neuroscience_based_critiques_of_mr

Neuroscience-Based Critiques of Multiple Realizability in Computational Functionalism: A Literature Review

Audio

Listen to this paper

Introduction

Multiple realizability (MR) is a central principle in philosophy of mind and cognitive science that underpins computational functionalism. In simple terms, MR is the idea that the same mental state or cognitive function can be implemented by different physical or biological systems. For example, the experience of pain or the function of memory could—at least in principle—occur in diverse creatures with very different neurobiology, or even in an artificial system like a computer. This thesis was famously introduced by Hilary Putnam in the 1960s and used to argue against strict mind-brain identity theories. If a single mental kind (say, pain) can correspond to many physical kinds (different neural or silicon states), then no one-to-one identity holds between a mental state and a specific brain state ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=In the philosophy of mind%2C,this premise and these arguments)). Jerry Fodor extended this idea in the 1970s to argue that psychology and other “special sciences” have autonomy from lower-level neurobiology, since their laws and generalizations cover phenomena that are multiply realized in different substrates ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Jerry Fodor ,that there are natural kind)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Fodor insists that it is,Multiple realizability thus)). In sum, MR became a foundational premise for functionalism – the view that mental states are defined by their functional or computational roles, not by their material composition ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=physical kinds,and other critics) quickly)). Functionalist models often liken the mind to software running on the “hardware” of the brain, implying that what matters is the abstract program (which could run on various platforms), rather than the specific neural circuitry. This conception encouraged cognitive scientists to focus on high-level computational descriptions of the mind while deemphasizing the importance of neural details. Indeed, nonreductive physicalists in the late 20th century treated multiple realizability as a defense of the autonomy of psychology and cognitive science, claiming that knowledge of the brain might be unnecessary for understanding the mind ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=Hilary Putnam’s original multiple realizability,thesis is the contention that)).

However, over the last few decades—especially in the past 20 years—this once-orthodox view has been challenged by philosophers and neuroscientists who argue that actual empirical evidence from neuroscience does not fully support rampant multiple realizability. A wave of neuroscience-based critiques has emerged, questioning whether mental functions are truly realized in wildly different ways across species or individuals. These critiques draw on findings about the brain’s organization and evolution: for instance, the observation that many cognitive functions are tied to particular neural structures (cortical specializations), that brain mapping reveals common patterns across species, and that even at the molecular level, diverse organisms share conserved mechanisms for neural processes. Such evidence seems to undercut the strong version of MR, suggesting that nature often uses the same neural solutions rather than completely independent ones to implement similar functions. At the same time, phenomena like neural plasticity (the brain’s ability to reorganize after injury or sensory loss) show that some functions can shift to new neural substrates – a potential case of multiple realization that invites careful analysis. This literature review will examine these issues in depth. We will: (1) outline the principle of multiple realizability in computational functionalism; (2) identify prominent critiques grounded in neuroscience that challenge the MR thesis; (3) discuss specific neuroscientific findings – such as neural plasticity, cortical specialization, lesion and imaging studies, and cross-species neural comparisons – that feature in these critiques; (4) analyze the implications of these neuroscientific challenges for the conceptual viability of computational functionalism; (5) highlight the contributions of key figures in this debate, including philosophers and neurophilosophers like William Bechtel, Jennifer Mundale, Lawrence Shapiro, Thomas Polger, John Bickle, and Patricia Churchland; and (6) conclude with a balanced assessment of how strong the neuroscience-based critiques are and what they might mean for future research in philosophy of mind and cognitive science. Throughout, the aim is to clearly articulate the arguments and evidence, with structured headings and supporting citations for each major point.

Multiple Realizability in Computational Functionalism

Before delving into the neuroscientific critiques, it is essential to understand the multiple realizability thesis and its role in computational functionalism. Multiple realizability is the claim that a particular mental state or process can be realized (instantiated or implemented) by many different kinds of physical substrates ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=In the philosophy of mind%2C,this premise and these arguments)). In philosophy of mind, this idea was introduced by Hilary Putnam (1967) and further developed by Jerry Fodor (1974) and others as a rebuttal to the earlier type-identity theory, which held that each mental state type corresponds to a specific brain state type. Instead, MR asserts a one-to-many mapping: a given mental kind (like “pain” or “belief”) might correspond to one neural configuration in humans, another in octopuses, and perhaps a silicon circuit in a computer, all resulting in an equivalent functional state. A classic illustration is the analogy of computing machines: just as the same program can run on different computer architectures (say, an Intel CPU versus an AMD CPU, or even an abacus, as long as the functional relations are preserved), a mental function could be implemented by different neural designs or materials ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=multiple realizability,that characterizing mental kinds as)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Ned Block and Jerry Fodor,explanations of behavior based on)).

This thesis has been enormously influential. It served as a key premise in arguments for functionalism, the view that what defines a mental state is its functional role in a causal network of inputs, outputs, and other states, rather than its internal makeup ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Ned Block and Jerry Fodor,explanations of behavior based on)). If mental states are multiply realizable, then it makes sense to abstract away from biological details and characterize them in terms of computations or information processing. Putnam famously suggested that psychological states could be seen as analogous to states of a Turing machine (a theoretical model of computation), where the focus is on state transitions given inputs and outputs, not on the machine’s physical construction ([Multiple Realizability, Mind and | Internet Encyclopedia of Philosophy](https://iep.utm.edu/mult-rea/#:~:text=By contrast with the Lewis,properties and then construct “idealized”)) ([Multiple Realizability, Mind and | Internet Encyclopedia of Philosophy](https://iep.utm.edu/mult-rea/#:~:text=provides abstract descriptions of real,features postulated by Turing machines)). For example, in functionalism a state like “hunger” might be defined by what it causes (e.g. seeking food) and what causes it (e.g. low blood sugar), rather than by any specific pattern of neuron firings. Computational functionalism in cognitive science embraced this idea, treating the mind as a kind of software: the algorithm or functional organization matters, and it could in principle be run on various hardware (biological or artificial). This view was bolstered by the apparent success of high-level psychological theories and computational models that ignored neural particulars. It also aligned with the intuitions of artificial intelligence research, which often assumed intelligence could be achieved on non-biological platforms.

Importantly, MR was also used to argue for the autonomy of psychology and cognitive science from neuroscience. Since the same psychological phenomena might be realized differently in different brains, it was argued that psychology should focus on formulating general laws at the cognitive level, without needing a one-to-one mapping to neurobiology (which might differ across species or individuals). As Fodor put it, if psychological categories like “memory” or “belief” are multiply realized in the brain, attempts to reduce psychology to neuroscience would face a hopeless task of trying to find neat neural equivalents for psychological kinds ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Jerry Fodor ,that there are natural kind)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Fodor insists that it is,Multiple realizability thus)). Thus, multiple realizability became a bulwark for nonreductive physicalism – the stance that mental states are ultimately physical but cannot be reduced to a single type of physical state. In practice, this meant cognitive scientists often proceeded with a functional taxonomy of mental processes, confident that neuroscientific details, while interesting, were not strictly necessary for explaining cognition.

