2. Theory avoidance (and how we might avoid it)

Fussing over the neglect of neuroscientific theories is only reasonable if there are, in fact, theories in neuroscience. A theory-in-neuroscience skeptic might call attention to passages where philosophers write that theory plays “little role” (compared to models) in neuroscience (Bechtel and Richardson [Reference Bechtel and Richardson2000] 2010, xvii) or characterize neuroscience as being “data rich and theory poor” (Churchland and Sejnowski Reference Churchland and Sejnowski2016, 667). Even if neuroscience is “theory poor,” sparse coffers are not the same as empty ones. If there are even some theories in neuroscience, it is reasonable to want an account of them. It is also reasonable to interrogate what notion of “theory” is being used when someone proclaims that theories are missing from neuroscience. There are many notions of “theory” (Sellars Reference Sellars, Robert and Marx1965). Neuroscience might appear more theoretically impoverished than it really is if we are working with a notion of “theory” ill-suited for thinking about theory within neuroscience. It would be foolish to dismiss this possibility out of hand. Philosophers of science have a time-honored tradition of using the physical sciences as a measure against which the biological sciences are compared. Adopting this approach has led to such conclusions as: The biological sciences don’t really have laws because these purported laws do not align with our understanding of those in the physical sciences (Smart Reference Smart1968). Alternative conclusions are certainly available. Perhaps drawing comparisons to the physical sciences is simply the wrong way to approach questions of knowledge or representation in the biological sciences (Mitchell Reference Mitchell2000). Biology may have laws but may require a different framework than the one used for the physical sciences to account for such laws.

What philosophers have said about theory in neuroscience is of a similar, discipline-relative spirit. Ken Schaffner has argued that illustrations of theory in the biomedical sciences and neuroscience only partially overlap with those in the physical sciences, the latter exhibiting theories whose structure—that is, the sort of entity that it is (Winther Reference Winther and Edward2021)—might be accounted for on a syntactic conception of theory, where theories are formally specified, deductive axiomatic systems that are universal in scope (Schaffner Reference Schaffner1993). Instead, we more often find instances of what Schaffner calls a “theory of the middle range” (Schaffner Reference Schaffner1993, 321), a theory that is narrower in scope and whose structure is better captured by a semantic conception on which theories specify a class of models. More specifically, Schaffner characterizes theories as a “series of overlapping interlevel temporal models” (Schaffner Reference Schaffner1993, 320) and, in doing so, highlights the existence of nonformal components. Subsequent work has continued to emphasize the nonformal components of theory in the biological sciences. Carl Craver, for example, has argued that many illustrations of theory in neuroscience are mechanistic in nature, where mechanistic theories are “abstract and idealized descriptions of a type of mechanism”—that is, “mechanism schemata” (Craver Reference Craver, Machamer and Silberstein2002, 69).

Both Schaffner and Craver can be read as advocating a type of pluralism. Theories may have formal components, nonformal components, or both. While their arguments for pluralism and the inclusion of nonformal components are well taken, I will later argue that plausible instances of neuroscientific theories fall outside the scope of either Schaffner or Craver’s accounts of structure. Even if my arguments fall flat, structure is only one aspect of theory. A further aspect of theory, one that is commonly discussed for theories in the physical sciences (Boyd and Bogen Reference Boyd, Bogen and Edward2021) but that neither Schaffner nor Craver explicitly address, is theory interpretation.

One understanding of what theory interpretation is likens it to the translation of a foreign text (Dewar Reference Dewar2023). A theory is imbued with meaning when we put the components of the theory into correspondence with symbols whose meaning has already been specified. This way of thinking about interpretation is based on the way this process works in the physical sciences, where theories are often formally specified in a way that the syntactic or semantic conceptions of theory may capture. As indicated in the preceding text, nonformal components are likely more characteristic of theories in the biological sciences, including neuroscience. Is there a comparable need for theory interpretation when considering theories with nonformal components? A formally specified theory is a calculus whose symbols lack significance prior to interpretation (Dewar Reference Dewar2023). But the symbols on a mechanistic diagram, for example, do seem to carry significance—for example, “→” represents direction and “Na⁺” represents sodium ions.

