Replicability and reproducibility of computational models has been somewhat understudied by “the replication movement.” In this paper, we draw on methodological studies into the replicability of psychological experiments and on the mechanistic account of explanation to analyze the functions of model replications and model reproductions in computational neuroscience. We contend that model replicability, or independent researchers' ability to obtain the same output using original code and data, and model reproducibility, or independent researchers' ability to recreate a model without original code, (...) serve different functions and fail for different reasons. This means that measures designed to improve model replicability may not enhance (and, in some cases, may actually damage) model reproducibility. We claim that although both are undesirable, low model reproducibility poses more of a threat to long-term scientific progress than low model replicability. In our opinion, low model reproducibility stems mostly from authors' omitting to provide crucial information in scientific papers and we stress that sharing all computer code and data is not a solution. Reports of computational studies should remain selective and include all and only relevant bits of code. (shrink)

In this paper, we argue that several recent ‘wide’ perspectives on cognition (embodied, embedded, extended, enactive, and distributed) are only partially relevant to the study of cognition. While these wide accounts override traditional methodological individualism, the study of cognition has already progressed beyond these proposed perspectives towards building integrated explanations of the mechanisms involved, including not only internal submechanisms but also interactions with others, groups, cognitive artifacts, and their environment. The claim is substantiated with reference to recent developments in the (...) study of “mindreading” and debates on emotions. We claim that the current practice in cognitive (neuro)science has undergone, in effect, a silent mechanistic revolution, and has turned from initial binary oppositions and abstract proposals towards the integration of wide perspectives with the rest of the cognitive (neuro)sciences. (shrink)

In artificial intelligence, recent research has demonstrated the remarkable potential of Deep Convolutional Neural Networks (DCNNs), which seem to exceed state-of-the-art performance in new domains weekly, especially on the sorts of very difficult perceptual discrimination tasks that skeptics thought would remain beyond the reach of artificial intelligence. However, it has proven difficult to explain why DCNNs perform so well. In philosophy of mind, empiricists have long suggested that complex cognition is based on information derived from sensory experience, often appealing to (...) a faculty of abstraction. Rationalists have frequently complained, however, that empiricists never adequately explained how this faculty of abstraction actually works. In this paper, I tie these two questions together, to the mutual benefit of both disciplines. I argue that the architectural features that distinguish DCNNs from earlier neural networks allow them to implement a form of hierarchical processing that I call “transformational abstraction”. Transformational abstraction iteratively converts sensory-based representations of category exemplars into new formats that are increasingly tolerant to “nuisance variation” in input. Reflecting upon the way that DCNNs leverage a combination of linear and non-linear processing to efficiently accomplish this feat allows us to understand how the brain is capable of bi-directional travel between exemplars and abstractions, addressing longstanding problems in empiricist philosophy of mind. I end by considering the prospects for future research on DCNNs, arguing that rather than simply implementing 80s connectionism with more brute-force computation, transformational abstraction counts as a qualitatively distinct form of processing ripe with philosophical and psychological significance, because it is significantly better suited to depict the generic mechanism responsible for this important kind of psychological processing in the brain. (shrink)

We analyse Hutto & Myin's three arguments against computationalism [Hutto, D., E. Myin, A. Peeters, and F. Zahnoun. Forthcoming. “The Cognitive Basis of Computation: Putting Computation In Its Place.” In The Routledge Handbook of the Computational Mind, edited by M. Sprevak, and M. Colombo. London: Routledge.; Hutto, D., and E. Myin. 2012. Radicalizing Enactivism: Basic Minds Without Content. Cambridge, MA: MIT Press; Hutto, D., and E. Myin. 2017. Evolving Enactivism: Basic Minds Meet Content. Cambridge, MA: MIT Press]. The Hard Problem (...) of Content targets computationalism that relies on semantic notion of computation, claiming that it cannot account for the natural origins of content. The Intentionality Problem is targeted against computationalism using non-semantic accounts of computation, arguing that it fails in explaining intentionality. Theion Problem claims that causal interaction between concrete physical processes and abstract computational properties is problematic. We argue that these a... (shrink)

