2.1. Objects of Research as Epistemically Blurry

We typically think of objects as three-dimensional things in the world, clearly delineated from their environments. Objects of research in psychology are not like that. Take memory, for example. First, while there is little doubt among researchers that memory is a legitimate object of scientific investigation, memory is not a physical object in the usual sense. Second, I claim that psychologists study memory in part because they are not entirely sure about its shape and contours (or even, perhaps, about whether at the end of the day this will be a viable scientific kind). Put differently, when we put ourselves in the epistemic position of practicing scientists, it is clear that they are not (or at least not exclusively or even primarily) in the business of explaining some already clearly delineated phenomenon. Rather, much of their empirical work is directed toward figuring out what they are actually investigating. Following Feest (Reference Feest2011a), I therefore refer to such objects of research as “epistemically blurry.”

With this thesis I do not mean to deny that ultimately the point of categories like “memory” is to explain behavior. In this vein, my approach is somewhat compatible with more traditional philosophy of psychology, according to which memory is a theoretical concept (an explanans), put into place to explain certain types of behaviors (explanandum phenomena). However, as I hope to make clearer in the course of this article, I think that this way of carving up the field (by dividing it into explananda, which are clearly delineated, and explanantia, which are to be determined) does not lend itself to an adequate analysis of the research process itself, which is importantly concerned with making taxonomic constructs and attempting to describe distinctive features of the (purported) objects in their extension. When I say that objects of research are epistemically blurry, I do not mean that scientists lack a theoretical explanation for an already clearly delineated empirical explanandum but that the very question that empirical data are even descriptively relevant to the object in question is part of the investigative project. Thus, it is my contention that often explanation is not the most immediate concern of experimental research in cognitive science.

I use the expression “object of research” to indicate the presumption of a research target that is independent of our own conceptualizations. Notice, though, that I am not committed to the claim that this presumption will pan out in every case. There may well be cases in which the “object” in question will turn out not to possess the inductive projectability required of a scientific kind. In that case, we will probably see the terms being eliminated in the long run. At the same time, this expression is also supposed to indicate that such (presumed) objects are research targets only by virtue of the fact that we direct our epistemic attention to them and hence that we tentatively use some concepts to describe our subject matter. The research process, as I construe it here, is thus one of simultaneously exploring a specific subject domain and of applying, revising, and extending existing concepts. In this respect, the outlook taken here is not so dissimilar from Wilson’s (Reference Wilson2006) study of the often unpredictable behavior of concepts when confronted with, or extended to, previously unimagined physical domains or circumstances.

2.2. Short-Term Memory as an Object of Research

I believe that the rough gloss just provided can make sense of research practices we encounter in psychology, where a great deal of time and effort are spent trying to arrive at adequate descriptions of the presumed research objects. Consider the example of working memory, which is typically thought of, or modeled, as a stage of information processing before items get committed to long-term memory (see, e.g., Baddeley and Hitch [Reference Baddeley, Hitch and Bower1974] for a classic formulation).Footnote ² I claim that the point of this model is that of delineating and empirically describing a particular object of research (working memory) in relation to long-term memory. To be sure, such work is theoretical, but it does not follow that its immediate aim is explanation. Rather, the empirical research in question aims at gaining a better descriptive understanding of the very objects under investigation.

Let me quickly illustrate the nature of such research (and thereby also of the epistemic blurriness of the research object in question) by looking at debates about two descriptive features long taken to be characteristic of short-term memory: duration and capacity, where the duration was thought to be about 10 to 20 seconds and the capacity $7\pm 2$ (Miller Reference Miller1956). The term “working memory capacity” refers to the number of items that can be stored in working memory at any given time. When talking about efforts to determine working memory capacity as illustrating my notion of an epistemically blurry research object, what I have in mind is not the fact that there is some variation between different individuals’ working memory capacities but rather that it is not a trivial (i.e., straightforwardly empirical) question how to determine a figure like $7\pm 2$. When scientists run a psychological experiment, by what rationale do they determine how many memory items are indicated by any given subject’s experimental response? My claim here is that these are questions that are quite central to researchers’ debates about their subject matter.

One reason why the question of how to determine working memory capacity is not trivial has to do with a particular phenomenon thought to be relevant to short-term (or working) memory: the phenomenon of chunking (Miller Reference Miller1956). According to the chunking hypothesis, working memory capacity is fixed. However, once chunking takes place, the amount of information that can be kept in working memory radically increases, because the items are now composed of chunks of smaller items. If we want to describe working memory capacity, chunking will inevitably pose obstacles, since the capacity limit may well be inflated by the effects of chunking. Thus, in order to produce an adequate measure of capacity, we need to ensure that we have a way of empirically identifying the chunks (as opposed to their parts) that subjects keep in working memory. None of these are explanatory questions. They are, rather, theoretical and methodological questions about how to investigate and describe a given object of research.

