1 Introduction

Once upon a time philosophy of science was all about physics. Debates about laws of nature, deductive nomological explanation and theory change all grew up around questions that occur naturally when considering physics, and were retrofitted for the biological and social sciences. The debate over scientific realism centred on the question of the existence of unobservables, one that occurs naturally for physics and chemistry, but seems irrelevant to many branches of biology. Nowadays, philosophy of science is as much about the special sciences as it is about physics. A consequence of this has been the downfall of D-N explanation, and the rise of mechanistic explanation. Should we think that there is no realism debate for most of the special sciences, because there are not enough unobservables? And that in these domains realism wins by default? No, what we need to do is reorient the realism debate so that we can better see what the issues are for the special sciences. At the heart of the realism debate is a basic curiosity about whether the best science tells us how the world really is? This question can meaningfully be raised across the board, as I will show in the course of this paper, using neuroscience as my representative special science.

My opponent is the traditional scientific realist. I will refer to the definition of Psillos (1999, xvii), with its helpful delineation into three claims that together constitute scientific realism, as usually understood:

The metaphysical stance asserts that the world has a definite and mind-independent natural-kind structure.

The semantic stance takes scientific theories at face-value, seeing them as truth-conditioned descriptions of their intended domain, both observable and unobservable. Hence, they are capable of being true or false. ….

The epistemic stance regards mature and predictively successful scientific theories as well-confirmed and approximately true of the world. So, the entities posited by them, or, at any rate, entities very similar to those posited, do inhabit the world.

At the heart of my quarrel with traditional realism is its claims for the mind-independence of the kinds described within successful science. Mind-independence effectively means human-independent: it is the assertion that at its best, scientific knowledge transcends the relatedness of knowledge to the scientists’ circumstances of discovery and technological motivations, and achieves something more like a God’s eye view of the natural world. Acceptance of the idea that science does not transcend the human perspective has been described by Massimi (2018) as the thread that connects perspectivist alternatives to traditional realism.

An important early presentation of perspectivism came from Ron Giere (2006b), who employed an analogy between scientific knowledge and colour vision. According to relationist theories,Footnote 1 colours are perceiver-dependent properties, which means that colour vision does not present objects to us as they are independently of perceiving humans. Likewise, science is said not to yield knowledge of nature as it is independently of human practical activities and conceptual frameworks. Traditional realism is absolutist rather than relationist about knowledge, and a nice example of this outlook is to be found in Peter Vicker’s recent use of the colour example. He tells us that strawberries are actually colourless, and that our visual experience is universally and systematically misleading (Vickers, 2023, pp. 8–10). Here absolutism amounts to the claim that if redness exists, it must be a perceiver-independent property instantiated by ripe strawberries, and other similarly coloured items. Instead, relationists hold that colours exist, and that our visual experience is not systematically misleading, but that we must accept that colours are perceiver-dependent properties. The wrong way to think about human perception is to take it as aiming to reveal to us how things are in themselves, instead of in terms of their human-relative properties.Footnote 2 The same goes for science.Footnote 3 The variety of perspectivism called haptic realism (Chirimuuta, 2016), is relationist about science: there is scientific knowledge but we must understand this as knowledge of things insofar as those things interact with people and their equipment. The kind of realism I oppose is absolutist about science: it says that science, at its best, gives us knowledge of human independent reality.

This paper has two aims. Firstly, to show that the realism debate can be made relevant to a special science, by turning away from the question of unobservables and towards the question of simplification of complex systems via abstraction and idealization in modelling. If we acknowledge that models vastly understate the complexity of their objects, and include known misrepresentations, can we say that they tell us how the world really is? My second aim is to show that haptic realism is a viable and fruitful alternative to traditional realism. It treats the epistemic aims of science more modestly, and therefore avoids the skeptical reverberations that accompany traditional realism once it is acknowledged how much simplification has to take place in order to build usable models. The next section reviews how consideration of the complexity of objects of scientific research can motivate alternatives to traditional realism. Section 3 presents haptic realism in more detail, while Sect. 4 shows how haptic realism is to be applied specifically to neuroscience. I conclude in Sect. 5 with a summary of points in favour of haptic realism.