Even if some nonformal components come equipped with already specified meanings, I argue that a need for interpretation still remains. One function of interpretation is to specify the commitments of a theory (Dewar Reference Dewar2023). If the commitments of neuroscientific theories are not completely transparent, there remains a plausible role for an interpretive project. Of course, such differences in the types of theory components may require a new consideration of what interpretation is in this context. For example, some might want to treat the interpretation of nonformal components as a type of interpretation*. So be it. What matters is whether there is some plausible notion of interpretation—or interpretation*—that remains relevant. In the next section, I show that the neuron doctrine admits of a variety of interpretations, making the case that the interpretive aspect of theory is still of central importance for neuroscience.

The reader may recall that my goal is to identify those aspects of theory that need accounting for to be able to put theory in its place—that is, situate theory in relation to extratheory aspects of neuroscience. My inclusion of an explicit interpretive aspect is an improvement over previous treatments that focus more exclusively on structure, yet this addition is not sufficient to accomplish the task at hand. An account of the functional role of theory in neuroscience is also needed.

Thinking about the function of various aspects of scientific practice—for example, scientific definitions (Knuuttila and Loettgers Reference Knuuttila and Loettgers2017)—is widespread. But the function of theory in particular has received more minimal explicit attention, typically from those who advocate a pragmatist view of theory (Winther Reference Winther and Edward2021). Some of this lack of attention may be due to an assumption that an account of the function of scientific theories is a misguided endeavor. Drawing a parallel to arguments against the plausibility of accounting for the uses of language on the grounds that there is an infinite number of uses to which language can be put (Cappelen Reference Cappelen2018), we might think that theories likely have so many potential uses that to account for them is a pointless exercise. It is likely the case that scientific theories can be used for all sorts of things and that trying to catalogue all these uses would be pointless. But, taking inspiration from those who have not been dissuaded from thinking about linguistic function (Thomasson Reference Thomasson2024), this myriad of uses does not entail that any account of theory function is misguided. Given a notion of function on which the function of some x is not equivalent to “any and every use to which some x may be put,” we may have some success in specifying a range of functions that neuroscientific theories can play.

3. Meet the neuron doctrine(s)

I now turn to the neuron doctrine as a case study for further illustrating how we should be thinking about neuroscientific theories.Footnote ² All statements of the doctrine portray the neuron, or neuronal activity, as some fundamental structure and/or functional unit. Yet, as I will discuss at some length, there are varied interpretations as to what this means. Consider the following examples: “The neuron doctrine represents nerve cells as polarized structures that contact each other at specialized (synaptic) junctions and form the developmental, functional, structural, and trophic units of the nervous system” (Guillery Reference Guillery2005, 1281); the doctrine claims that the neuron is “the structural and functional unit of computation” (Saxena and Cunningham Reference Saxena and Cunningham2019, 103); the doctrine claims that neurons are “the relevant functional units in the brain” (Cao Reference Cao2014, 891); the doctrine claims that “the only functionally relevant aspect of the brain, what gives it its behavioral capacities and gives us our mental states, is neuronal activity” (Cao Reference Cao2022, 3). These interpretations are not equivalent. Being the unit of computation (Saxena and Cunningham Reference Saxena and Cunningham2019) does not necessarily entail that only neuronal activity is functionally relevant to our behavior and mental states (Cao Reference Cao2022). Charting the dimensions along which interpretations vary is one goal of this section. A briefer discussion of the structure of the doctrine and potential functions bookends this interpretive analysis.

There are two reasons why I have chosen the neuron doctrine in particular to illustrate the limitations of current treatments of neuroscientific theories, and the sorts of questions that we will encounter as we try to improve and expand upon these accounts. The first reason is the recognized centrality and prominence of the doctrine within neuroscience. The neuroscientist Marina Bentivoglio writes, “Will there ever be another doctrine as fundamental as the neuron doctrine in what are now the neurosciences and cell biology in general?” (Bentivoglio Reference Bentivoglio and Gordon2016, xvii). Similarly, the neuroscientist Rafael Yuste writes, “For over a century, the neuron doctrine…has provided a conceptual foundation for neuroscience” (Yuste Reference Yuste2015, 487). The second reason is that, despite its prominence, the doctrine remains largely overlooked by philosophers. The most notable exception is Rosa Cao, who has argued that it is time for neuroscience to leave the doctrine behind (Cao Reference Cao2014, Reference Cao2022).