To summarize, the principle of multiple realizability in computational functionalism holds that mental functions are substrate-neutral: what makes something a pain, a memory, or a decision is the role it plays, not the stuff it’s made of. This principle justified a focus on abstract computational explanations of the mind and implied that minds could exist in creatures or machines with very different physiology. It was a powerful idea that challenged reductionist viewpoints and encouraged the flourishing of cognitive-level theories independent of biology. But how well does this principle stand up against the findings of neuroscience? The next sections review a series of challenges arising from neuroscientific research that have prompted many philosophers to reconsider just how pervasive multiple realizability really is in actual biological minds.

Neuroscience-Based Critiques of Multiple Realizability

Beginning in the late 1990s and accelerating through the 2000s, scholars in philosophy of mind and cognitive science began to re-examine the multiple realizability thesis in light of accumulating neuroscientific evidence. Neuroscience-based critiques argue that the brain’s structural and functional organization shows more uniformity and constraint than the naive MR thesis would suggest. These critiques do not claim that MR never occurs, but they cast doubt on its scope and frequency in the domain of mental phenomena, especially across biological organisms. Below, we identify several major lines of argument and evidence used to question or undermine multiple realizability:

• Cross-Species Neural Homologies: Comparative neuroscience indicates that different species often share homologous brain structures and circuits for similar cognitive functions, rather than each species employing entirely distinct neural solutions ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=because they were likely to,across species%2C and persistence of)). This continuity suggests a common realization of mental functions across species, challenging the idea of radically different realizations.
• Cortical Specialization and Localization: Within a given species (e.g. humans), the brain exhibits specialized regions dedicated to certain functions (vision, language, memory, etc.). Lesion studies and neuroimaging consistently show that specific mental capacities depend on particular neural substrates, implying a one-to-one mapping in practice (at least under normal conditions) rather than interchangeable implementations.
• Neural Plasticity and Reorganization: Findings in neural plasticity (such as brain reorganization after injury or sensory reassignment experiments) demonstrate that some functions can be taken over by different neural areas. At face value, these cases seem to support MR (since a function moves to a new substrate). However, critics analyze whether these are true examples of the same function realized in a different way, or whether the brain region taking over actually comes to resemble the original area’s processing, thus maintaining a common underlying mechanism ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to)).
• Conserved Molecular Mechanisms: At a more fine-grained level, neurobiology has revealed that very disparate organisms share surprisingly similar cellular and molecular machinery for basic processes like synaptic plasticity, learning, and memory. These universal mechanisms (e.g. specific neurotransmitters or genetic pathways) suggest that even when high-level structures differ, the underlying biochemical realization of cognitive functions might be the same across species ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Work with fruit flies%2C sea,statements like the following%2C from)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=insect biologists Josh Dubnau and,Tom Tully)). Such evidence is used to argue that evolution re-uses core neural components, limiting the degree of multiple realization in nature.

We will now explore each of these points in detail, highlighting prominent critiques and studies, and noting contributions of key philosophers and neuroscientists who have shaped the debate.

Cross-Species Homologies and Brain Uniformity

One powerful challenge to the MR thesis comes from comparative neuroscience: the study of different species’ brains. If multiple realizability were true in a broad and pervasive way, we would expect that different organisms might employ very different neural structures to accomplish similar psychological functions. To some extent, this is the intuition behind MR – for example, an octopus and a human can both perceive their environment and learn, yet their brains are anatomically quite distinct. However, philosophers William Bechtel and Jennifer Mundale (1999) argued that the actual practice of neuroscience assumes and finds a great deal of cross-species uniformity in the realization of mental functions ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=The claim of the multiple,of multiply realized psychological functions)). In their influential paper “Multiple Realizability Revisited: Linking Cognitive and Neural States,” Bechtel and Mundale examine how neuroanatomists and neurophysiologists map brain regions and functions across different animals. They point out that neuroscientists routinely use psychological criteria to identify corresponding brain areas in different species, effectively aligning mental functions to homologous neural structures across taxa ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=century,of multiply realized psychological functions)) ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=two noteworthy points%3A ,function and brain mapping research)). For example, early brain mappers like Korbinian Brodmann (1909) compared the microscopic anatomy of mammalian brains (from rodents to primates) and labeled regions with a common numbering system. Brodmann assumed that structurally similar cortical areas across species were likely doing analogous things; indeed, he explicitly sought to advance a theory of function through his anatomical maps ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=areas and developing maps of,55 species ranging over 11)) ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=because they were likely to,across species%2C and persistence of)). He found a constancy in the overall layout of cortex – e.g. primary visual cortex, auditory cortex, etc. – suggesting that mammals share a basic blueprint for sensory and cognitive processing ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=because they were likely to,across species%2C and persistence of)).

Bechtel and Mundale note that this comparative method undermines the notion that psychological capacities are realized in arbitrarily different ways in each species. On the contrary, it “relies on the assumption that there is a common realization of mechanisms for processing visual information across species.” ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(1999%3A 201)) Neuroscience has been successful precisely because researchers expect and find that, say, vision in a cat, monkey, or human relies on an analogous set of neural areas and pathways (like a retina projecting to a visual thalamus and then to a visual cortex organized in columns). If multiple realizability were rampant, such cross-species brain mapping would be futile – one could not use what we know about the monkey visual system to guide understanding of the human visual system. But in practice, that transfer of knowledge works quite well, implying that similar cognitive states are not realized in utterly dissimilar ways, at least within the range of species studied ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=vision demonstrate%2C neuroscientists have successfully,neuroscientific goal has been to)). As Bechtel and Mundale put it, “this project has not been impaired by multiple realization of psychological states; rather, it relies on the assumption that there is a common realization” across species for the functions being compared ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(1999%3A 201)). Thus, they caution against drawing strong anti-reductionist conclusions from the abstract possibility of MR. If the brains of different creatures often use equivalent parts for equivalent functions, psychology may not be as independent from neuroscience as the standard MR argument suggests ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=grains,holding across species are found)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=that kind,octopus eyes—one can appeal to)).

Philosopher John Bickle and others have echoed this point, noting that a guiding principle in contemporary neuroscience is the continuity of underlying neural mechanisms across species ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Jaegwon Kim ,many neuroscience experimental techniques and)). Bickle (1998, 2003) observes that many experimental techniques in neurobiology (from studies of simple organisms like sea slugs and fruit flies, to rodents and primates) are predicated on the idea that findings will generalize across species because of shared biology ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Jaegwon Kim ,many neuroscience experimental techniques and)). For example, researchers study memory in sea slugs or flies expecting that fundamental insights will apply to mammalian memory, which would make little sense if the realizations were completely different. The success of these approaches suggests that psychological kinds (e.g. “learning” or “fear”) are not as wildly multiple-realizable as once assumed ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=match at L455 So Kim,out that even celebrated neural)). In a modus tollens form of argument: if psychological states were truly multiply realized in vastly different ways, cross-species neuroscience should not work as smoothly as it does – but since neuroscience does progress by comparing species, it indicates similar realization of those states ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=match at L455 So Kim,out that even celebrated neural)). As philosopher Jaegwon Kim succinctly put it, science individuates kinds by their causal powers; if two organisms have mental states with different neural bases that confer different causal powers, then those states are not actually the same kind ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro’s dilemma is in the,what makes something a scientific)). By this view, the very idea of a mental kind abstracted from any structure (as MR would have it) is scientifically suspect, because causal roles are implemented via structure.