2 Complexity and the challenge to traditional scientific realism

As mentioned at the outset, the classic scientific realism debate is physics-centric. It focuses on the question of the existence of unobservable entities that feature heavily in foundational physical theories such as quantum mechanics. What’s peculiar about physics is that it investigates the most simple stuff around: nonliving matter, not undergoing chemical reactions. The stuff under consideration for the physicist is quite homogenous (particles of a certain type are all the same, unlike members of a biological species), unchanging (not subject to growth, plasticity and age), and insensitive to its surroundings. Neurons, the ‘elementary particles’ of neuroscience, are very much unlike this.

Both Catherine Elgin and Paul Teller motivate their Goodman-inspired alternatives to traditional realism by beginning with the point that nature is complicated in ways that outstrip the precise conceptual schemes of scientific theories. This leads to the view that the entities and relationships represented in scientific theories are not mind-independent kinds, but are in some sense actively produced or constructed in the process of doing science. Elgin contrasts traditional realism with what she calls a nominalism:

The issue that divides realists and nominalists is whether—or perhaps to what extent—such scientific work is creation or discovery. Do the scientists draw the lines, or discover where nature draws them? Realists maintain that nature draws the lines. The scientists and other investigators simply find what is there to be found. The problem, though, is that there are too many things there to be found. (Elgin, 2019, p. 523)

The complexity of nature overflows what can be encompassed within a representation interpretable to human scientists. On this I agree with Elgin. One refinement is that I want to resist a dichotomy of creation versus discovery. More will be said in Sect. 3 about why I assert that scientific kinds are not only constructed or created, but also constrained by their objects of investigation.

Teller argues that even in physics—depicted, a moment ago, as the science of simple stuff—the prevalence of idealization shows us that theory is forced to simplify reality:

A system of such idealizations (Newtonian mechanics, classical electrodynamics...) nicely illustrates the idea of a framework in which, often, the idealized statements are treated as true. Such generalizations are taken as true because they are true (would be true) in worlds (if the world were one) in which the idealizations hold, while the gap between such idealized worlds and the real world are irrelevant for present purposes. But usually.… we put aside that we are working within a framework that simplifies the world with the idealizations in question. (Teller, 2021, p. S5028)

By implication, traditional realism is guilty of neglecting the difference between the “idealized world” and the more complex reality that the theory represents.Footnote 4 Importantly, idealization is pursued, and the “gap” ignored because the practical purposes of science (e.g. prediction, isolation of causal dependencies) are better served that way. Philosophically, when we are confronted with the question of how to interpret the knowledge claims of science, bothered by the question of whether things really are as depicted, we fail if we ignore the gap between the abstract and concrete, neglecting the mundane reality of recalcitrant complexity and the attendant human effort that goes in to having to deal with it.

The objects of neuroscience are much more complicated than the objects of physics, which means that the gap between the object of investigation and the idealised representation will be even larger. The central challenge of neuroscientific research is to find effective ways to simplify the brain. The brain is an ultra-hyper-complex object: it is very heterogeneous, its parts are always changing, and it is highly sensitive to context. One definition of complexity, due to Murray Gell-Mann, is that these are systems in between randomness and order: there is pattern there, but never exact repetition (Mitchell, 2009a, pp. 98–99). The weather is like that. So are financial markets. This is a useful way to understand complexity in the brain. The most challenging way that the brain is complex is in its changeability (Chirimuuta, 2020). Exact science needs fixed targets but the brain is constantly changing itself in response to how things are in the body and the rest of world—which is necessarily so for it to have a role in sustaining intelligent behaviour in a complex world. Against the ambitions of traditional realism, we should countenance that each neuroscientific representation will be no more than a rough, rigid caricature of a protean object that can never be fully characterised.

My highlighting changeability (heterogeneity across time) as the critical dimension of neural complexity means that I also highlight the way that the methods associated with particular neuroscientific perspectives are involved with the active production of their targets of investigation, through stabilisation of phenomena. The point is that discussions of perspectivism and complexity should not be restricted to the issue of partial sampling: due to the fact that natural systems comprise so many different entities, properties, and causal relationships (heterogeneity at a time), a given scientific perspective must restrict its view to a subset of these, e.g. by observing and modelling phenomena at particular scales. Both kinds of complexity and notions of perspectives are very important, and teach a lesson that the content of models of complex systems should not be mistaken for presentations of how those systems are in themselves. I emphasise the former issue (temporal heterogeneity and stabilisation of phenomena) because it has had less attention elsewhere, and it is this one that justifies the more radical sounding claims of haptic realism—where it is denied that representations of such complex systems convey to us human-independent facts concerning the natural world. This is to be explained in the next section.