To emphasize once more the nature and scope of my aims, an evaluative project like Cao’s is not the kind of project I am engaged in here. Whether the doctrine is a theory worth keeping around is a different question from whether the doctrine is a mechanism schemata. It is likely unnecessary to have a full account of the structure, interpretation, and function of the doctrine in view before we embark on an evaluative project, yet some understanding is necessary. Structure influences the relevant notion of interpretation and a central role of interpretation is to specify the commitments of a theory. These commitments are certainly relevant to an evaluative project. Just as we need a better understanding of theory before we can situate it in relation to extratheory aspects of science, we need a better understanding of the doctrine before we can evaluate it. After a brief discussion of structure, I return to the interpretive issue.

Treatments of theory structure in neuroscience have tended toward pluralism and have emphasized nonformal theory components. It seems plausible that some neuroscientific theories are a series of “overlapping interlevel temporal models” that bear a resemblance relation to each other and are narrow in scope, as Schaffner suggests. Or, that some theories are abstracted and idealized descriptions of mechanisms, as Craver suggests. But the doctrine does not appear to be the type of theory that either Schaffner or Craver had in mind when they developed their accounts of structure. For one thing, the doctrine is not narrow in scope. Its scope is not restricted to specifying a type of structure-function relationship that only holds for some organisms but that bears resemblance to a structure-function relationship found in other organisms. For another thing, the doctrine is not a mechanism, abstract or otherwise: It is not interlevel, lacks temporal stages, and there are no set-up or termination conditions (Machamer et al. Reference Machamer, Darden and Craver2000). I do not want to entirely rule out any possibility of these accounts being modified such that they could account for the doctrine. The main takeaway is that, even if this is possible, the necessary work has yet to be done.

Returning to the analysis of interpretation, I suggest that interpretations of the doctrine can vary along the following dimensions: (1) structure versus function, (2) lower versus higher functions, and (3) causal versus methodological notions of fundamentality.

Structure Versus Function: Interpretations of the doctrine differ in their emphasis on its structural or functional components; these differences potentially change the theory’s commitments. For example, the introduction and initial interpretation of the doctrine near the end of the nineteenth century was predominantly structural, a claim about the tissue in the nervous system being composed of discrete nerve cells (Shepherd Reference Shepherd2016). The Spanish anatomist and histologist Santiago Ramón y Cajal (1852–1934) was among the most vocal proponents of this theory, defending it against the so-called reticular theory on which nervous tissue is composed of a continuous network of neurons, the view championed by Cajal’s rival, the Italian histologist Camillo Golgi (1843–1926) (Ramón y Cajal Reference Ramón y Cajal1906). Interpretations of the doctrine that foreground structural concerns, like Cajal’s, have functional implications (Guillery Reference Guillery2005). In contrast, interpreting the doctrine as committed to the neuron being the unit of computation in the nervous system (Saxena and Cunningham Reference Saxena and Cunningham2019), for example, foregrounds functional commitments in a way that the interpretation in the Cajal–Golgi debate does not. This leads to the second dimension along which interpretations may vary, the nature of the functions in question.

Lower Versus Higher Functions: At the risk of stating a triviality, the nervous system is thought to be subserving an enormous range of functions, from the motor to the sensory, the cognitive to the conative, the organizational and integrative to the nutritive, and so forth. Philipp Haueis draws a useful distinction between cognitive and noncognitive functions supported by neural activity, these notions identifying the ends of a continuum of neural functioning rather than discrete categories (Haueis Reference Haueis2018). The former pertains to more complex “functions that process behaviorally relevant information,” the latter to functions that “do not process behaviorally relevant information but maintain and repair neural and other systems of the body” (Haueis Reference Haueis2018, 5373). While this is a rough distinction in need of further explication, it is sufficient to note that different interpretations of the doctrine appeal to different notions of function. For example, Cao appeals to behavioral and mental functions (Cao Reference Cao2022), Horace Barlow to sensory and perceptual functions (Barlow Reference Barlow1972), Sir Charles S. Sherrington to lower order—what Haueis might consider “noncognitive”—functions (Sherrington Reference Sherrington1906).

Distinguishing interpretations along the dimensions of structure versus function or lower versus higher functions does not, however, illuminate the notion of fundamentality at play. While specifications of structure-function relationships are widespread in neuroscience, the structure-function relationship in the doctrine is unique insofar as it is regarded as fundamental in some (largely uninterpreted) sense. To put it glibly, no one is championing an amygdala doctrine as the “conceptual foundation” of neuroscience. Multiple interpretations of fundamentality are available. Here, I will limit myself to discussing two: causal and methodological. While such distinctions have been drawn in other contexts—for example, adaptationism (Godfrey-Smith Reference Godfrey-Smith, Orzack and Sober2001)—they have not been applied to the doctrine.