To illustrate, consider the example of vision across species. Insects have compound eyes and very different neural layouts compared to vertebrates with camera-type eyes. From a broad functional perspective, both provide vision (they enable the organism to see), so one might cite this as a case of multiple realizability of the function “sight.” However, critics argue that at the level of detail relevant to each species’ ecology, the visual processing in insects vs. mammals is not identical – each has unique capabilities and limitations tied to their distinct hardware. If we define the function coarsely as “obtaining visual information,” then indeed multiple systems achieve that. But such a broad characterization may gloss over functionally relevant differences. Lawrence Shapiro (2000) emphasizes the importance of the level of description or grain at which we declare two processes the “same” ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bechtel and Mundale ,output similarities)). Proponents of MR often choose a very coarse-grained description of mental states (e.g. both a frog and a human have “pain” or “vision”), while describing neural states at a fine-grained level, thereby making the mappings look many-to-many ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bechtel and Mundale ,output similarities)). Shapiro and others urge that we must compare like with like: if we hold the taxonomic level constant for mental and neural descriptions, we often find one-to-one correspondences rather than one-to-many ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bechtel and Mundale ,output similarities)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=match at L613 grains,holding across species are found)). In practice, when cognitive scientists align human and animal behaviors, they typically ensure the behaviors are analogous, and then find the neural analogies too. Cases where realizations truly differ tend to be those where the functional equivalence is only superficial. For instance, an octopus learning to navigate a maze and a rat doing the same might both be said to “learn,” but the strategies and representations might differ so greatly (due to their divergent brains) that we might question whether it’s the same psychological kind or two different kinds of learning adapted to different organisms.

In summary, cross-species evidence used by neuroscience shows a pattern of conserved brain organization for shared functions, which challenges the idea that mental states are multiply realized in fundamentally different ways among organisms. Philosophers like Bechtel, Mundale, Bickle, and Shapiro argue that this undermines the strong MR thesis. While multiple realizability may be logically possible, the empirical reality is that evolution tends to reuse and tweak the same components for cognition, leading to deep commonalities. From this perspective, psychology might not enjoy complete independence from biology; rather, cognitive kinds often align with specific biological kinds (or at least families of related kinds). This doesn’t mean there is never any variation – certainly, species have unique adaptations – but it suggests that variance is constrained and often occurs within a common architectural framework (e.g. all mammals rely on a hippocampus for memory encoding). We rarely see totally novel neural solutions for the same problem in closely related species. As one commentator put it, expectations of ubiquitous multiple realization in psychology are “greatly exaggerated.” ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=Instead%2C they are interested in,multiple realization are greatly exaggerated))

Cortical Specialization and Functional Localization

Closely related to the cross-species uniformity argument is evidence for cortical specialization and localization of function within the brain of a single species. If mental functions were highly plastic with respect to their neural realization, one might expect that the brain could implement a given function in many different ways or locations. Instead, decades of neuroscientific research have revealed a remarkable consistency in how functions map to structure in the brain. The human brain (and that of other mammals) is modular to a significant degree: different regions are specialized for vision, hearing, language, memory, emotion, and so on. These specializations are seen across individuals and even, to some extent, across species (as discussed above).

Lesion studies from neuropsychology provide striking examples. Damage to specific brain areas produces characteristic deficits: for instance, injury to Broca’s area (a region of the left frontal lobe) impairs the ability to produce fluent speech (Broca’s aphasia), while damage to Wernicke’s area (left posterior temporal lobe) impairs language comprehension. These findings, dating back to the 19th century, suggest that normal language function is realized in those particular neural circuits – not just anywhere in the brain that might take up the job. Similarly, lesions of the hippocampus reliably produce severe anterograde amnesia (inability to form new memories), indicating the hippocampus is the critical substrate for declarative memory encoding. Patients like H.M., who had both hippocampi removed, did not simply have some other part of the brain step in to realize memory; instead, the function was essentially lost. Such cases imply a one-to-one linkage of cognitive function to neural substrate in the typical adult brain. If MR were true in a strong form within a species, we might expect the brain to be more interchangeable – e.g. perhaps if the hippocampus is gone, another circuit could entirely take over forming new memories. But with certain core functions, that does not happen (at least not without extraordinary interventions). The tight coupling of deficits to specific lesions strongly supports localized realization of those mental capacities.

Neuroimaging studies (using fMRI, PET, etc.) bolster this view by showing that when people engage in particular mental tasks, there are reliable patterns of brain activation. For example, viewing faces consistently activates the fusiform face area in the ventral temporal cortex ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=Recently%2C neuroimaging ,see%2C for)), whereas reading words activates the visual word form area in the left occipitotemporal region. Different individuals show activation in roughly the same locations for the same tasks, and these locations correspond to what lesion studies find as necessary for those tasks. The existence of such function-specific regions (sometimes called “modules”) indicates that the human cognitive architecture is not radically degenerate (i.e. it’s not the case that any random part of cortex could equally well realize face recognition or language). Instead, there is a preferred neural basis for each function, likely reflecting both genetic developmental patterns and efficient specialization. Neuroscientists like Nancy Kanwisher, Lesley Ungerleider, and others have catalogued many such correspondences, and while there is debate about how strict the modularity is, the evidence of repeatable structure–function mapping is robust.

Philosophers engaging with neuroscience have noted that this kind of evidence tempers the multiple realizability idea. If all normally functioning humans use more or less the same neural circuitry for a given cognitive task, then, at least within our species, that mental state is not multiply realized (it’s realized in basically the same way each time). For instance, Patricia Churchland, a philosopher and neuroscientist, has long argued that understanding the mind requires understanding the brain’s specialized circuits, and she has been skeptical of functionalist assumptions that ignore neural details. Churchland suggests that many mental state categories from folk psychology (like “memory” or “fear”) might in fact fractionate into more precise neural kinds once we investigate the brain (a view sometimes called neurophilosophy or even eliminative materialism when it posits replacing folk categories). In her work Neurophilosophy (1986), Churchland pointed out that multiple realizability did not prevent successful intertheoretic reductions in other sciences – for example, temperature is a high-level property that is multiply realized at the molecular level (the kinetic energy of molecules can be realized in gas, liquid, or solid states differently), yet the reduction of thermodynamics to statistical mechanics was achieved ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=help of scientific examples,example is the concept of)). By analogy, the fact that a mental function could be implemented in different neural ways doesn’t guarantee it resists reduction or one-one mapping; scientists might still find bridging laws or deeper commonalities. Churchland also emphasized that as science advances, we often insert intermediate levels of description that connect the higher-level functional account to the lower-level structural account ( Multiple Realizability (Stanford Encyclopedia of Philosophy) %2C Patricia Churchland ,example%2C the physiology of individual)). In neuroscience terms, this means we develop frameworks (like computational neuroscience models, or network theories) that link psychological functions to neural organization, effectively reducing the gap that MR initially seemed to create.