Recent discussions of emergence in physics have focussed on the use of limiting relations, and often particularly on singular or asymptotic limits. We discuss a putative example of emergence that does not fit into this narrative: the case of phonons. These quasi-particles have some claim to be emergent, not least because the way in which they relate to the underlying crystal is almost precisely analogous to the way in which quantum particles relate to the underlying quantum field theory. But there (...) is no need to take a limit when moving from a crystal lattice based description to the phonon description. Not only does this demonstrate that we can have emergence without limits, but also provides a way of understanding cases that do involve limits. (shrink)

When does a model explain? When does it promote understanding? A dominant approach to scientific explanation is the interventionist view. According to this view, when X explains Y, intervening on X can produce, prevent or alter Y in some predictable way. In this paper, I argue for two claims. First, I reject a position that many interventionist theorists endorse. This position is that to explain some phenomenon by providing a model is also to understand that phenomenon. While endorsing the interventionist (...) view, I argue that explaining and understanding are distinct scientific achievements. Second, I defend a novel theory of scientific understanding. According to this view, when some model M promotes understanding, M makes available a distinctive mental state. This state is of the same psychological kind as when we grasp events in a narrative as bearing on some ultimate conclusion. To conclude, I show that, given this view, mechanistic explanations often provide a powerful source of understanding that many causal-historical models lack. This paper will be of interest to both philosophers of science and epistemologists engaged in the topics of explanation and understanding. (shrink)

One of the fundamental research goals for explanation-based Natural Language Inference (NLI) is to build models that can reason in complex domains through the generation of natural language explanations. However, the methodologies to design and evaluate explanation-based inference models are still poorly informed by theoretical accounts on the nature of explanation. As an attempt to provide an epistemologically grounded characterisation for NLI, this paper focuses on the scientific domain, aiming to bridge the gap between theory and practice on the notion (...) of a scientific explanation. Specifically, the paper combines a detailed survey of the modern accounts of scientific explanation in Philosophy of Science with a systematic analysis of corpora of natural language explanations, clarifying the nature and function of explanatory arguments from both a top-down (categorical) and a bottom-up (corpus-based) perspective. Through a mixture of quantitative and qualitative methodologies, the presented study allows deriving the following main conclusions: (1) Explanations cannot be entirely characterised in terms of inductive or deductive arguments as their main function is to perform unification; (2) An explanation typically cites causes and mechanisms that are responsible for the occurrence of the event to be explained; (3) While natural language explanations possess an intrinsic causal-mechanistic nature, they are not limited to causes and mechanisms, also accounting for pragmatic elements such as definitions, properties and taxonomic relations; (4) Patterns of unification naturally emerge in corpora of explanations even if not intentionally modelled; (5) Unification is realised through a process of abstraction, whose function is to provide the inference mechanism for subsuming the event to be explained under recurring patterns and high-level regularities. The paper contributes to addressing a fundamental gap in classical theoretical accounts on the nature of scientific explanations and their materialisation as linguistic artefacts. This characterisation can support a more principled design and evaluation of explanation-based AI systems which can better interpret, process, and generate natural language explanations. (shrink)

A central goal of research in explainable artificial intelligence (XAI) is to facilitate human understanding. However, understanding is an elusive concept that is difficult to target. In this paper, we argue that a useful way to conceptualize understanding within the realm of XAI is via certain human abilities. We present four criteria for a useful conceptualization of understanding in XAI and show that these are fulfilled by an abilities-based approach: First, thinking about understanding in terms of specific abilities is motivated (...) by research from numerous disciplines involved in XAI. Second, an abilities-based approach is highly versatile and can capture different forms of understanding important in XAI application contexts. Third, abilities can be operationalized for empirical studies. Fourth, abilities can be used to clarify the link between explainability, understanding, and societal desiderata concerning AI, like fairness and trustworthiness. Conceptualizing understanding as abilities can therefore support interdisciplinary collaboration among XAI researchers, provide practical benefit across diverse XAI application contexts, facilitate the development and evaluation of explainability approaches, and contribute to satisfying the societal desiderata of different stakeholders concerning AI systems. (shrink)