3 Traditional versus haptic realism

In my opposition to traditional realism, I am not supposing that the entities and relations depicted in scientific theories and models are fully constructed or mind-dependent, but just that there is an ineliminable human component in all scientific representations, due to the fact that they do not encompass the full complexity of their target systems. As such, they are the result of human decisions about how to simplify, and these decisions are shaped by the practical demands of the research. Furthermore, the ineliminable human component of scientific representations is due to the fact that these representations are the product of an interaction between human investigators and a target system. I use the perceptual metaphor of touching, haptics, for the investigative process. With touch, the material traffic between sensing organ and object sensed is salient and undeniable. I cannot know an object by touch without in some way making contact with it, and thereby altering it and being altered by it, even if it is the lightest brush. With vision, in contrast, one has the impression of a lack of commerce between perceiver and perceived, which fosters what Dewey called the “spectator” ideal of knowledge (Hacking, 1983): the aspiration to know the world as it is, unaffected by the presence and actions of the viewer. In haptic realism, when I talk of the activity of the scientist that goes into the interaction, I mean both the very concrete processes of preparing objects in laboratories, and the cognitive process of creating and refining the conceptual schemes that yield the most apt and useful simplifications.Footnote 5

Building on the haptic metaphor, a central idea is that scientific models are like hands. Hands are both sensory organs for touch, and our means for manipulating the world. With the sense of touch these two roles are not separable: how we perceive is conditioned by how we desire to engage with the object touched, and vice versa. You touch a peach differently, depending on whether you are trying to pick it or test if it is ripe, and the feeling, the information that you get is quite different, in turn. Likewise, the thought is that in science, how people know is conditioned by how they desire to make changes in the world, and vice versa. Neurobiologists aiming to treat memory loss will develop an account of the hippocampus that differs from a model of the same brain region produced by researchers in a machine learning collaboration, working on artificial cognitive agents. In addition, the scope of practical ambitions will be shaped by the prior history of discovery. Many tools, such as chisels, tweezers and spoons, extend the capabilities of hands in specialized ways. Scientific models should be considered, likewise, as tools that extend and potentiate human sensing and acting capabilities.

Scientific representations are the products both of the activity of scientists, and constraints imposed by objects of investigation. Not any conceptual scheme will effectively aid the kinds of interventions that a scientific project aims at. Acknowledgment of this constraint means that we can deny full constructivism.Footnote 6 Reality comes into the picture as what pushes back against human agency. It is better—more “realistic” in Chang’s (2022) sense—to construe reality this way, rather than as what is absolutely mind-independent. But is this notion of realism too weak, since reality figures merely as something ‘out there’, influencing conceptual schemes but not itself conceptualized and known? I discuss the Kantian dimensions of my position elsewhere (Chirimuuta forthcoming). Here I would like to mention that this position is not anomalous within philosophy of science following the ‘practice turn’, and indeed it chimes with some recent work on natural kinds. Kendig (2016) describes how there has been a break from the tradition of taking mind-independence to be the standard for the reality or objectivity of natural kinds. Reydon (2016, 59) proposes an account in which there is “co-creation” of natural kinds, so that “both the contributions from nature and from us to the making of classifications and kinds should equally be taken into account”. This is analogous to my claims about the status of entities and relationships depicted in neuroscientific models.

3.1 Perspectival realism

As mentioned previously, haptic realism is a variant of perspectivism. Recently, Massimi (2022) has sought to establish perspectival realism as a distinct and credible version of scientific realism. Earlier presentations of perspectivism, like Ron Giere’s (2006a, 2006b), faced criticism for not being realist enough. Since haptic realism also grew from considering Giere’s perspectivism, and probably stays closer to it, it is relevant to say something about this comparison. Giere’s is a construal of perspectivism that Michela now calls perspectival1Footnote 7 and she considers it to have the worrying metaphysical implications that there are no non-perspectival facts, and that all properties are dispositional (or relational) as Chakravartty (2017) has pressed.Footnote 8 Massimi’s account is in the middle of a spectrum between absolutism and relationism (see Sect. 1): it insists that scientific knowledge is bound to the situated human perspective, but at the same time seeks to avoid the more radical implications of relationism and therefore recover a realism in which perspectival representations provide a “window on reality” (2022, p. 33).