Another philosopher, Lawrence Shapiro, contributes here with the notion of causally relevant properties. Shapiro argues that not every difference in physical implementation counts as a different realization; only differences that affect the relevant functional properties do ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=are relevant to the purposes%2C,As Shapiro notes)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Establishing relevant multiple realizability requires,for that kind of thing)). For example, two people’s brains might have slight anatomical differences, but if those differences don’t change how a cognitive function operates, they shouldn’t be counted as distinct realizations of that function. What matters are differences that produce a different method of performing the function. This concept helps clarify that while each human brain is unique (no two brains are identical down to the synapse), we normally treat them as the same kind of realization for cognition because they share the key causal organization. Shapiro famously gave the example of corkscrews: a metal corkscrew and a plastic corkscrew operate on the same principle (their material difference doesn’t affect their cork-pulling function), so functionally they are not interesting different realizations ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=functional kind can be “multiply,As Shapiro notes)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro then points out that,realizability—of a single functional kind)). By contrast, a corkscrew versus, say, a cork-pulling robot that uses a completely different mechanism would be distinct realizations. Applying this to brains, small individual differences (like one person having a slightly larger visual cortex than another) are like the aluminum vs. steel corkscrew – not a meaningful new realization. Real multiple realization would require fundamentally different neural strategies achieving the same result.

Shapiro’s analysis leads to a dilemma for the MR thesis: if the supposed differences in realization are not causally relevant, then they don’t count as genuine multiple realization (it’s the “same” mechanism in relevant respects); if they are causally relevant (truly different mechanisms), then likely the functional outcome is slightly different too, meaning we’re no longer talking about exactly the same mental state or process ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro then points out that,functional kind with distinct realizations)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=The usual justification for grouping,Shapiro remarks)). Either way, the strong claim of “one mental kind, many physical kinds” loses its force. With respect to cortical specialization, one might say: brains that implement language via the typical left perisylvian network and brains that (hypothetically) implemented language via a completely different circuit would arguably have some differences in language processing. But we don’t observe such radical differences in humans – almost everyone uses the same network – so we don’t have a clear case of the “same” language function realized differently (barring rare exceptions or pathology, which we’ll address under plasticity). Thus, the neuroscientific evidence of consistent localization aligns with Shapiro’s view that shared functional kinds tend to share relevant physical properties, otherwise we demarcate them as different kinds.

In conclusion, the localization evidence reinforces the idea that mental functions map onto specific neural substrates in a uniform way, at least for a given species and set of conditions. This fact constrains multiple realizability: it’s not that anything can realize anything; rather, the brain has an organized architecture where certain structures are intrinsically involved in certain functions. While computational functionalism treated the brain as an implementer of algorithms that could in principle be executed elsewhere, neuroscience shows that the brain’s “hardware” has unique, evolved structures fine-tuned for those algorithms. The implication is that computational functionalism must acknowledge these biological constraints. Purely abstract models of cognition might oversimplify if they assume the function is independent of the implementing tissue. Many philosophers (including Bechtel, Mundale, Churchland, Shapiro, Polger, and others) argue that acknowledging this reality doesn’t eliminate functional description, but it does mean that higher-level theories should be informed by neural details – or at least not assume that neural details are irrelevant.

Neural Plasticity and Functional Reorganization

An interesting twist in the debate comes from neural plasticity – the brain’s capacity to reorganize itself by forming new connections or repurposing areas in response to development, experience, or injury. On the face of it, neural plasticity might be thought to support multiple realizability: if one part of the brain is damaged, sometimes another part can take over the lost function, suggesting that the function is not inexorably tied to a single neural substrate. Indeed, functional reorganization is a key feature of brains (especially developing ones), and there are striking examples in the scientific literature. For instance, in congenitally blind individuals, the visual cortex (which normally processes sight) can be taken over by auditory or tactile functions (such as enhanced hearing or Braille reading touch processing). Such individuals show activation in occipital cortex when reading Braille by touch, implying that “touch reading” is being realized in a cortical area usually devoted to vision. Likewise, stroke patients who lose function in one cortical area can sometimes retrain another part of the brain to assume some of that function through rehabilitation.

One particularly famous experimental example is the cross-modal rewiring study in ferrets. Neuroscientist Mriganka Sur and colleagues (von Melchner et al. 2000) rerouted the visual input pathways of newborn ferrets such that signals from the retina were sent to the auditory cortex instead of the visual cortex. Amazingly, these ferrets developed the ability to respond to visual stimuli using what is normally “auditory” brain tissue ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to)). In other words, the auditory cortex, when fed visual information, reorganized and took on a role in visual processing sufficient to guide visual behaviors ([Visual behaviour mediated by retinal projections directed … - PubMed](https://pubmed.ncbi.nlm.nih.gov/10786793/#:~:text=Visual behaviour mediated by retinal,cortex can mediate visual behaviour)). This seems like a textbook case of multiple realization: the function “seeing” was realized in a different cortical region (auditory cortex) than usual (visual cortex). Advocates of MR could point to this as proof that the brain is quite flexible – it’s the pattern of connections and computations that matters, not the gross identity of the area. If auditory cortex can support vision, then vision is not tied to a singular neural substrate.

However, philosophers such as Shapiro and Polger have analyzed this case and urge caution in interpretation. Polger and Shapiro (2016) discuss the ferret experiment and conclude that it “falls short of multiple realization because the visual and auditory cortexes do not end up doing things in relevantly different ways.” ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to)). What does this mean? Essentially, although the location of the function changed, the way the function was carried out in the rewired ferret might not have been fundamentally novel. In fact, studies found that the auditory cortex in these ferrets developed neural response properties resembling those of a visual cortex (for example, it formed a two-dimensional retinotopic map of visual space, much like a normal visual cortex would). The organization of the auditory cortex was altered by the visual inputs to mirror the functional architecture needed for vision ([[PDF] Cross-Modal Projections from Auditory to Visual Cortices in the Ferret](https://scholarscompass.vcu.edu/cgi/viewcontent.cgi?article=1888&context=etd#::text=These data indicate that cross,between the primary representations)). To the extent that it succeeded, the auditory cortex became, functionally, more like a visual cortex. Thus, one could argue we don’t have a case of the “same psychological function implemented by two inherently different mechanisms,” but rather the new mechanism transformed to adopt the standard strategy. Conversely, to the extent that the auditory cortex remained different (perhaps it processed vision less accurately, or retained some auditory-style processing), the ferrets’ visual ability was limited or not quite the same as normal vision ([Cross-Modal Projections from Auditory to Visual Cortices in the Ferret](https://discovery.researcher.life/article/crossmodal-projections-from-auditory-to-visual-cortices-in-the-ferret/e4faba3b09f73165ab0aa610bbd5d381#::text=Cross,of Visual and Auditory)). In fact, the rewired ferrets achieved only a modest level of visual discrimination – their “sight” was not as effective as a normal ferret’s. This suggests that simply having any cortex take on vision is not as good as having the right cortex. So, the ferret example can be interpreted as showing conditional plasticity but within limits: the brain can reroute functions, especially early in development, but it often does so by making the new area functionally imitate the old area.

Another oft-cited example is the case of sensory substitution in humans. When one sense is lost, the brain can sometimes use alternative input to drive the same perceptions. There are devices that translate visual information into tactile or auditory signals for blind individuals (e.g. a camera that converts images into soundscapes). Users of such devices can learn to “see” via hearing or touch. Does this mean vision (the perceptual experience of a spatial layout) is multiply realized – once by the visual pathway, and once by an auditory pathway? Perhaps, but again the line is blurry. Some studies have shown that using these devices, the visual cortex itself can become active, as if the information from touch or sound is ultimately being fed into the visual areas to produce a visual-like experience. That might indicate that the brain is leveraging the existing visual circuitry, just with a different input channel, rather than creating vision wholly in auditory cortex per se. In other cases, the person may learn a new skill that is functionally similar to vision but phenomenologically different (some argue the qualia or subjective aspect might differ, or it might be a cognitive inference rather than genuine visual perception). These nuances make it challenging to declare a clear verdict of multiple realization.