Cowan (Reference Cowan2000), in an influential paper (in which he argues that Miller’s famous $7\pm 2$ estimate should be corrected downward to $4\pm 1$), tries to develop a method for distinguishing between memory chunks and the total number of items stored in memory, arguing that chunks are associations between concepts and that one needs to control for items that are already “chunked” in long-term memory. However, as the peer responses to Cowan’s article show, there are also problems with this approach. For example, one question that comes up is how it can be established that higher-order coding has in fact been prevented (Beaman Reference Beaman2000, 118), suggesting that Cowan is begging the question by presupposing a particular understanding of chunking. Others argue that Cowan’s theoretical definition of a chunk is too weak (Halford, Phillips, and Wilson Reference Halford, Phillips and Wilson2000; Milner Reference Milner2000; Pascual-Leone Reference Pascual-Leone2000). Yet others worry whether empirical findings about chunking can be generalized across modalities (Schubert and Frensch Reference Schubert and Frensch2000, 146).

I take the debate to be one about how to adequately describe a particular feature (storage capacity) of a specific research object (working memory). My aim here is not to take sides in the debate but to use it to illustrate my thesis that scientists are often concerned with providing descriptions of research objects and that there is frequently a great deal of epistemic uncertainty about the objects in question.

3.1. Two Meanings of “Phenomena.”

According to Bogen and Woodward, phenomena are “relatively stable and general features of the world” (Woodward 1988), which “have stable, repeatable characteristics” (Bogen and Woodward Reference Woodward1989, 393). In this vein, one might suppose that memory is a phenomenon that has certain repeatable characteristics, such as the characteristic of chunking. However, when we look at the examples provided, it is clear that Bogen and Woodward regard the chunking effect itself as a phenomenon, not simply as a characteristic of some other phenomenon. So, it seems that they use the term in two distinct senses: (1) as a stable feature or kind that has repeatable characteristics and (2) as simply a repeatable regularity.

I contend that by running these two senses together, we give up an important analytical distinction, a distinction that will help us—among other things—clarify how objects of research are conceptualized in psychology, why they are epistemically blurry, and (most importantly) how they are investigated. In this vein, I suggest restricting the use of the term “phenomenon” to things like the chunking effect, the melting point of lead, and other events that occur regularly under specified circumstances. Now, it is certainly conceivable that such types of phenomena can become objects of research in the sense outlined above; that is, scientists can turn their epistemic attention to them, explore them, probe them by varying the experimental conditions under which they occur, and so forth. However, I contend that objects of research in psychology are often more variegated than that. For example, I suggest that scientists think of working memory as a cluster of phenomena, such as chunking, encoding, retrieval, duration. Thinking of objects of research in this way highlights once again what makes them epistemically blurry: they are epistemically blurry both because the phenomena they are thought to be composed of are ill understood in their own right and because the ways in which those phenomena are related to one another are ill understood as well. Figuring out the exact contours of an object of research involves providing descriptions of the various phenomena tentatively thought to constitute the object of research and providing an account of how they hang together.

3.2. Phenomena versus Effects?

My use of the term “phenomenon” includes regularities both at the experimental level and at the level of processes presumed to exist independently of experimental creation. In this respect, my account differs from Bogen and Woodward’s (Reference Bogen and Woodward1988), who argued that phenomena are mind-independent regularities in the world, to be distinguished from experimental data, which are idiosyncratic to specific experimental contexts. My account also differs from Hacking’s (Reference Hacking1983) contention that phenomena often only exist under the very specific experimental circumstances of their creation. Responding to statements such as this one, in turn, Bogen and Woodward suggested that what Hacking referred to as phenomena were really data in their sense.

The distinction is not so clear, however. I argue that experimental effects should be considered phenomena. Consider the chunking effect. Bogen and Woodward use this as an example of a phenomenon by virtue of its stability and repeatability and by virtue of the assumption that it exists “in the world.” Nonetheless, it is also clearly an experimental effect. Experimental effects, insofar as they are stable and repeatable, are phenomena in the sense I use the term here. Hence, I argue that phenomena can occur on a scale between the context-specific and local and the context-transcendent (see Feest [Reference Feest2011b] for a similar point), although it is typically assumed that local effects instantiate context-transcendent ones. If, as I contend here, objects of research (such as the presumed referents of the taxonomic categories in psychology) are often composed of multiple phenomena, the question is how can experimentally produced phenomena contribute to the ways in which objects of research are described and constructed. I argue that experimentally created phenomena often function as evidence in the course of scientific attempts to provide empirical descriptions of objects of research. In this vein, I suggest a shift away from the prevailing concern with phenomena as explananda toward a consideration of phenomena as experimental evidence (see also Cummins Reference Cummins, Keil and Wilson2000).