Interpretations of Fundamentality: On the causal interpretation of fundamentality, the doctrine involves an empirical commitment to the structure of the biological world being a certain way, a commitment to the neuron as the most causally important unit in the nervous system for the relevant function(s). On the methodological interpretation of fundamentality, the doctrine involves a commitment to a neuron-centric approach to the study of the nervous system, a way of structuring or organizing investigations within neuroscience, rather than an empirical claim about the causal structure of the nervous system.Footnote ³

As Peter Godfrey-Smith notes in his application of similar distinctions to the adaptationist debate, these interpretations are logically distinct (Godfrey-Smith Reference Godfrey-Smith, Orzack and Sober2001). Still, in actual practice, the causal interpretation might have some bearing on the methodological one. The latter may involve some implicit empirical commitments as to the causal structure of the nervous system, as this causal structure must have some relevance to one’s evaluation of whether a neuron-centric approach is worthwhile. At a minimum, the epistemic payoffs of a neuron-centric approach seem partially dependent on facts about the nervous system. Even so, the empirical commitments on the methodological interpretation are more minimal than those on the causal interpretation. The methodological interpretation may also take into consideration epistemic aims and interests, cognitive and technological limitations, and so forth in a way that the causal interpretation does not.

To illustrate, consider the sort of evidence drawn upon in criticisms of the doctrine. In recent decades, findings from experimental work and computational modeling indicate that astrocytes (a type of nonneuronal brain cell) and populations of neurons, respectively, may make greater contributions to information-processing in the nervous system than previously thought (Cao Reference Cao2014, Reference Cao2022; Yuste Reference Yuste2015). Growing evidence suggests that astrocytes play an important role in synaptic transmission using their ability to both release neurotransmitters into, and reuptake them from, the synaptic cleft (Araque Reference Araque2008; Araque et al. Reference Araque, Parpura, Sanzgiri and Haydon1999). Neural network models that analyze neural dynamics distributed across an entire population of neurons challenge the idea that a single neuron is the fundamental unit of structure and function (Barack and Krakauer Reference Barack and Krakauer2021; Ebitz and Hayden Reference Ebitz and Hayden2021; Yuste Reference Yuste2015). Criticisms like these are relevant to, and problematic for, some interpretations of the doctrine—for example, a causal interpretation of fundamentality where the relevant functions of a neuron are taken to include higher-order cognitive and behavioral ones (Cao Reference Cao2022). Alternative interpretations of the doctrine, such as the methodological interpretation, may not be equally affected. Suppose astrocytic processes are centrally involved in nervous system functioning. This does not preclude the possibility that organizing our investigations around the neuron may still best facilitate our (to-be-specified) aims.

Even so, one might question whether the doctrine under the methodological interpretation should still be considered a type of theory. One avenue through which to approach this question is a consideration of theory function. Can the doctrine, under the methodological interpretation, play the sort of functional role that we expect of neuroscientific theories? To answer this, we would need an account of theory function. As with structure and interpretation, I aim to explore the sorts of questions that we will encounter as we develop a framework for thinking about the function of neuroscientific theories.

The reader may recall that whatever notion of theory function we settle on cannot be a notion on which the function of theory is equivalent to how a theory is used on any given day by any given scientist. This is compatible with aiming to develop an account that captures a plurality of roles that theory may be thought to play—for example, explanation, prediction, control, representation, unification, posing research questions, identifying mechanisms, experimental design, fitting models (Morrison Reference Morrison2007; Winther Reference Winther2012, Reference Winther and Edward2021). A list of functions is far from illuminating. A more promising approach is to try and identify higher-order categories of functions, or macrofunctions, which may reveal underlying commonalities and relations among them. As a first pass, we might try to distinguish between backward- and forward-facing macrofunctions. The former, backward-facing macrofunction might include such subfunctions as figuring in explanations or supporting predictions. To play these sorts of roles, a theory must have relatively rich content. The latter, forward-facing macrofunction might include such subfunctions as organizing or structuring current and future investigations, perhaps even ones that are more exploratory in nature. In this role, theories might be thought of as analogous to directives, with the sort of content required being more minimal.