What these plasticity cases highlight is the need to distinguish functional equivalence from implementational similarity. The brain shows functional redundancy to a degree – especially in youth, it has the ability to reorganize such that if one path is blocked, another can carry the load. But the reorganization often works by recruiting neural tissue that, while not originally designated for that function, can alter itself to perform it, effectively converging on a similar neural solution. It is not always the case that a completely different neural process suddenly solves the problem in a novel way. Moreover, plasticity has its limits: in adults, if a critical area is destroyed, recovery is often partial at best, indicating that some functions simply cannot find an alternate realization easily after development.

Philosophers Thomas Polger and Lawrence Shapiro argue that many purported cases of multiple realization, when examined closely, do not meet the strict criteria for being one kind realized by many distinct kinds ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro then points out that,functional kind with distinct realizations)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=The usual justification for grouping,Shapiro remarks)). The plasticity scenarios are a good test. They would say: if the auditory cortex truly remained an “auditory cortex” in all its causal properties yet somehow delivered vision, that would be a genuine different realization. But if the auditory cortex had to become functionally like a visual cortex, then we haven’t seen a new type of realization, just the usual realization occurring in an atypical spot. In their framework, the differences in neural properties in the new implementation must be causally irrelevant to the function for it to count as the same kind of function. If differences are causally relevant, then the function is not exactly the same (or at least not performed to the same degree).

Patricia Churchland and others have also noted that plasticity often reinforces how organization dictates function. Churchland discussed cases of cortical remapping to emphasize that the brain is an adaptive system, but its adaptations follow certain principles of connectivity and development. She would likely interpret that the ability of one area to take over another’s role is because the cortex is somewhat generic early on and can be instructed by input; yet the end result is still a brain that has specialized circuits for the task, just in a non-standard location.

In sum, neural plasticity provides a nuanced perspective. It shows that the relationship between mental functions and neural structures is not entirely rigid (one function can sometimes occur in a different structure), which functionalists often cite as evidence for substrate-independence. But the critics respond that these cases do not fully vindicate the MR thesis in its original intent. Instead of demonstrating completely independent realizations of a cognitive function, such cases often demonstrate the brain’s resilience and ability to find a way to realize the function using available resources, often by making those resources functionally resemble the usual ones. The neuroscientific critique here is that while MR is possible (the brain has some flexibility), it is still constrained by underlying commonalities – you can’t get vision out of just any tissue unless that tissue can assume the functional organization required. Thus, plasticity does not imply that mental functions are indifferent to their biological implementation; rather, it underscores the structured, rule-governed nature of how implementations can change. Polger and Shapiro conclude that clear-cut examples of multiple realizability in the brain are rarer than they might seem at first glance ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to)). Even the best candidates (like the ferret experiment) turn out to involve significant commonalities in mechanism.

Conserved Molecular Mechanisms and the Depth of Realization

Another line of neuroscientific critique of multiple realizability dives beneath the level of brain regions and systems down to the cellular and molecular mechanisms of neural function. The question here is: even if different species have different brain anatomies, might they share common biochemical or physiological pathways that implement key psychological functions? If yes, then at a deeper level the realizations are not different at all – they are fundamentally the same, just wrapped in varied packaging. John Bickle has been a leading voice in this arena, advocating what he calls “ruthless reductionism.” Bickle suggests that to find true unity (or lack thereof) in the realization of mental states, we should look at neuroscience’s discoveries at the cellular/molecular level across species ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bickle ,increasingly being found at these)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=inquiries at the systems level,the role of the cyclic)).

Bickle (2003) argues that while at the large scale (whole brain systems) different species might appear to have different neural implementations – giving an initial impression of multiple realization – neuroscience does not stop at that level ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bickle ,increasingly being found at these)). As research progresses into the micro-level, scientists have uncovered identities of mechanism across species ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=inquiries at the systems level,the role of the cyclic)). In other words, the lower-level processes (neuronal signaling, synaptic plasticity mechanisms, gene expression changes) often turn out to be highly conserved. He notes that many fundamental aspects of neurophysiology are the same from simple invertebrates to mammals: the way neurons fire (action potentials via ion channels), how synapses strengthen or weaken (long-term potentiation mechanisms), and even how neural circuits store memory, all show common biochemical underpinnings ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=physiology and increasingly into the,element)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Work with fruit flies%2C sea,term memories into long)). This is crucial for psychology because cognitive functions like learning, memory, and even aspects of emotion or perception ultimately depend on these cellular processes (e.g. memory consolidation relies on synaptic changes).

Memory consolidation is a prime example Bickle uses. The process of converting a short-term memory into a long-term memory (stabilizing it so it persists) has been studied in organisms ranging from the sea slug Aplysia and fruit flies (Drosophila), to mice and other mammals. Remarkably, research has implicated the same molecular signaling pathway in all these cases: the cyclic AMP (cAMP) signaling cascade that activates protein kinase A (PKA) and leads to changes in gene transcription via CREB (cAMP response element-binding protein). Experiments show that disrupting this pathway in these animals interferes with long-term memory formation ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Work with fruit flies%2C sea,term memories into long)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=insect biologists Josh Dubnau and,Tom Tully)). Bickle highlights a quote from biologists Josh Dubnau and Tom Tully: “In all systems studied, the cAMP signaling cascade has been identified as one of the major biochemical pathways involved in modulating both neuronal and behavioral plasticity. … [Evidence] suggest that CREB may constitute a universally conserved molecular switch for long term memory.” ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=insect biologists Josh Dubnau and,Tom Tully)). This statement, based on data from insects, mollusks, and vertebrates, strongly indicates a universal realizer at the molecular level for the function of forming long-term memories. So, while an octopus brain and a human brain look very different, and their memory circuits are organized differently, at the level of molecules the process of making memories might be running the same “software” (the cAMP-CREB pathway). If that is true, then the differences in higher-level realization are partly superficial – deep down, the realization is the same, undermining the claim of truly distinct realizations.

Bickle extends this reasoning with principles from evolutionary biology ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=that argument%2C Bickle turns to,share the common ancestor that)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=“universally conserved” molecular switch,cellular cognition” proceeds%2C we should)). Critical molecular mechanisms tend to be conserved by evolution because changes to them usually harm the organism’s fitness. If a protein or pathway is fundamental to learning, evolution will not reinvent it in each lineage; rather, anything with a common ancestor that had that mechanism will likely retain it (barring extraordinary circumstances). Thus, we should expect that many cognitive kinds (like basic forms of learning, sensory processing, etc.) share common molecular bases across species ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=constant across existing biological species,these shared mechanisms of memory)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=species—changes to it%2C especially its,species for shared psychological kinds)). Evolutionary conservation essentially means that multiple realizability is constrained at the genetic and molecular level. You might have many surface variations (different brain sizes, shapes, network layouts), but under the hood, the neurons operate with conserved mechanisms. In Bickle’s view, as neuroscience probes these deeper levels (what he sometimes calls the “new wave” reductionism), it finds unitary realizers for psychological phenomena in surprising places ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=individual neurons,unitary realizers—ruthless reductions—across species for)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=expect that the molecular mechanisms,species for shared psychological kinds)).