At first sight, this latter claim (that phenomena can function as evidence) may be seen as flying in the face of the very distinction at the heart of Bogen and Woodward’s account, namely, that between phenomena (which are the explananda of general theory) and data (which are thought to provide evidence in favor of specific phenomena). While this distinction is very instructive, I think the notion of an object of research puts into sharper relief what purposes experimental evidence serves in the first place. And while I agree with Bogen and Woodward that experimental data are typically not explananda, this does not mean that they are not phenomena. It just means that in the context of research practices in psychology, phenomena are not always “objects of discovery and explanation,” as Bogen and Woodward initially thought. They are, rather, evidence in relation to particular descriptive questions about objects of research.

I also wish to highlight that the local and context-specific phenomena that are created as part of the investigation of specific objects of research are—at least for the duration of the research process—conceptually tied to the research object in question. Put differently, experimentally produced phenomena can function as evidence in relation to specific objects of research, because of an (however preliminary and defeasible) assumption that they are relevant to this object of research.

In this regard, consider, again, the chunking effect. Qua experimental effect it is possible to use it to gather data about an object of research precisely because of the assumption that the empirical effect of chunking provides epistemic access to some presumed cognitive process of chunking, which in turn is thought to be a central descriptive feature of working memory. Notice that these assumptions, however tentative, already need to be in place in order to even investigate issues like the capacity of working memory, thereby highlighting what I have in mind when I speak of working memory as an example of an epistemically blurry research object.

4.1. Phenomena as the Explananda of Mechanistic Explanations

Carl Craver, Bill Bechtel, and other philosophers of neuroscience, when using the term “phenomenon,” often do so in the context of talking about mechanistic explanation, where by “phenomena” they typically mean explanandum phenomena. There are three problems with this. First, the notion of phenomenon, as it is used in this literature, is not entirely clear. Second (and relatedly), it is typically not explicated what the relevant explanandum phenomena of a given mechanistic explanation are. Third, even with a reasonable construal of the explananda of psychological research, I will argue in a moment that the research process is still not very well portrayed as being focused on explanation.

Consider Craver’s (Reference Craver2007) well-known discussion of spatial memory as an object of explanation in cognitive neuroscience. Central assumptions about spatial memory are that it enables us to navigate in space, that it involves spatial maps, and that such maps are formed in the hippocampus by means of the mechanism of long-term potentiation. Now, given Craver’s claim that this is a good example of how mechanistic explanations function in cognitive neuroscience, we may ask what the explanandum phenomenon is in this case. The answer Craver and Darden (Reference Craver, Darden, Machamer, Grush and McLaughlin2001, 121) give to this question is that “the phenomenon to be explained” is “spatial memory,” suggesting that it will be explained by a multilevel mechanism. However, this explication of the explanandum phenomenon does not fit well with their notion of phenomenon a few pages earlier, where they suggest that they use the term “phenomenon” in the sense of Bogen and Woodward (Reference Bogen and Woodward1988), “as relatively stable and repeatable properties or activities that can be produced, manipulated or detected in a variety of experimental arrangements” (Craver and Darden 2001, 114). As examples, they name “the acquisition, storage, and retrieval of spatial memories; the release of neurotransmitters, and the generation of action potentials.” Notice that all of these “phenomena” are more specific and confined than “the phenomenon” of spatial memory, thus raising the question what precisely the authors take to be the explanandum phenomenon of this research.

It seems to me that the authors run together two meanings of “phenomenon” outlined above, that is, on the one hand, specific regularities presumed to belong to a research object (acquisition, storage, retrieval, etc., and perhaps also patterns of search behavior of rats in mazes) and, on the other hand, more complex objects (e.g., spatial memory), conceivably composed of many specific regularities. Related to this, they also do not clearly distinguish between two possible (kinds of) explananda of psychology, namely, (1) the behavioral regularities triggered by particular experimental setups and (2) the capacity of an organism to execute a variety of behaviors in more complex environments (see Chow [Reference Chow2015] for a similar point). I will now briefly consider each of these two candidate explanandum phenomena, arguing that the former are often used as empirical data in support of specific descriptions of the latter.

We begin with the first interpretation of explanandum phenomenon. There are several possible readings of Craver’s (Reference Craver2007, 165) claim that the navigational behavior of mice is “the spatial memory phenomenon.” First, scientists might actually want to explain a particular experimental mouse’s behavior. Second, they want to explain the behaviors of mice more generally, extrapolating from an explanation of experimental mice behavior to an explanation of the behavior of mice in the wild. Third, they want to explain human navigational behavior, extrapolating from the explanation of experimental mice behavior to the explanation of human behavior. Now, again, I am not questioning the assumption that we ultimately want the concept of spatial memory to be an explanatory construct. Nor do I (in this article) question mechanistic accounts of explanation. What I do question, however, is the underlying assumption that experimentally generated phenomena, such as the stable and repeatable regularities of navigational behavior exhibited by mice in mazes, are typically intended to instantiate explananda. Instead, I suggest that they are created to provide evidence in favor of particular descriptive accounts of spatial memory, understood here as an object of research in the sense outlined above.

