Beyond reduction: mechanisms, multifield integration and the unity of neuroscience
Introduction
Neuroscience is a multifield research program.1 Its departments, journals, societies, and textbooks include perspectives from anatomy, biochemistry, computer science, developmental, evolutionary and molecular biology, electrophysiology, experimental psychology, ethology, pharmacology, psychiatry, and radiology to name just a few. The Society for Neuroscience (SfN) was founded in 1969 with the mission to:
Advance the understanding of the brain and the nervous system by bringing together scientists of diverse backgrounds, by facilitating the integration of research directed at all levels of biological organization, and by encouraging translational research and the application of new scientific knowledge to develop improved disease treatments and cures.2
Understanding the structure of contemporary neuroscience requires understanding how these multiple fields, embodying distinct perspectives, techniques, and vocabularies, manage to integrate their work.
Most philosophers who have discussed interfield integration in neuroscience (e.g., Bickle, 1998, Bickle, 2003, Churchland, 1986, Oppenheim and Putnam, 1958, Schaffner, 1993a, Schaffner, 1993b) describe it using models of intertheoretic reduction. According to the ‘classical’ model of reduction, from which each of these authors’ models descends, reduction is a species of deductive nomological explanation: one theory is reduced to another when it is possible to identify the theoretical terms of the first with those of the second and to literally derive the first from the second. On the assumption that fields and theories correspond, reduction then serves as a model of interfield integration as well.
There are many reasons why philosophers of neuroscience have found reduction attractive for thinking about interfield integration. First, the reduction relation can be defined precisely using formal logic (e.g., Schaffner, 1993a, Schaffner, 1993b) or set theory (e.g., Bickle, 2003), and so the thesis that fields are integrated through reduction is clear and testable. Second, there is a long tradition of using reduction models in the philosophy of physics, chemistry, and biology, and it is natural to suggest that the models can be extended to the neurosciences. Finally, at least since Oppenheim and Putnam’s manifesto (1958), reduction has been nearly synonymous with the explanatory unity of science: the unity of science is achieved by reducing the theories of all fields to the theories of the one field describing fundamental ontology. For these reasons, reduction seems natural as a model of interfield integration in the neurosciences.
But reduction models have well rehearsed shortcomings. Most philosophers now accept that nonfundamental kinds are multiply realizable — that is, the same kind of phenomenon can be realized on different occasions by wildly different substances, parts, or organized processes. Arguments concerning multiple realization are largely responsible for the fact that non-reductive physicalism is now a standard view in the philosophy of mind (Putnam, 1975, Fodor, 1974, Fodor, 1997). Furthermore, derivational models of explanation have received sustained attack for decades, and many now recognize the importance of non-derivational and especially causal forms of scientific explanation (e.g., Salmon, 1984, Salmon, 1989). And finally, historians of science have identified numerous non-reductive forms of interfield integration including the development of interfield theories (Maull, 1977, Darden and Maull, 1977, Bechtel, 1986), interfield relations (Darden, 1991), and pidgin languages in the development of scientific techniques (Galison, 1987; see also Wylie, 2003). Reduction seems to survive as a model of interfield integration in neuroscience despite these shortcomings largely because there is no alternative of comparable scope and clarity.