Another example: consider emotion and neurotransmitters. Mammalian brains and even simpler vertebrate brains share common neurotransmitters like dopamine, serotonin, oxytocin, etc., that are involved in reward, mood, and social behaviors. If an octopus shows a learning behavior, and a mammal does too, they might both utilize serotonin in the process (indeed, experiments on mollusks show serotonin’s role in learning in Aplysia). So one could argue the “realizer” of certain reward learning might be the serotonin-modulated synaptic change, present in both. The structural context differs, but the chemical basis is shared.

Philosophically, this line of argument suggests that the multiple realizability premise might fail at the lowest levels. Even if one remains agnostic at the brain-systems level, the discovery of common molecular mechanisms means the supposed different realizations are not different all the way down – they converge. This has implications for reduction: it hints that cross-species reduction of psychological phenomena to neurobiology is feasible because at some level, it’s the same neurobiology. Patricia Churchland had pointed this out with analogies to other fields: e.g., heat is multiply realized (motion of molecules in gas vs. liquid vs. solid), but at the deeper level of statistical mechanics, it’s all kinetic energy – a unifying basis ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=help of scientific examples,example is the concept of)). Similarly, life is implemented differently in different organisms, but at the molecular level, all life uses DNA/RNA and the same genetic code. So, MR doesn’t necessarily prevent unification or reduction when we find the common platform beneath the variations.

It’s worth noting, however, that one could argue this cuts both ways. A defender of MR might say: “Yes, they share molecular pathways, but the organization at higher levels is still different – and that organization matters for the cognitive phenomenon.” Bickle acknowledges that memory consolidation is just one example, and not every psychological kind may reduce neatly to one pathway ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Memory consolidation is just one,metabolic)). The risk of cherry-picking exists: perhaps we find a few conserved mechanisms (especially for very basic processes tied to cell biology), but more complex cognitive abilities (like language or abstract reasoning) might not have such clear molecular commonality across species (indeed, those higher functions might have evolved more recently and divergently). Bickle’s optimistic view is that as “molecular and cellular cognition” progresses, more cases of common realization will be found ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=species—changes to it%2C especially its,species for shared psychological kinds)). But critics of reduction might remain skeptical until such cases are shown for a wider range of cognitive functions.

In any case, the conserved mechanism argument adds another weight on the side of the scale against a naive multiple realizability thesis. It underscores that the depth of analysis matters: a mental function might appear multiply realized at one level of description (different brain anatomies) but not at a lower level (same molecular process). So, whether something counts as multiply realized can depend on what level of mechanism we consider the “realizer.” Polger and Shapiro (2016) also highlight this, suggesting that one must specify which level of physical description is at issue. They tend to favor looking for differences that are significant enough to count as different realizations. Minor variations or shared cores indicate a single kind of realization in their view ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=functional kind is that such,Shapiro remarks)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(2000%3A 649)).

In summary, neuroscientific findings about molecular conservation and cross-species commonalities in neural processes provide a potent empirical challenge to multiple realizability. The more we find such universals, the less convincing it is that mental kinds float free of physical specifics. Instead, it appears that biology often gives us “the same wires under the hood,” even if the chassis looks different. This evidence aligns with a more reductionist or identity-friendly perspective, suggesting that at least some mental states correspond to single neurobiological kinds after all (just maybe not at the gross anatomical level, but at a finer grain). Philosophers like Bickle and Churchland use these points to argue that computational functionalism must be tempered: the mind may be computational, but the computations are implemented in very particular neural hardware that has common features across instances of that computation. Ignoring those features can lead to incomplete theories.

Implications for the Viability of Computational Functionalism

The neuroscience-based critiques outlined above have significant implications for computational functionalism as a philosophy of mind and as a guiding assumption in cognitive science. Computational functionalism held that mental processes are essentially medium-independent computations, and thus multiple realizability was expected, if not assumed. Given the critiques, we must ask: if mental functions are not as freely multiply realizable as once thought, what does that mean for computational functionalism? Does it undermine the approach, or merely refine it?

1. Redefining Functional Kinds with Neural Constraints: One immediate effect of these critiques is the realization that our definitions of mental kinds might need to be more fine-grained and informed by neuroscience. If, as Shapiro argues, many alleged instances of a single mental kind across species are actually different kinds when you consider their causally relevant properties ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro’s dilemma is in the,what makes something a scientific)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(2000%3A 649)), then functionalists may have to narrow what counts as “the same” mental state. Computational functionalism might survive, but it would work with more biologically constrained notions of function. For example, instead of saying “memory” in general is a functional kind multiply realized in sea slugs and humans, we might distinguish types of memory and their mechanisms (perhaps “cAMP/CREB-dependent memory consolidation” as a functional kind that is realized similarly across those species). In other words, cognitive science may become more like neuroscience in carving nature at its joints. This doesn’t eliminate functional explanation, but it links it more tightly to mechanism. Philosophers Thomas Polger and Lawrence Shapiro in their 2016 book stress that acknowledging limited MR does not mean denying the reality of psychological states or the usefulness of higher-level science ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=depression%2C preferences%2C and institutions%3F More,the case for multiple realization)) ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=defend the integrity and realism,the case for multiple realization)). They see psychology as still autonomous in some respects, but argue that its autonomy doesn’t hinge on an exaggerated MR thesis. Reduction (or integration) is “no threat to psychological realism,” in their words ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=defend the integrity and realism,the case for multiple realization)). So computational functionalism can still claim that mental states are functional as opposed to purely behavioral or purely spiritual, but it may have to concede that those functional states often line up one-to-one with specific neural states after all (making them closer to identity theory than originally admitted).

2. The Fate of Multiple Realizability as an Argument: Historically, MR was used as a weapon against reductionist identity theory. If that weapon is blunted (because MR is rarer or less radical than thought), one consequence is that reductive approaches gain plausibility. As we have seen, John Bickle welcomes this: he believes the erosion of the MR premise gives “aid and comfort” to psychoneural reductionism ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bickle ,reductionism was one of the)). If each mental kind (or many of them) corresponds to basically one neural kind (perhaps at some appropriate level of description), then the classic functionalist stance that “only the software matters” is weakened. Computational models of the mind might need to be continually tested and constrained by neural evidence. We see this trend in fields like computational neuroscience and cognitive neuroscience, where researchers build models that incorporate known neural architecture (e.g. neural network models that mimic layers of visual cortex). This is a shift from pure computational functionalism (which might have built a vision model without any reference to the visual cortex structure) to a more integrated approach. It suggests that the conceptual viability of a purely substrate-neutral functionalism is questionable. If all actual intelligence we know (animal minds) share certain neural design principles, one might doubt whether a completely different design (like a digital computer) could realize the exact same repertoire of mental states, at least without replicating those principles. This has even been discussed in AI: some argue that to achieve human-like understanding, AI might need to mimic human neural processes more closely, rather than just any general programming. That is a speculative extension, but it flows from the idea that medium matters.