Research in pharmacogenomics and precision medicine has recently introduced the concept of Polygenic Scores (PGSs), namely, indexes that aggregate the effects that many genetic variants are predicted to have on individual disease risk. The popularity of PGSs is increasing rapidly, but surprisingly little attention has been paid to the idealisations they make about phenotypic development. Indeed, PGSs rely on quantitative genetics models and methods, which involve considerable theoretical assumptions that have been questioned on various grounds. This comes with epistemological and (...) ethical concerns about the use of PGSs in clinical decision-making. In this paper, I investigate to what extent idealisations in genetics models can impact the data gathering and clinical interpretation of genomics findings, particularly the calculation and predictive accuracy of PGSs. Although idealisations are considered ineliminable components of scientific models, they may be legitimate or not depending on the epistemic aims of a model. I thus analyse how various idealisations have been introduced in classical models and progressively readapted throughout the history of genetic theorising. Notably, this process involved important changes in the epistemic purpose of such idealisations, which raises the question of whether they are legitimate in the context of contemporary genomics. (shrink)

Changes in synaptic strength are described as a unifying hypothesis for memory formation and storage, leading philosophers to consider the ‘synaptic efficacy hypothesis’ as a paradigmatic explanation in neuroscience. Craver’s mosaic view has been influential in understanding synaptic efficacy by presenting long-term potentiation as a multi-level mechanism nested within a multi-level structure. This paper argues that the mosaic view fails to fully capture the explanatory power of the synaptic efficacy hypothesis due to assumptions about multi-level mechanisms. I present an alternative (...) approach that emphasizes the explanatory function of unification, accounting for the widespread consensus in neuroscience regarding synaptic efficacy by highlighting the stability of synaptic causal variables across different multi-level mechanisms. (shrink)

The relationship between topological explanation and mechanistic explanation is unclear. Most philosophers agree that at least some topological explanations are mechanistic explanations. The crucial question is how to make sense of this claim. Zednik (Philos Psychol 32(1):23–51, 2019) argues that topological explanations are mechanistic if they (i) describe mechanism sketches that (ii) pick out organizational properties of mechanisms. While we agree with Zednik’s conclusion, we critically discuss Zednik’s account and show that it fails as a general account of how and (...) when topological explanations are mechanistic. First, if topological explanations were just mechanism sketches, this implies that they could be enriched by replacing topological terms with mechanistic detail. This, however, conflicts how topological explanations are used in scientific practice. Second, Zednik’s account fails to show how topological properties can be organizational properties of mechanisms that have a place in mechanistic explanation. The core issue is that Zednik’s account ignores that topological properties often are _global_ properties while mechanistic explanantia refer to _local_ properties. We demonstrate how these problems can be solved by a recent account of mechanistic completeness (Craver and Kaplan in Br J Philos Sci 71(1):287–319, 2020; Kohár and Krickel in Calzavarini and Viola (eds) Neural mechanisms—new challenges in the philosophy of neuroscience, Springer, New York, 2021) and use a multilayer network model of Alzheimer’s Disease to illustrate this. (shrink)

Among the many functions of models, explanation is central to the functioning and aims of science. However, the discussions surrounding modeling and explanation in philosophy have largely remained separate from each other. This chapter seeks to bridge the gap by focusing on the puzzle of model-based explanation, asking how different philosophical accounts answer the following question: if idealizations and fictions introduce falsehoods into models, how can idealized and fictional models provide true explanations? The chapter provides a selective and critical overview (...) of the available strategies for solving this puzzle, mainly focusing on idealized models and how they explain. (shrink)

In light of recent criticisms by Woodward (2017) and Rescorla (2018), we examine the relationship between mechanistic explanation and phenomenological laws. We disambiguate several uses of the phrase “phenomenological law” and show how a mechanistic theory of explanation sorts them into those that are and are not explanatory. We also distinguish the problem of phenomenological laws from arguments about the explanatory power of purely phenomenal models, showing that Woodward and Rescorla conflate these problems. Finally, we argue that the temptation to (...) pit mechanistic and interventionist theories of explanation against one another occludes important and scientifically relevant research questions. (shrink)

In this essay I propose that what design principles in systems biology and systems neuroscience do is to present abstract characterizations of mechanisms, and thereby facilitate mechanistic explanation. To show this, one design principle in systems neuroscience, i.e., the multilayer perceptron, is examined. However, Braillard contends that design principles provide a sort of non-mechanistic explanation due to two related reasons: they are very general and describe non-causal dependence relationships. In response to this, I argue that, on the one hand, all (...) mechanisms are more or less general, and on the other, many design principles are causal systems. (shrink)