Returning to the discussion of the doctrine through which we entered the topic of function, it would seem that, if the doctrine under the methodological interpretation is going to support any category of function, it will be the forward-facing kind—for example, organizing investigation. The plausibility of this claim hinges on a much more elaborate characterization and defense of the various macrofunctions that theory can play and the relation among them. With regard to the particular question at hand, this would necessarily involve an argument illustrating that the forward-facing macrofunction is not dependent upon the features required for the backward-facing macrofunction. If there is such a dependency, the doctrine under the methodological interpretation would likely be incapable of having any meaningful functional role because it lacks the richer content in virtue of which a theory can support a backward-facing macrofunction. More work is needed to explore this question in neuroscience. However, these live questions about theory function already reveal how the choice of an interpretation of a theory in the first place affects one’s commitments about the doctrine.

4. Conclusion: The theoretical infrastructure of neuroscience

I began this article by noting a relative naïveté regarding theory in neuroscience. In light of the subsequent analysis, are we any less naïve? I hope that the answer is affirmative. I have identified three aspects of theory—structure, interpretation, and function—that we need an account of if we want to make and evaluate claims like: Neuroscientific theories are “secondary to and dependent upon new tool development, both temporally and epistemically” (Bickle Reference Bickle, Bickle, Craver and Barwich2022, 14; emphasis in original). I have discussed limitations of extant treatments of neuroscientific theories. And I have illustrated, and started to explore, some of the questions that we should consider when providing an account of the structure, interpretation, or function of the neuron doctrine in particular, and neuroscientific theories more generally. Are forward-facing functions independent of backward-facing ones? Can the doctrine under the methodological interpretation play a forward-facing functional role? If so, does this entail that the doctrine has significant epistemic value despite the criticisms that have been raised against it?

Rather than continuing to catalogue the questions that we will eventually need to answer, I want to conclude by briefly introducing the metaphor of theoretical infrastructure that I believe holds value as a guide for continuing to think about neuroscientific theories, and perhaps scientific theories more generally. While our stereotype of infrastructure likely features things like roadways, telecommunications networks, or bridges, infrastructure comes in two main varieties, hard and soft. Hard infrastructure refers to the sorts of things I listed just now, physical entities that are necessary for some enterprise to run smoothly. Soft infrastructure refers to the services, frameworks, rules and regulations, systems of governance, and so forth that are sometimes less explicitly identifiable but whose implementation is also necessary for the smooth operation of some enterprise—for example, health care services, social security, institutions for environmental protection, policies governing market competition, standards for safety inspections. Combined, hard and soft infrastructure are the material and conceptual connections that organize and hold things together.

The metaphor of theoretical infrastructure is valuable for at least two reasons. First, the features and functions of soft infrastructure are useful guides for further developing the notion of a forward-facing macrofunction, and the features needed to support this function. Thinking about soft infrastructure may help make explicit the legitimate, and perhaps independent, role of a framework or policy implemented for the purpose of directing, organizing, and structuring an enterprise. Second, the relation between hard and soft infrastructure suggests a way of thinking about the relation between theory—that is, soft/theoretical infrastructure—and the extratheory, material aspects of scientific practice—that is, hard/material infrastructure.

As a very tentative sketch, the infrastructural metaphor may help us consider ways in which theory might be integrated into frameworks developed for thinking about extratheory, material aspects of scientific practice. Hans-Jörg Rheinberger, for instance, has spent decades developing a framework for thinking about the biological sciences on which these sciences are oriented around an “experimental system,” comprised of what Rheinberger calls “epistemic things” and “technical objects” (Rheinberger Reference Rheinberger1997). Epistemic things, the targets of scientific inquiry, are embedded within, and constrained by, a wider arrangement of “instruments, inscription devices, model organisms, and the gloating theorems or boundary concepts attached to them” (Rheinberger Reference Rheinberger1997, 29). Utilizing the infrastructural metaphor, these so-called technical objects might be repackaged as hard/material infrastructure. But hard/material infrastructure is not enough for complex systems to run smoothly, soft/theoretical infrastructure is also needed. Even on this breezy and preliminary discussion, the metaphor might be suggestive of a more nuanced view on which putting theory in its proper place means not minimizing it but integrating it with other aspects of neuroscientific practice. How might this work? At present, we are still too naïve to say. But I hope to have shown how we can be less naïve about theory and its potential contributions to such infrastructure.