3. Autonomy of Psychology Revisited: If MR was the main reason to keep psychology autonomous from biology, then downplaying MR opens the door for a more unified science of mind and brain. Philosophers like William Bechtel have advocated for a mechanistic philosophy of science, where psychological functions are explained via mechanisms that often span multiple levels (from cognitive algorithm down to neural circuit) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=vision demonstrate%2C neuroscientists have successfully,neuroscientific goal has been to)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(1999%3A 201)). The neuroscience critiques reinforce this mechanistic, integrative approach. Instead of seeing cognitive science and neuroscience as separate enterprises (with cognitive science focusing on functional software and neuroscience on implementational hardware), the modern trend—sometimes called neurocognitive science—sees them as collaborating to describe the same mechanisms at different grains. Computational functionalism in a revised form can still play a role: we still develop computational models of cognition, but we treat them as ultimately realized in a specific class of neural mechanisms. This may actually improve the models, as it imposes biologically plausible constraints.

4. The Limits of Functional Equivalence: Another implication concerns the limits of medium-independence. The critiques suggest that mental processes might not be completely medium-independent after all. They could be modestly medium-independent: there is some flexibility (as plasticity shows), but not an anything-goes situation. For philosophy of mind, this might weaken thought experiments like “imagine an alien with a silicone-based brain that has pain.” The question becomes, how would we know it’s the same pain? The neuroscience perspective might say: unless that alien’s brain does something analogous (perhaps releasing something like substance P, activating analogous circuits, etc.), it might not actually have what we call pain, but something only superficially similar. The debate shifts to what the essence of a mental state is – purely functional organization, or functional organization plus some material constraints. Some philosophers (e.g., Ned Block with his notion of psychofunctionalism) have allowed that our notion of mental states might be partially informed by the actual human realizations, not just abstract functional roles. The empirical findings lend weight to that view.

5. Balancing Reduction and Pluralism: It is important to note that not everyone interprets the neuroscience as a total victory for reductionism. Many acknowledge that multiple realizability does occur in some form – especially when considering technology or very different life forms. For instance, while all mammals may share a memory mechanism, one could argue that an octopus (with its radically different nervous system layout) still learns and remembers in a way that has important differences; or that an AI could implement memory in a non-neural way but still functionally mimic human memory. Philosophers Robert A. Wilson and others have cautioned that we not overstate the demise of MR. What the last 20 years have shown is that one must demonstrate MR with detailed analysis, not simply assume it. This has led to a more moderate position: multiple realizability is not the default truth of cognitive science, but a hypothesis to be tested case by case. Some cases will confirm it (especially at a coarse grain or across very dissimilar systems), others will refute it (showing uniformity or identity). Computational functionalism can thus be reinterpreted as saying mental states are in principle realizable in multiple ways, but in our actual world, evolution may have funneled cognition into a narrower set of mechanisms – a contingency that functionalist theory needs to accommodate.

6. Revision of Computational Metaphors: The critiques also urge caution in how we use the computer metaphor for the brain. Early functionalism likened brain to a general-purpose computer running software. Neuroscience suggests the brain is not a general-purpose Von Neumann machine; it’s more of a highly specialized, parallel, self-organizing system. Some computational functionalists have updated the metaphor to neural networks and parallel distributed processing, which are indeed more neurally inspired. Those models implicitly accept that how the computation is done (architecture) matters. For example, deep learning networks today draw from how visual cortex is structured. This can be seen as a vindication of the neurophilosophical critique: understanding the brain’s algorithms requires understanding its hardware constraints. The computational models that succeed (like convolutional neural networks for vision) are those that incorporate principles from actual visual neuroscience. This trend might indicate that computational functionalism has not been abandoned, but it has evolved – it is becoming neurocomputational functionalism, if you will, where the “multiple realizability” is limited to variations on a theme, not arbitrary implementations.

In conclusion, the conceptual viability of computational functionalism in light of neuroscientific critiques is a topic of ongoing negotiation. The strong, classic form of functionalism – which treated the physical implementation as irrelevant – appears less tenable. But a weaker form, which still emphasizes functional description but acknowledges the implementing mechanisms, remains viable and indeed is the norm in cognitive science today. This shift parallels how other fields matured: early chemistry didn’t care about quantum physics, but eventually we integrated them. Likewise, psychology can maintain its own level of explanation while increasingly connecting to neuroscience. The multiple realizability thesis was a useful corrective against crude reductionism, but taken to an extreme it isolated cognitive theory from biological reality. The neuroscience evidence has swung the pendulum back, encouraging a more integrative approach that promises a deeper understanding of mind-brain relationships.

Conclusion

How strong are the neuroscience-based critiques against the multiple realizability thesis, and what are the implications for the future? Based on this review, the critiques are substantial and have significantly eroded the once-dominant view that mental states are multiply realized across brains in an unconstrained way. Empirical findings from the last two decades strongly indicate that, in practice, nature often achieves similar cognitive and behavioral functions using similar neural solutions. Cross-species comparisons show homologous brain structures carrying out analogous roles, suggesting common realizations rather than diverse ones ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=because they were likely to,across species%2C and persistence of)). Within humans and other species, cortical specialization and consistent lesion/imaging correlations point to a fixed mapping between many mental functions and specific neural substrates. Where variability exists, it usually does not amount to wholly different realizations, but rather minor variations or compensation within the same framework. Even neural plasticity, which at first glance supports functional flexibility, upon close inspection tends to reinforce how the brain can reconfigure itself to approximate the normal circuitry for a function, instead of demonstrating a brand-new method of realization ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to)). At the microscopic scale, conserved molecular mechanisms like the universal role of the cAMP-CREB pathway in memory show a deep unity of implementation across diverse life forms ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=insect biologists Josh Dubnau and,Tom Tully)). All these lines of evidence converge on a message: the multiple realizability of mental states is more limited than philosophers once assumed.

That said, the story is not one of outright dismissal of multiple realizability. There remain important nuances and open questions. For one, multiple realizability does occur in some respects – for instance, artificial systems can emulate certain human cognitive functions (think of computer vision or speech recognition) using very different hardware and algorithms. The philosophy of AI will need to grapple with whether those counts as the “same” mental state or only a functional analogue. Additionally, even within biology, as we broaden our scope to extremely different organisms (e.g. the octopus, whose brain architecture is completely unlike a vertebrate’s), we may yet find cases where similar capacities (like problem-solving or play behavior) arise from truly distinct neural designs. The current critiques have focused on known species and well-studied functions, mostly within mammals or comparing mammals to simpler model organisms. It is possible that evolution might, in some instances, discover genuinely different solutions to similar problems (convergent evolution in cognition, analogous to how a squid’s eye and a human’s eye are similar in function but evolved separately). The extent of such convergences in neural terms is still an empirical question. Thus, multiple realizability is down but not out – it has lost its status as an unquestionable axiom and become an empirical hypothesis subject to verification.

The strength of the neuroscience critiques lies in making us rethink how we categorize mental states and what we expect of a scientific theory of mind. They encourage a more interdisciplinary approach: philosophers of mind must pay attention to neuroscientific details, and cognitive scientists can benefit from philosophical clarity about levels of explanation and criteria for sameness of function. The debate has prompted the development of more sophisticated accounts of realization and reduction. For example, researchers are investigating formal ways to describe what it means for two systems to implement the same computation and where differences in implementation become differences in function (this involves work in computational neuroscience and philosophy of computation). The outcome is likely a more refined functionalism – one that is compatible with (and enriched by) neuroscience, rather than aloof from it. In practical terms, this means cognitive models will increasingly incorporate constraints from neural architecture, and neuroscientific theories will explicitly link to cognitive constructs.