The primary aim of this paper is to articulate such an alternative, to compare it to reduction, and to display some of its relative merits. My target is reduction as a model of interfield integration and the unity of neuroscience. There are further worries about reduction as a model of explanation or as a metaphysical thesis. Although such worries cannot be sidestepped entirely, my present project is limited to evaluating reduction as a model of interfield integration and the unity of neuroscience.3
This paper is partly historical and partly constructive. In the historical component (Sections 3 Technical background: Long-Term Potentiation and LM mechanisms, 4 LTP’s origins: not top-down search but intralevel integration, 5 The mechanistic shift), I consider the development of a prominent multifield research program in contemporary neuroscience: the learning and memory (LM) research program.4 Several philosophers of neuroscience have described the LM research program as an exemplar of reduction. I argue, in contrast, that the development of the LM research program reveals at least three limitations of reduction models: first, that they neglect upward-looking aspects of interlevel interfield integration; second, that they ignore intralevel forms of interfield integration; and third, that they gloss over the fact that scientific progress has sometimes been achieved by abandoning reduction as an explanatory goal. Indeed, the LM research program needed to move beyond reduction to formulate an adequate explanation of learning and memory. Through the 1970s, as interdisciplinary fervor culminated in the SfN’s foundation, the LM research program began to pursue multilevel and mechanistic forms of explanation. This mechanistic shift provided an abstract multilevel structure both onto and around which diverse fields could integrate their results. Scientists engaged in this pursuit integrate fields by adding constraints on mechanisms. The constructive portion of the paper examines this constraint-based and mechanistic form of scientific integration. In Section 6, I show how the pursuit of mechanisms scaffolds interfield integration at a given level. In Section 7, I illustrate how attention to mechanistic relations clarifies constraints on interlevel integration. I conclude that mechanistic models of interfield integration are more historically accurate than reduction in describing the LM research program, that they reveal neglected varieties of interfield integration, and that they are more informative about the kinds of constraints needed to build the bridges from molecules to behavior.
Section snippets
Reductive views of interfield integration
There are many different reduction models (see, e.g., Nagel, 1949, Hooker, 1981, Schaffner, 1969, Schaffner, 1993a, Bickle, 1998), and there is no need to review the details of each. It will be more instructive for my purposes to develop a prototype of reduction. By prototype I mean a list of features that hold for many reduction models even if possibly no model has all of them. The mechanistic approach rejects each of these features.
First, reduction is an intertheoretic relationship. This
Technical background: Long-Term Potentiation and LM mechanisms
I will focus specifically on the protracted discovery of an electrophysiological phenomenon now known as Long-Term Potentiation (or LTP). LTP is the enhancement of a synapse resulting from rapid and repeated electrophysiological stimulation (a tetanus) to the pre-synaptic neuron. Many believe that this laboratory phenomenon demonstrates the existence of processes in the brain that might underlie synaptic plasticity during learning or memory.
LTP was first encountered in the hippocampus (shown in
LTP’s origins: not top-down search but intralevel integration
Most neuroscientists date the discovery of LTP to a series of 1973 papers by Tim Bliss, Terje Lømo, and Tony Gardner-Medwin. These papers became a watershed in the development of the LTP research program. However, the story of LTP begins much earlier. In recovering the earlier history, we see the origins of LTP not as an instance of reduction, but rather as an instance of intralevel (and so nonreductive) interfield integration.
Electrophysiologists first produced and reported synaptic plasticity
The mechanistic shift
How did LTP gain its association with learning and memory? This upward-looking phase of the research program, which continues to the present, only developed after the initial discovery of long lasting (i.e., greater than ten minutes) forms of potentiation in the hippocampus. Some neuroscientists in the 1950s and 1960s had a reductive view of the connection between synaptic changes and learning or memory. However, with the LTP watershed, these reductive aspirations were subtly replaced by
Mechanisms and intralevel integration
Researchers in the LM research program are not just building theories simpliciter; rather, they are building theories about mechanisms. Perhaps mechanisms can be described using formal accounts of theories—perhaps they can be axiomatized in predicate logic or reconstructed as set theoretic predicates. But such formal accounts of the structures of scientific theories gloss over the mechanistic structures crucial for understanding how these theories are constructed and evaluated. We cannot hope
Mechanisms and interlevel integration
The mechanistic approach also has many advantages over reduction for thinking about interlevel forms of interfield integration. First, it provides a straightforward way to interpret the talk of ‘levels’, which is a problem almost entirely neglected in much of the literature on reduction. Second, it offers significantly more insight into what interlevel integration is, into the kinds of evidence by which interlevel bridges are evaluated, and into the forces driving the co-evolution of work at
Conclusion
Reductive models of interfield integration describe it as an abstract relationship between theories at different levels and as a relationship that involves establishing homomorphism between the reduced and reducing theories. According to the mechanistic model of interfield integration, fields are integrated by adding constraints on the organization of a mechanism. Given the centrality of the search for mechanisms in contemporary neuroscience and beyond, it seems fair to say that reduction has a
Acknowledgements
William Bechtel, Lindley Darden, Ilya Farber, Kenneth Schaffner, and Alison Wylie provided helpful comments on earlier drafts. Kim Haddix and Phil Valko provided editorial assistance. Any errors are my own.