Autoinhibition is a design principle realized in many molecular mechanisms in biology. After explicating the notion of a design principle and showing that autoinhibition is such a principle, we focus on how researchers discovered instances of autoinhibition, using research establishing the autoinhibition of the molecular motors kinesin and dynein as our case study. Research on kinesin and dynein began in the fashion described in accounts of mechanistic explanation but, once the mechanisms had been discovered, researchers discovered that they exhibited a (...) second phenomenon, autoinhibition. The discovery of autoinhibition not only reverses the pattern in terms of which philosophers have understood mechanism discovery but runs counter to the one phenomenon-one mechanism principle assumed to relate mechanisms and the phenomena they explain. The ubiquity of autoinhibition as a design principle, therefore, necessitates a philosophical understanding of mechanisms that recognizes how they can participate in more than one phenomenon. Since mechanisms with this design are released from autoinhibition only when they are acted on by control mechanisms, we advance a revised account of mechanisms that accommodates attribution of multiple phenomena to the same mechanism and distinguishes them from other processes that control them. (shrink)

Why is it rational for scientists to pursue multiple models of a phenomenon at the same time? The literatures on mechanistic inquiry and scientific pursuit each develop answers to a version of this question which is rarely discussed by the other. The mechanistic literature suggests that scientists pursue different complementary models because each model provides detailed insights into different aspects of the phenomenon under investigation. The pursuit literature suggests that scientists pursue competing models because alternative models promise to solve outstanding (...) empirical and conceptual problems. Looking into research on visual processing as a case study, we suggest an integrated account of why it is rational for scientists to pursue both complementary and competing models of the same mechanism in scientific practice. (shrink)

Recent findings in different areas of psychology and cognitive science have brought the unconscious mind back to center stage. However, the unconscious mind worry remains: What renders unconscious phenomena mental? I suggest a new strategy for answering this question, which rests on the idea that categorizing unconscious phenomena as “mental” should be scientifically useful relative to the explanatory research goals. I argue that this is the case if by categorizing an unconscious phenomenon as “mental” one picks out explanatorily relevant similarities (...) to a corresponding paradigmatically mental conscious phenomenon. Explanatory relevance is spelled out in terms of mechanistic norms. (shrink)

We provide two programmatic frameworks for integrating philosophical research on understanding with complementary work in computer science, psychology, and neuroscience. First, philosophical theories of understanding have consequences about how agents should reason if they are to understand that can then be evaluated empirically by their concordance with findings in scientific studies of reasoning. Second, these studies use a multitude of explanations, and a philosophical theory of understanding is well suited to integrating these explanations in illuminating ways.

The new mechanists and the autonomy approach both aim to account for how biological phenomena are explained. One identifies appeals to how components of a mechanism are organized so that their activities produce a phenomenon. The other directs attention towards the whole organism and focuses on how it achieves self-maintenance. This paper discusses challenges each confronts and how each could benefit from collaboration with the other: the new mechanistic framework can gain by taking into account what happens outside individual mechanisms, (...) while the autonomy approach can ground itself in biological research into how the actual components constituting an autonomous system interact and contribute in different ways to realize and maintain the system. To press the case that these two traditions should be constructively integrated we describe how three recent developments in the autonomy tradition together provide a bridge between the two traditions: (1) a framework of work and constraints, (2) a conception of function grounded in the organization of an autonomous system, and (3) a focus on control. (shrink)

The history of developmental biology is interwoven with debates as to whether mechanistic explanations of development are possible or whether alternative explanatory principles or even vital forces need to be assumed. In particular, the demonstrated ability of embryonic cells to tune their developmental fate precisely to their relative position and the overall size of the embryo was once thought to be inexplicable in mechanistic terms. Taking a causal perspective, this Element examines to what extent and how developmental biology, having turned (...) molecular about four decades ago, has been able to meet the vitalist challenge. It focuses not only on the nature of explanations but also on the usefulness of causal knowledge – including the knowledge of classical experimental embryology – for further scientific discovery. It also shows how this causal perspective allows us to understand the nature and significance of some key concepts, including organizer, signal and morphogen. This title is also available as Open Access on Cambridge Core. (shrink)