For the future research in philosophy of mind and cognitive science, several implications and avenues emerge:

• Refining Taxonomies of Mind: We may need to revise our taxonomies of mental kinds in light of neural evidence. Future work can explore if some psychological categories should be split or redefined because what we thought was one kind of state is realized differently in different contexts (or vice versa, unify categories that share mechanisms).
• Cross-Disciplinary Case Studies: Detailed case studies of specific mental functions (e.g. visual perception, episodic memory, tool use, emotional feelings) across multiple species and implementations (biological and artificial) can illuminate how much variability in realization exists. This will inform the MR debate with concrete data. Philosophers like Ken Aizawa and Carl Craver have engaged in such empirical examinations, sometimes pushing back on earlier critiques by arguing that proponents like Bechtel & Mundale oversimplified neuroscience practice ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=These responses quickly attracted critical,Carl Gillett (2003) argues)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=anatomical studies used multiple species,Bechtel and McCauley remind us)). Ongoing dialogue between empirical findings and philosophical interpretation is crucial.
• Computational Neuroscience Models: The development of biologically realistic computational models that can be run in different configurations (e.g. simulate a function with one architecture vs. another) might help test multiple realizability in silico. If two very different models both produce the same cognitive performance, what does that tell us about necessary vs. contingent features of realization?
• Philosophy of AI and Synthetic Minds: As artificial intelligence grows more sophisticated, the question of whether AI systems realize mental states in the same sense as brains do will become more pressing. The neuroscience-based perspective might predict that without the patterns present in biological brains, an AI might not truly duplicate certain mental phenomena (like conscious experience or certain intuitive cognitive skills). This is speculative but a fertile ground for philosophers like Paul Churchland (who in the past speculated about neural networks being needed for mind-like intelligence, as opposed to symbolic AI). The MR debate provides a framework to tackle these questions: are we seeing a new realization of intelligence in AI, or just a crude mimicry missing the essence captured by neural processes?
• Metaphysics of Science: On a more philosophical note, the fate of multiple realizability ties into discussions about the metaphysics of science – specifically, the nature of natural kinds and the unity of science. Some argue that higher-level kinds (like mental states) won’t neatly map to lower-level kinds (neural states), resulting in a kind of pluralism (each science has its own kinds). Others, buoyed by the recent critiques, argue for a more integrated picture where at least some kinds line up across levels, enabling cross-level generalizations. The truth might be intermediate: a patchwork where some mental kinds reduce to neural kinds while others remain somewhat autonomous. Philosophers such as Carl Gillett and William Wimsatt have contributed to this nuanced view.

In weighing the strength of the neuroscience critiques, it’s fair to say they have delivered a serious blow to the mythos of ubiquitous multiple realizability. They shifted the burden of proof: it is no longer convincing to simply assert MR without evidence; one must show how a given mental function is realized differently and why that difference is not just superficial. Many influential philosophers (Bechtel, Shapiro, Polger, Bickle, Churchland, among others) now converge on the view that multiple realizability is a more limited phenomenon ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=Instead%2C they are interested in,multiple realization are greatly exaggerated)). As Polger and Shapiro memorably conclude, “expectations of multiple realization are greatly exaggerated.” ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=Instead%2C they are interested in,multiple realization are greatly exaggerated)) Still, a balanced perspective acknowledges that MR has not been refuted in principle – it remains a conceivable feature of minds, especially as we imagine minds beyond the ones evolution gave us. Thus, the prudent approach for future inquiries is empirically open-minded: investigate each case of a purported mental kind across various implementations, and determine empirically whether it is a single kind with multiple realizers or a cluster of related but distinct kinds.

For computational functionalism, this means its legacy endures in the commitment to functional analysis, but with a renewed appreciation that function cannot be divorced from structure. The brain is not just an arbitrary computer – it embodies solutions shaped by biology. The more our theories of mind incorporate those solutions, the more accurately they will capture the nature of cognition. In the end, the reconciliation of functionalist and neuroscientific perspectives promises a more comprehensive science of the mind, one that respects both the software and the hardware of cognition, and understands their intimate interdependence.

References:

• Bechtel, W. & Mundale, J. (1999). “Multiple Realizability Revisited: Linking Cognitive and Neural States.” Philosophy of Science, 66(2), 175-207. ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=The claim of the multiple,of multiply realized psychological functions)) ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=two noteworthy points%3A ,function and brain mapping research))
• Bickle, J. (2003). Philosophy and Neuroscience: A Ruthlessly Reductive Account. Dordrecht: Kluwer. ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Bickle ,increasingly being found at these)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=insect biologists Josh Dubnau and,Tom Tully))
• Churchland, P. S. (1986). Neurophilosophy: Toward a Unified Science of the Mind-Brain. Cambridge, MA: MIT Press. ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=help of scientific examples,example is the concept of)) ( Multiple Realizability (Stanford Encyclopedia of Philosophy) %2C Patricia Churchland ,example%2C the physiology of individual))
• Fodor, J. (1974). “Special Sciences (Or: The Disunity of Science as a Working Hypothesis).” Synthese, 28, 97-115. ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Jerry Fodor ,that there are natural kind)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Fodor insists that it is,Multiple realizability thus))
• Polger, T. W. & Shapiro, L. A. (2016). The Multiple Realization Book. Oxford: Oxford University Press. ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=Instead%2C they are interested in,multiple realization are greatly exaggerated)) ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to))
• Putnam, H. (1967). “Psychological Predicates,” in W. H. Capitan & D. D. Merrill (eds.), Art, Mind, and Religion. Pittsburgh: University of Pittsburgh Press. (Reprinted as “The Nature of Mental States” in Putnam, 1975). ([Multiple Realizability Revisited](https://mechanism.ucsd.edu/~bill/research/multiple.htm#:~:text=Hilary Putnam’s original multiple realizability,thesis is the contention that)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=In the philosophy of mind%2C,this premise and these arguments))
• Shapiro, L. (2000). “Multiple Realizations.” The Journal of Philosophy, 97(12), 635-654. ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=,(2000%3A 644)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=Shapiro then points out that,functional kind with distinct realizations))
• Sur, M., Pallas, S. L., & Roe, A. W. (1990). “Cross-modal plasticity in cortical development: differentiation and specification of sensory neocortex.” Trends in Neurosciences, 13(5), 227-233. (See also: von Melchner, L., Pallas, S. L., & Sur, M. (2000). “Visual behaviour mediated by retinal projections directed to the auditory pathway.” Nature, 404, 871-876.) ([Polger and Shapiro’s The Multiple Realization Book // Reviewed by Marion Godman](https://www.thebsps.org/reviewofbooks/thomas-w-polger-and-lawrence-a-shapiro-the-multiple-realization-book/#:~:text=(von Melchner et al. ,this can be reduced to))
• [Additional sources and discussion integrated throughout ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#::text=vision demonstrate%2C neuroscientists have successfully,neuroscientific goal has been to)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#::text=,(2000%3A 649)) ([ Multiple Realizability (Stanford Encyclopedia of Philosophy) ](https://plato.stanford.edu/entries/multiple-realizability/#:~:text=meets the road,species for shared psychological kinds)).]