However, Giere’s original kind of perspectivism is better motivated if we begin our enquiry with the thought that nature is immeasurably complex, that it is so complex and changeable that scientists must interact with it in careful ways in order to stabilise phenomena,Footnote 9 and get representable facts. This is a way of saying that there are not perspective independent facts—i.e. facts that just hold independently of choices made about how to investigate certain portions of the world. Nature is not, in its entirety, a passive, stable thing, waiting to be mirrored in scientific representations, but in many cases something that actively responds to processes of investigation.Footnote 10 It may seem like a weird, post-modern claim if your paradigm science is classical physics, but in neuroscience it is just stating the obvious. Brains plastically change with the experience of the whole animal or person in the lab, and neurons behave differently when subjected to invasive procedures and examined in vitro. It makes sense to think of all of the brain’s properties as relational, and to forget the question of how the brain is intrinsically, independently of its interactions. My concern about invoking the metaphor of the “window on reality” is that it reverts perspectivism to an account of knowledge in which scientific activity is—at its best—a transparent medium through which a human-independent world can be viewed. The metaphor of touch is intended to break with that ideal: we have knowledge of nature in virtue of our activity, which means that the objects of knowledge are things that are affected by us.

3.2 Standing out against scientific realism and empiricism

Most discussions of scientific realism only consider it in relation to the ‘anti-realisms’ of logical empiricism and instrumentalism. This is unfortunate because it has obscured the important thing that the traditional realist and anti-realist, especially the empiricist, have in common. Their shared starting point is a normative picture of science being an absorption of facts; they differ over whether they restrict the facts to empirical observations (empiricism), or whether they take those facts to be the unobservable states of affairs that are, in the best case, represented by mature theories (scientific realism). It is a conception of the task of knowledge formation in which it is at its best when it is most passive: science succeeds in its epistemic goals when either the empirical data or unobservable reality exclusively shape the theory, so that scientific representations can mirror their objects. Haptic realism breaks with this shared commitment and instead emphasises activity and the constructive engagement that brings about knowledge.Footnote 11 The basic point is that it rejects the passivity that goes with the other positions: the knower has to be doing something, engaged with things, for anything to be known scientifically.

One pay-off of this reorientation is that it helps make sense of the puzzle over how it is that models of phenomena can be so successful (yield satisfying explanations, be predictively powerful, and fruitful in the development of new experiments and models), while being so full of “distortions” such as idealizations and grossly simplifying abstractions. Philosophers like Batterman (2010), Bokulich (2012), and Potochnik (2017) have frequently argued that these distortions are essential to the models’ explanatory success, and yet it has been hard to envisage any connection between deliberate distortion and somehow getting a better account of the target, when thinking within the traditional constraints. Once knowledge-building activity is acknowledged and emphasised, we can think of models as devices that aim to achieve a certain compromise or balance between a natural system, the scientific (collective) mind, and some material purposes. Explanatory, predictive, and practical success are a matter of achieving the right kind of fit, not of the attainment of a God’s-eye view on the subject. There can be various ways to be successful (a plurality of perspectives), and sometimes the best way to achieve alignment between the target system, human conceptual resources, and the material goals, is through deliberate distortion.

The primary argument in favour of traditional scientific realism, originally put forward by Hilary Putnam, is that every other account leaves the technological successes of science an inexplicable miracle (Psillos, 1999, chapter 4). But this assumes that technological success can only be a downstream consequence of acquisition and application scientific theories that represent mind independent states of affairs to a close approximation. It is in fact question begging, since the understanding of the connection between practical efficacy and knowledge of the mind-independent world is precisely what is at stake between traditional and haptic realism. Given human cognitive limitations, haptic realism contends, practical efficacy can only be achieved by strip** back the complexity that is there in nature, prior to human efforts to simplify it, and not by most closely approximating it. Technological success is attributed to the achievement of an adequate fit between the constraints stemming from the parts and processes at work in a target system, and those due to the limitations of human cognition and know how. Technological success is achieved more often than not by simplification and deliberate distortions, and for this reason should not be taken as dispositive evidence that the theory employed has encapsulated some absolute truth about the workings of nature, as the no miracles argument supposes.

4 Neuroscience and realism

In this section I will show how the contrast between traditional and haptic realism is to be articulated when we focus on neuroscience. This also shows how the realism debate works when we consider the theoretical claims of our representative special science, neuroscience, rather than physics. A large part of the theoretical content of neuroscience consists in computational models of the brain, and there is not a clear theory/model distinction, as there is in physics. Therefore, when showing how the question of realism applies to neuroscience I will mostly focus on computational models as the equivalent to theories in the realism debate for physics. Following Psillos (1999) it is customary to think of the realism debate as divided up in to three strands, an ontological, semantic and epistemic one. I will discuss each of these in turn.

