recall that not a systematic approaches could be taken to address single article in our sample contains “fMRI” in this question. Yet, while the magnitude of the the title or abstract). “neuroscience advantage” is variable across A second possibility is that cognitive neuro- categories, in each case there is a trend in the scientists are more likely to cite articles (or same direction. This provides some preliminary authors) who cite them, perhaps via an implicit evidence that the overall effect is not spuriously or explicit reciprocity, or because they find it easier to relate such work to their own interests.

8 Is cognitive neuroscience an oxymoron? 9 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner

This hypothesis is more likely, and also harder to articles that did not refer to neuroscience). This rule out based on our data. Because the citation analysis, however, is compromised by a marked advantage that we observe is not eliminated reduction in statistical power. when controlling for the total number of In summary, our analysis provides strong citations, we can at least provide some evidence evidence that references to neuroscience articles against a pure effect of peer-to-peer reciprocity are associated with higher citation rates for in citation (assuming that neuroscientists are no cognitive research in Cognition. Yet, this data is more prone to such reciprocity than non- compatible with a number of rival hypotheses neuroscientists), but other versions of the and does not provide exclusive support the hypothesis are not addressed by this analysis. conclusion that neuroscience has an important There is a third version of this alternative influence on cognitive research. that is perhaps the most likely, but is friendly to the overall claim that cognitive research is 5. Conclusion improved by engagement with the neuroscienc- es. Possibly, cognitive research that engages Cognitive neuroscience aims to characterize seriously and directly with neuroscience is also the interface between computation and , more useful to neuroscientists who are working and it is not an oxymoron. A priori, there is good on cognitive questions—i.e., cognitive reason to believe that cognitive neuroscience neuroscientists. For instance, research that uses will advance cognitive theory. Although it is not behavioral methods to investigate early visual clear whether “theories” of neuroscience will representations would likely benefit from often play this role, there is every reason to research on this topic using neuroscience believe that neuroscience methods will be methods; in turn, it is especially likely to inform pivotal. This depends, of course, on the future research using neuroscience methods. In existence of theories that map between physical this case, our effect might be driven by citations properties of the brain and computational from neuroscientists, and yet it would still be properties of the mind. A case study of an early diagnostic of the underlying quality and broad advance in cognitive neuroscience, the mapping applicability of the research for the development of receptive fields in the retinal ganglion and of cognitive theories. early visual cortex, shows that this approach can Based on the set of citation records available succeed in practice. This case study underscores to us we could not develop a reliable method to the utility of neuroscience methods for directly assess whether our effect was due describing representations and computations entirely to enhanced rates of citation by that are many stages of processing removed from neuroscience for those articles in our sample that behavior and, thus, difficult to characterize by referenced neuroscience. The most relevant behavioral methods alone. Of course, this case evidence we found targets a different, but related study may be anomalous. Yet, in an analysis of a question: Are references to neuroscience related sample of articles developing cognitive theories, to increased citation from articles that are not we find that references to past neuroscience neuroscience? In order to assess this, we research are associated with a greater impact as analyzed the number of citations for each article measured by citation index. This is at best an in our sample, focusing exclusively on indirect proxy for the underlying “quality” of subsequent articles also published in Cognition. cognitive research, however, and the evidence In other words, we asked, “For three years’ presented here is consistent with several worth of articles published in Cognition, if they alternative hypotheses. referenced neuroscience, were they then more Each of the methods used above—a priori likely to be cited by subsequent articles also analysis, case study and quantitative analysis— published in Cognition?” In this case we found has its own drawbacks. Together, however, their no significant difference in citation rates; indeed, complementary strengths provide a unified if anything the citation rates trended in the account of the important role that evidence from opposite direction (i.e., more citations for

9 Is cognitive neuroscience an oxymoron? 10 Forthcoming: Current Controversies in Philosophy of Cognitive Science, Eds. S. Cullen, S.J. Leslie & A. Lerner the neurosciences can play in advancing cognitive research.

Acknowledgements

Thanks to Matthew Cashman, who played a pivotal role in compiling the database of citation records used in this article, and to members of the Research Lab for valuable feedback.

References

Barlow, H. B. (1953). Summation and inhibition in the frog's retina. The Journal of , 119(1), 69- 88. Campbell, F. W., & Robson, J. (1968). Application of Fourier analysis to the visibility of gratings. The Journal of physiology, 197(3), 551-566. Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge: MIT Press. DiCarlo, J. J., Zoccolan, D., & Rust, N. C. (2012). How does the brain solve visual object recognition? , 73(3), 415-434. Gallistel, C. R. (1990). The organization of learning. Cambridge: MIT Press. Gordon, I. E. (2004). Theories of : Psychology Press. Hartline, H. (1940). The effects of spatial summation in the retina on the excitation of the fibers of the . American Journal of Physiology. Hartline, H. K. (1938). The response of single optic nerve fibers of the vertebrate eye to illumination of the retina. American Journal of Physiology. Hubel, D. H., & Wiesel, T. N. (1962). Receptive fields, binocular interaction and functional architecture in the cat's visual cortex. The Journal of physiology, 160(1), 106. Kuffler, S. W. (1953). Discharge patterns and functional organization of mammalian retina. Journal of , 16(1), 37-68. Lettvin, J., Maturana, H., McCulloch, W., & Pitts, W. (1959). What the frog's eye tells the frog's brain. Proceedings of the IRE, 47(11), 1940-1951. Marr, D. (1982). Vision. New York: Freeman. McCabe, D. P., & Castel, A. D. (2008). Seeing is believing: The effect of brain images on judgments of scientific reasoning. Cognition, 107(1), 343-352. Weisberg, D. S., Keil, F. C., Goodstein, J., Rawson, E., & Gray, J. R. (2008). The seductive allure of neuroscience explanations. Journal of Cognitive Neuroscience, 20(3), 470-477. Wurtz, R. H. (2009). Recounting the impact of Hubel and Wiesel. The Journal of physiology, 587(12), 2817- 2823.

10