This brings me to my second interpretation of “the explanandum phenomenon of spatial memory,” according to which the explananda in cognitive science are often not isolated behavioral phenomena but the complex behavioral capacities themselves. Craver (Reference Craver2007) distances himself from Cummins’s use of the term “capacity,” but it strikes me as fairly innocuous, as long as we appreciate that the term can be understood as either (1) referring to an explanandum phenomenon (a complex behavioral capacity of a system) or (2) a placeholder for a detailed descriptive account of a multilevel mechanism (i.e., an explanans). Here, my interest is in the former. I argue that once we are asking what a given behavioral capacity really is, we are inquiring into the empirical characteristics of an elusive research object, and the answer is not simply going to be a matter of filling in the mechanical details of an already clearly delineated explanandum. Again, this draws our attention to the fact that what is at stake here are taxonomic and descriptive questions. In the context of ongoing research, I claim, experimentally generated instances of phenomena often function as evidence with regard to accounts of specific behavioral capacities, not as explananda of mechanistic explanations.

4.2. Phenomena and Mechanisms as Objects of Discovery

It is to the credit of mechanist philosophers that they have put not only explanation but also discovery on the agenda of philosophy of science (e.g., Bechtel and Richardson Reference Bechtel and Richardson1993; see also Darden Reference Darden2006; Bechtel Reference Bechtel2008; Craver and Darden Reference Machamer, Grush and McLaughlin2013). In this context, Bechtel and Richardson have early on highlighted the conceptual openness of the research process, suggesting that phenomena frequently get “reconstituted” in the course of their investigation. It is noticeable, though, that these works are very much focused on the discovery of (explanatory) mechanisms, thus leaving intact the basic idea that the explanation of phenomena, rather than the description of research objects, is the main focus of research.

A widely held assumption in the literature is not only that phenomena are the explananda of mechanistic explanations but also that the discovery of mechanisms typically proceeds by way of constructing “higher level” (phenomenological) descriptions of the relevant explanandum phenomena, with the aim of later supplying the mechanistic explanation, which in turn can “reconstitute” the explanandum phenomenon (e.g., Bechtel and Richardson Reference Bechtel and Richardson1993). In a similar vein, Craver and Darden (2001, Reference Machamer, Grush and McLaughlin2013) describe the process of discovery as a process of “going down one level” to search for the mechanisms and “going up one level” to try to describe again precisely what the mechanisms do (i.e., to describe again the phenomenon).

This seems like a sensible idea, and the authors provide some nice illustrations of how this can work (pointing to top-down experiments that manipulate rats/mice at the behavioral level and observe effects on the level of the brain and, vice versa, bottom-up experiments). However, I argue that this account at best captures a small part of research practices that play a role in discovery in memory research, which puts a lot of effort into obtaining empirical descriptions of the research objects and the phenomena that pertain to them. As an example of such descriptive questions, recall the discussion in section 2.2 about duration and capacity as descriptive features of working memory.

I suggest that, in spatial memory research, too, there are a host of unresolved descriptive questions of the following kinds: How literally should the metaphor of the spatial map be interpreted? Are spatial maps more like images or like architectural models? Are they accurate? Are they consistent? Insofar as there are distortions in spatial memory, what are these distortions caused by? What are cognitive reference points for spatial information, and how can they vary by context? Do spatial maps encode perspective, or do they represent a bird’s-eye view? These are questions that have been raised in the spatial knowledge literature in the past 20 years or so. In this vein, one psychologist has argued that “despite its considerable appeal, as traditionally used, the ‘cognitive map’ metaphor does not reflect the complexity and richness of environmental knowledge. That knowledge comes in a variety of forms, memory snippets of maps we’ve seen, routes we’ve taken, areas we’ve heard or read about, facts about distances or directions” (Tversky Reference Tversky, Frank and Campari1993, 21).

This quotation expresses the sentiments I have tried to argue for in this article, as it does not talk about mechanisms or explanations. Rather, the author raises descriptive questions about a research object that is not very well understood or that is, as I have called it in this article, epistemically blurry. While I cannot pursue this case study here, I suggest that it offers a glimpse into the kinds of descriptive questions psychologists often concern themselves with, raising philosophical issues about the ways in which experimentally produced phenomena are used in their research.