What kinds of norms constrain mechanistic discovery and explanation? In the mechanistic literature, the norms for good explanations are directly derived from answers to the metaphysical question of what explanations are. Prominent mechanistic accounts thus emphasize either ontic or epistemic norms. Still, mechanistic philosophers on both sides agree that there is no sharp distinction between the processes of discovery and explanation. Thus, it seems reasonable to expect that ontic and epistemic accounts of explanation will be accompanied by ontic and epistemic (...) accounts of discovery, respectively. As we will show here, however, recent discovery accounts implicitly rely on both ontic and epistemic norms to characterize the discovery process. In this paper, we develop an account that makes explicit that, and how, ontic and epistemic norms work together throughout the discovery process. By describing mechanism discovery as a process of pattern recognition we demonstrate that scientists have to develop epistemic activities to distinguish a pattern from its background. Furthermore, they have to determine which epistemic activities successfully describe how the pattern is implemented by identifying the pattern’s components. Our approach reveals that ontic and epistemic norms are equally important in mechanism discovery. (shrink)

Alex Rosenberg (2020). Reductionism and Mechanism. Cambridge University Press, 2020, pp. 75, £15 (paperback).

Philosophers of psychology debate, among other things, which psychological models, if any, are (or provide) mechanistic explanations. This should seem a little strange given that there is rough consensus on the following two claims: 1) a mechanism is an organized collection of entities and activities that produces, underlies, or maintains a phenomenon, and 2) a mechanistic explanation describes, represents, or provides information about the mechanism producing, underlying, or maintaining the phenomenon to be explained (i.e. the explanandum phenomenon) (Bechtel and Abrahamsen (...) 2005; Craver 2007). If there is a rough consensus on what mechanisms are and that mechanistic explanations describe, represent, or provide information about them, then how is there no consensus on which psychological models are (or provide) mechanistic explanations? Surely the psychological models that are mechanistic explanations are the models that describe, represent, or provide information about mechanisms. That is true, of course; the trouble arises when determining what exactly that involves. Philosophical disagreement over which psychological models are mechanistic explanations is often disagreement about what it means to describe, represent, or provide information about a mechanism, among other things (Hochstein 2016; Levy 2013). In addition, one's position in this debate depends on a host of other seemingly arcane metaphysical issues, such as the nature of mechanisms, computational and functional properties (Piccinini 2016), and realization (Piccinini and Maley 2014), as well as the relation between models, methodologies, and explanations (Craver 2014; Levy 2013; Zednik 2015). Although I inevitably advocate a position, my primary aim in this chapter is to spell out all these relationships and canvas the positions that have been taken (or could be taken) with respect to mechanistic explanation in psychology, using dynamical systems models and cognitive models (or functional analyses) as examples. (shrink)

An important strand in philosophy of science takes scientific explanation to consist in the conveyance of some kind of information. Here I argue that this idea is also implicit in some core arguments of mechanists, some of whom are proponents of an ontic conception of explanation that might be thought inconsistent with it. However, informational accounts seem to conflict with some lay and scientific commonsense judgments and a central goal of the theory of explanation, because information is relative to the (...) background knowledge of agents. Sometimes we make lay judgments about whether a model is an explanation simpliciter, not just an explanation relative to some particular agent. And as philosophers of explanation, we would like a philosophical account to tell us when a model is an explanation simpliciter, not just when a model is an explanation relative to some particular agent. Thus, even if one’s account of explanation is not concerned with explanation qua communicative or speech act, the account’s reliance on the concept of information generates a prima facie conflict between the claims that 1) explanation is the conveyance of information, 2) information is relative to the background knowledge of an agent, and 3) some models are explanations not relative to the background knowledge of any particular agent. I sketch a solution to this puzzle by distinguishing informationally what I call “explanation simpliciter” from what I call “explanation-to,” relativizing the latter to an individual’s background knowledge and the former to what I call “total scientific background knowledge”. (shrink)