4.1 The ontological strand

The ontological question is whether at least some of the entities and relations depicted in the best neuroscientific models have a mind-independent existence. The answer may seem obviously, ‘yes’, vindicating traditional realism, but building on Teller’s (2021) account of the trade-off between precision and accuracy in scientific representation, we can see that the situation is not so clear. Teller’s point is that no useful conceptual framework (such as a taxonomic scheme) can handle all of the complexity that is there in nature, in a maximally precise way. A taxonomic scheme can enhance its accuracy by being less precise in what it states about the entities, effectively blurring out its view on things. Once we appreciate that even the best representations of what there is in the brain are imprecise and simplified, the realist hope that those representations simply depict the mind-independent entities starts to fade. Haptic realism is attractive because it recognises that the entities posited in scientific representations come to us through the medium of the scientists’ classificatory work, but still they are not simply made up.

To see the relevance of these points, consider that all neuroanatomy is an abstraction. Not only is it partial, capturing some structures but not others, but more significantly it employs a classificatory framework which must abstract away from the differences between members of a class of cells (e.g. Purkinjie cells in the cerebellum), and overstate their similarity. Part of the complexity of the brain is that no two neurons are exactly alike in their elaborate branching forms, and these structural differences are probably functionally important, and yet they get treated as if they were not there. Each cell class is something of a neuroanatomist’s ‘ideal type’—like any idealization it is not a denizen of concrete reality. While the old debate about the existence of unobservables seems spurious in neuroscience, when we are dealing with many structures visible to the eye or under light microscopy, we should recognize that the important question remains, over whether neurons exist, as catalogued and conceptualized by neuroanatomy. Here, the haptic realist denies that the scientific schema is at its best a report of an order that is there in mind-independent reality, as opposed to a useful rendering of an excessively complex reality.

We can also consider what is at stake in computational neuroscience. The question is whether the computations or principles of neural coding, depicted by the best models, exist independently of the simplifications that come about through experimental practice and through modelling techniques such as idealization. Consider a deep convolutional neural network model of visual processing in the ventral stream of the primate brain (e.g. Yamins et al., 2014). Note that the scientists referenced do not claim that they have arrived at a highly accurate model of computations in this brain area. But the realist idea is that as models improve, this improvement consists in the model converging on the actual computations performed by the primate visual system. Many suppose that the structure represented in the model mathematically as a computation is there in the brain activity as a series of physical state transitions, to the extent that the model is accurate. The traditional realist invokes a structure in common (a homomorphism) claimed to exist in model and target, a view I have called formal realism (Chirimuuta, 2021). This stands in contrast to an alternative way to cash out the ontological status of the model called formal idealism. According to the latter view, which fits with haptic realism, whatever processes exist in the brain are vastly more complicated than the relations represented in the computational models. Indeed, all computational models can be treated as idealizations of physical systems. The aim of neurocomputational modelling is to achieve an acceptable simplification of the brain processes, which helps the scientists to achieve their explanatory, predictive, and technical goals. Thus, the success of the research is more a matter of structuring a much too complex set of processes than of discovering pre-existing structures. A successful model is a compromise between the vast complexity of nature and the limitations and demands of human enquiry.

4.2 The semantic strand

We will now move on to the semantic strand. According to the formal realist, neural computational models should be interpreted literally as attempts at representation of a real order hidden behind the manifest complexity of neural anatomy and physiology. That is, as representations of computations actually performed by the brain. The formal idealist holds, in contrast, that neural computational models should be interpreted analogically: there is a constructed similarity between brains and computations, that provides neuroscientists with a useful simplification of the brain. Accordingly, the content of the computational model is an ideal pattern. This pattern is to some extent dependent on the activity of the scientists and is not a structure or phenomenon that exists in the brain independently of experiment, data processing, and data analysis.