References (61)
- et al.
Control of hippocampal output by afferent volley frequency
Progress in Brain Research
(1967) - et al.
Synaptic tagging: Implications for late maintenance of hippocampal long-term potentiation
Trends in the Neurosciences
(1998) - et al.
Electrophysiological studies of hippocampal neurons III: Responses of hippocampal neurons to repetitive perforant path volleys
Electroencephalography and Clinical Neurophysiolology
(1964) - et al.
Neurophysiological studies of hippocampal connections and excitability
Electroencephalography and Clinical Neurophysiology
(1956) Unifying science without reduction
Studies in History and Philosophy of Science
(1977)- et al.
Impaired hippocampal representation of space in CA1-Specific NMDAR1 knockout mice
Cell
(1996) Interhippocampal impulses II. Apical dendritic activation of CA1 neurons
Acta Physiologica Scandinavica
(1960)Interhippocampal impulses III. Basal dendritic activation of CA3 neurons
Acta Physiologica Scandinavica
(1960)LTP—an exciting and continuing saga
A prelude to long-term potentiation
Philosophical Transactions of the Royal Society of London, Series B
(2003)
Activation of the dentate area by septal stimulation
Acta Physiologica Scandinavica
Discovering complexity: Decomposition and localization as strategies in scientific research
Psychoneural reduction: The new wave
Philosophy of neuroscience: A ruthlessly reductive approach
Long-lasting potentiation of synaptic transmission in the dentate area of the unanaesthetized rabbit following stimulation of the perforant path
Journal of Physiology
Synaptic plasticity in the hippocampal formation
Long-lasting potentiation of synaptic transmission in the dentate area of the anaesthetized rabbit following stimulation of the perforant path
Journal of Physiology
Neurophilosophy
The computational brain
Being there: Putting brain, body and world together again
Some commissural and septal connexions of the hippocampus in the rabbit: A combined histological and electrical study
Journal of Physiology
Interlevel experiments and multilevel mechanisms in the neuroscience of memory
Philosophy of Science
The making of a memory mechanism
Journal of the History of Biology
Discovering mechanisms in neurobiology: The case of spatial memory
Theory change in science: Strategies from Mendelian genetics
Interfield theories
Philosophy of Science
The neurophysiological basis of mind
The physiology of synapses
Cited by (104)
Expanding the notion of mechanism to further understanding of biopsychosocial disorders? Depression and medically-unexplained pain as cases in point
2024, Studies in History and Philosophy of ScienceEpistemic cultural constraints on the uses of psychology
2023, New Ideas in PsychologyIdealisation, genetic explanations and political behaviours: Notes on the anti-reductionist critique of genopolitics
2021, Studies in History and Philosophy of ScienceDetail and generality in mechanistic explanation
2020, Studies in History and Philosophy of Science Part ACitation Excerpt :After outlining some clarifications and potential criticisms of my position, I finish by considering how open questions regarding what constitutes a mechanism may bear on these arguments. Although the “new mechanist philosophy” engages with many different aspects of scientific practice, such as investigation and representation (e.g., Bechtel & Richardson, 1993; Bechtel, 2006; Craver, 2005; Darden & Craver, 2002; Menzies, 2012), a predominant focus has always been scientific explanation. While earlier philosophical accounts of explanation had often highlighted the role of laws or very broad causal generalisations, mechanist philosophers pointed out that many scientific disciplines explain phenomena through descriptions of the (possibly quite specific) mechanisms that underlie those phenomena (e.g., Bechtel & Abrahamsen, 2005; Craver, 2006, 2007; Glennan, 2002).