Borsboom and colleagues have recently proposed a “network theory” of psychiatric disorders that conceptualizes psychiatric disorders as relatively stable networks of causally interacting symptoms. They have also claimed that the network theory should include non-symptom variables such as environmental factors. How are environmental factors incorporated in the network theory, and what kind of explanations of psychiatric disorders can such an “extended” network theory provide? The aim of this article is to critically examine what explanatory strategies the network theory that includes (...) both symptoms and environmental factors can accommodate. We first analyze how proponents of the network theory conceptualize the relations between symptoms and between symptoms and environmental factors. Their claims suggest that the network theory could provide insight into the causal mechanisms underlying psychiatric disorders. We assess these claims in light of network analysis, Woodward’s interventionist theory, and mechanistic explanation, and show that they can only be satisfied with additional assumptions and requirements. Then, we examine their claim that network characteristics may explain the dynamics of psychiatric disorders by means of a topological explanatory strategy. We argue that the network theory could accommodate topological explanations of symptom networks, but we also point out that this poses some difficulties. Finally, we suggest that a multilayer network account of psychiatric disorders might allow for the integration of symptoms and non-symptom factors related to psychiatric disorders and could accommodate both causal/mechanistic and topological explanations. (shrink)

It is generally acknowledged by proponents of ‘new mechanism’ that mechanistic explanation involves adopting a perspective, but there is less agreement on how we should understand this perspective-taking or what its implications are for practising science. This paper examines the perspectival nature of mechanistic explanation through the lens of the ‘mechanistic stance’, which falls somewhere between Dennett’s more familiar physical and design stance. We argue this approach implies three distinct and significant ways in which mechanistic explanation can be interpreted as (...) perspectival: ‘phenomenon perspectivism’, ‘pattern perspectivism’ and ‘hierarchy perspectivism’. We evaluate the strength of the perspective-dependency implied by each of these, and along the way, discuss their significance for wider debates within the new mechanism literature, such as the nature of function attribution and an ontic vs epistemic understanding of explanation. (shrink)

This paper adds to the philosophical literature on mechanistic explanation by elaborating two related explanatory functions of idealisation in mechanistic models. The first function involves explaining the presence of structural/organizational features of mechanisms by reference to their role as difference-makers for performance requirements. The second involves tracking counterfactual dependency relations between features of mechanisms and features of mechanistic explanandum phenomena. To make these functions salient, we relate our discussion to an exemplar from systems biological research on the mechanism for countering (...) heat shock—the heat shock response system—in Escherichia coli bacteria. This research also reinforces a more general lesson: ontic constraint accounts in the literature on mechanistic explanation provide insufficiently informative normative appraisals of mechanistic models. We close by outlining an alternative view on the explanatory norms governing mechanistic representation. (shrink)

The mechanistic view of computation contends that computational explanations are mechanistic explanations. Mechanists, however, disagree about the precise role that the environment – or the so-called “contextual level” – plays for computational explanations. We advance here two claims: Contextual factors essentially determine the computational identity of a computing system ; this means that specifying the “intrinsic” mechanism is not sufficient to fix the computational identity of the system. It is not necessary to specify the causal-mechanistic interaction between the system and (...) its context in order to offer a complete and adequate computational explanation. While the first claim has been discussed before, the second has been practically ignored. After supporting these claims, we discuss the implications of our contextualist view for the mechanistic view of computational explanation. Our aim is to show that some versions of the mechanistic view are consistent with the contextualist view, whilst others are not. (shrink)

I analyze a tension at the core of the mechanistic view of computation generated by its joint commitment to the medium independence of computational vehicles and to computational systems possessing teleological functions to compute. While computation is individuated in medium-independent terms, teleology is sensitive to the constitutive physical properties of vehicles. This tension spells trouble for the mechanistic view, suggesting that there can be no teleological functions to compute. I argue that, once considerations about the relevant function-bestowing factors for computational (...) systems are brought to bear, the tension dissolves: physical systems can have the teleological function to compute. (shrink)

This paper considers the way mathematical and computational models are used in network neuroscience to deliver mechanistic explanations. Two case studies are considered: Recent work on klinotaxis by Caenorhabditis elegans, and a longstanding research effort on the network basis of schizophrenia in humans. These case studies illustrate the various ways in which network, simulation and dynamical models contribute to the aim of representing and understanding network mechanisms in the brain, and thus, of delivering mechanistic explanations. After outlining this mechanistic construal (...) of network neuroscience, two concerns are addressed. In response to the concern that functional network models are non-explanatory, it is argued that functional network models are in fact explanatory mechanism sketches. In response to the concern that models which emphasize a network’s organization over its composition do not explain mechanistically, it is argued that this emphasis is both appropriate and consistent with the principles of mechanistic explanation. What emerges is an improved understanding of the ways in which mathematical and computational models are deployed in network neuroscience, as well as an improved conception of mechanistic explanation in general. (shrink)