The ideal pattern helps scientists make sense of brain data, by offering them an analogy with processes that occur in artificial computers. Thus neural computational models should not be interpreted as literal claims about the neural system implementing those computations.Footnote 12 This account is influenced by Marry Hesse’s classic work on analogies and models in science. Scientific analogies domesticate what is incomprehensible in nature by fixating on similarities with what is to the scientist more familiar and better understood (Hesse, 1955, p. 353). A basic, first pass way of thinking about analogies is to take the similarities highlighted by models to be pre-existing facts about the objects of investigation. Comparing two objects, some properties will be shared and others will not be. You might picture a Venn diagram where two classes of properties, each associated with one of the objects, partially overlap. The greater the extent of the overlap, the greater the number of analogical inferences one may draw, and so the more revealing the comparison will be.

Hesse’s early account of analogy did in fact posit an identity of structure between an analogy source, the model system, and the analogy target (Hesse, 1962, p. 22). This account of analogy presupposes a formal realism. My account rejects formal realism, and so the treatment of analogy will diverge from this one. I do not assume that epistemically appropriate, useful scientific analogies depend upon there being pre-existing structure in common, a homomorphism, between the model and the target system. As discussed above, determining structure in the target of investigation is an active, and to some extent constructive business. Complex systems do not come pre-structured, ready to be depicted in the model.

My position is consistent with some of the themes from Hesse’s later work. She concurs with Gadamer’s rejection of formal realism, in his “assumption that no ultimate order can be apparent to finite minds,” which motivates the position, “that there is a fundamental inexactness of all human knowledge,” and therefore that the commonly held ideal of scientific language being univocal, literal, and maximally precise, is misguided (Hesse, 1995, p. 363). Instead, Hesse argues that all language, including scientific language, is rooted in the figurative rather than the literal (1994, p. 453), and can never completely abandon metaphor and analogy. This does not prevent it from conveying knowledge (Hesse, 1995, p. 352), or its being the medium for rational argument (Hesse, 1994, p. 453). What it does stand against is the literal interpretation of scientific models and descriptions, and the tendency to take scientific accounts of the world as providing a purified record of the plain facts. In this way, Hesse’s later work attacked the common idea that there is a sharp separation between literal and figurative language, the former being factual and the latter expressive. This picture of scientific language, as imbued with analogy and metaphor allows for a richer understanding of neuroscientists’ use of everyday psychological terms, as well as terms borrowed from the technical domain of computer science, and has profound implications for how philosophers should interpret this research.

4.3 The epistemic strand

Finally, we come to the epistemic strand. The traditional, formal realist asserts that there is knowledge of the brain to the extent that the entities and relations depicted in simplified neuroscientific models correspond to a real order buried in the manifestly complex neural system. In contrast, the haptic realist holds that there is knowledge of the brain to the extent that scientists are able to strike a compromise between the unfathomable complexity of the brain, and their limited human capacities, develo** theories and models that allow them to solve problems at hand (including biomedical applications) and advance their research. The emphasis is on the practical utility, and indeed necessity, of simplification in the process of knowledge. In a Kantian vein, an implication of this position is that the Brain-in-Itself (in its full complexity) is not an object of knowledge.

We should not be over-concerned by this implication. It is another way of expressing the perspectivist thought that the absolute, God’s eye view of nature cannot be attained by limited human beings. Especially because of the mismatch between the complexity of the world and number of variables and interactions that make sense to us, we should accept that the need for simplification gives a human dimension to all scientific representations. Thus, what we know is how things are for-us—and of course through technology, the range of accessible phenomena keeps expanding—not as they are and would be entirely independent of our actions and conceptualisations. Hasok Chang calls this a shift towards a humanistic epistemology:

instead of taking ‘natural’ as something transcending humans and seeking forever to erase all traces of humanity in our concepts, we can embrace the humanity of our concepts and assess their merits in terms of how well they enable human scientific inquiry. Nature conceived in the anti-humanist manner is impossible for us to access, though the aim of learning about such ‘nature’ may make sense as a sort of forlorn regulative ideal worthy of Sisyphus or Tantalus. (Chang, 2016, p. 43)

Here, Chang is dismissing the goal of attaining knowledge of the human-independent world, even as a regulative idea guiding researchers. This is a point at which I might disagree. Chang aims to give an account of scientific knowledge that is useful not only to philosophers and broader communities, but to scientists themselves. The suggestion is that by giving up on absolutist ideals, and thinking differently about their acquisition of knowledge, scientific communities will be more empirically active and pluralistic, and therefore more successful (2022, p. 7). My proposal of haptic realism for neuroscience is broadly motivated by the question of how philosophers should interpret neuroscientific knowledge claims for the purposes of doing philosophy of mind. The question of interpretation is also important for non-academic outsiders to science, and scientists when they popularise their work, but I am not treating it as something that needs to, or should even, impact the working scientist. In other words, I have not developed this account in order to prescribe it to scientists themselves. Haptic realism is for the better understanding of what the neurosciences (and other special sciences) tell us about the world, but not necessarily for neuroscientists. I suspect that scientists themselves might be better off sticking with absolutism rather than buying into the relationist claims of haptic realism. It is possible that the search for absolute truths about nature, and adherence to the anti-pluralist aim of convergence towards the one best theory,Footnote 13 have served scientists by giving them focus and conviction, and providing a heuristic to help them find regularities that are more stable and therefore useful for interventions, amongst all the patterns that are empirically accessible. Having made it clear that the God’s eye view is unobtainable for science, we perspectivists should ask ourselves whether achievement of the God’s eye view was ever really the point: maybe the goal was pragmatic all along.Footnote 14

5 Conclusion: why haptic realism?

The two aims of this paper were to show (1) how the problem of realism is relevant to the special sciences, turning away from the question of the existence of unobservables towards the issue of simplification; and (2) how a reoriented debate pits absolutist traditional realism against a relationist haptic realism. My account of this debate was mostly descriptive. But it is worth considering at the end what reasons there are for replacing traditional realism with haptic realism.

Two reasons came up in the course of the paper: the ability of haptic realism to make sense of the idea that models explain because of, not in spite of, idealization; and the point that traditional realism is committed to the unattainable goal of humans achieving human-independent knowledge of nature. In addition, it was argued that the central positive argument for traditional realism (‘no miracles’) is question begging against haptic realism, because it presupposes that technological success is due to the attainment of knowledge of how things are independently of their relations to humans, a point denied by the haptic realist, who instead sees applied science as the result of finding an adequate compromise between nature’s complexity and scientists’ ability to understand, stabilise and utilise a certain number of dependencies.

I will now present one more argument in favour of haptic realism, returning to the analogy with colour vision. In the debate over colour realism, the traditional realist’s claim that colours are perceiver-independent properties of objects elicited a counter-reaction from anti-realists perturbed by the undeniable pervasiveness of the activity of the perceptual system in sha** how we see things in colour, and the lack of good grounds for identifying perceiver-independent, physical properties with the properties that colour vision presents us with (Hardin, 1993). Anti-realism is an error theory: it tells us that colour perception lands us with a lifelong, systematic illusion or misperception of reality. In this context, colour relationism is a voice of sanity: it diagnoses the problem as stemming from the realists having set up an inappropriate goal for colour vision, that of presenting to us the perceiver-independent spectral properties of objects, as opposed to a set of perceiver-dependent properties somewhat grounded in physical ones, but tailored to suit the needs of the perceiving creature.

By analogy, the problem with traditional scientific realism is that it opens the way to a pervasive skepticism about scientific knowledge, because it sets such a high standard for what is to count as knowledge: correspondence with how things are in a human-independent, absolute reality. This problem has not been so obvious to people, as with the case of colour vision, because there was lack of full recognition of the pervasiveness of scientists’ activity in sha** their objects of knowledge and conceptual frameworks.Footnote 15 By shifting the realism debate onto the terrain of special sciences of highly complex systems, acknowledging this becomes unavoidable. Neuroscience cannot be held up as giving us the unvarnished truth about the brain, because the brain is so very complex, and the gap between it and the simplified scientific representation is so large. There are countless details of neural (and bodily, and environmental) systems that are plausibly relevant to the explanation of any given phenomenon in this multi-scale, highly interactive organ, and yet these variables are mostly ignored within any one project. A perennial challenge for neuroscientists is to try to figure out which of the details matter most to their particular explanatory and technical projects. All neuroscientific representations are, very obviously, gross simplifications, but the likely response of the traditional realist is to assert that the ignored details can safely be neglected because the simple model isolates just those entities and relations that are, in an absolute sense, relevant to explaining the phenomenon at hand. My point is that there are not good scientific grounds for claiming this. The more the brain is investigated, the more interactions and potential explanatory factors come to light. Thus, relevance is pragmatic and must be relativized to the scientists’ aims. I suspect that traditional realism has harboured an implicit Platonism, a belief in there being a simpler order and reality masked by the observable complexity of nature. When philosophy shifts its focus from the elegant mathematical symmetries of physics to the unruly domains of the special sciences, this becomes a dubious article of faith.