In the last two decades few topics in philosophy of science have received as much attention as mechanistic explanation. A significant motivation for these accounts is that scientists frequently use the term “mechanism” in their explanations of biological phenomena. While scientists appeal to a variety of causal concepts in their explanations, many philosophers argue or assume that all of these concepts are well understood with the single notion of mechanism. This reveals a significant problem with mainstream mechanistic accounts– although philosophers (...) use the term “mechanism” interchangeably with other causal concepts, this is not something that scientists always do. This paper analyses two causal concepts in biology–the notions of “mechanism” and “pathway”–and how they figure in biological explanation. I argue that these concepts have unique features, that they are associated with distinct strategies of causal investigation, and that they figure in importantly different types of explanation. (shrink)

Recent work in explainable artificial intelligence (XAI) attempts to render opaque AI systems understandable through a divide-and-conquer strategy. However, this fails to illuminate how trained AI systems work as a whole. Precisely this kind of functional understanding is needed, though, to satisfy important societal desiderata such as safety. To remedy this situation, we argue, AI researchers should seek mechanistic interpretability, viz. apply coordinated discovery strategies familiar from the life sciences to uncover the functional organisation of complex AI systems. Additionally, theorists (...) should accommodate for the unique costs and benefits of such strategies in their portrayals of XAI research. (shrink)

Richard Levins’s (Am Sci 54(4):421–431, 1966) paper sets a landmark for the significance of scientific model-making in biology. Colombo and Palacios (Biol Philos 36(5):1–26. 10.1007/s10539-021-09818-x, 2021) have recently built their critique of the explanatory power of the Free Energy Principle on Levins’s insight into the relationship between generality, realism, and precision. This paper addresses the issue of the plausibility of biological explanations that are grounded in the Free Energy Principle (FEP) and deals with the question of the realist fortitude of (...) FEP’s theoretical framework. It indicates that what is required for establishing the plausibility of the explanation of a target system given a model of that system is the dosage or the harmony between the generality and accuracy of explanatory models. This would also provide a basis for seeing how scientific realism could be a viable option with respect to FEP. (shrink)

In this paper we identify six theses that constitute core results of philosophical investigation into the nature of mechanisms, and of the role that the search for and identification of mechanisms play in the sciences. These theses represent the fruits of the body of research that is now often called New Mechanism. We concisely present the main arguments for these theses. In the literature, these arguments are scattered and often implicit. Our analysis can guide future research in many ways: it (...) provides critics of New Mechanism with clear targets, it can reduce misunderstandings, it can clarify differences of opinion among New Mechanists and it helps to define a research agenda for New Mechanists. (shrink)

Mechanistic explanations are often said to explain because they reveal the causal structure of the world. Conversely, dynamical models supposedly lack explanatory power because they do not describe causal structure. The only way for dynamical models to produce causal explanations is via the 3M criterion: the model must be mapped onto a mechanism. This framing of the situation has become the received view around the viability of dynamical explanation. In this paper, I argue against this position and show that dynamical (...) models can themselves reveal causal structure and consequently produce nonmechanistic, dynamical explanations. Taking the example of cell fates from systems biology, I show how dynamical models, and specifically the attractor landscapes they describe, identify the causes of cell differentiation and explain why cells select particular fates. These dynamical features of the system better fit Woodward’s (2010, 2018) criteria of specificity and proportionality and make them the best candidate causes of cell fates than mechanisms. I also show how these causes are irreducible and inaccessible to mechanistic models, making 3M unworkable and counterproductive in this case. Dynamical models can reveal dynamical causes and thereby provide causal explanations. (shrink)

Enactivism advances an understanding of cognition rooted in the dynamic interaction between an embodied agent and their environment, whilst new mechanism suggests that cognition is explained by uncovering the organised components underlying cognitive capacities. On the face of it, the mechanistic model’s emphasis on localisable and decomposable mechanisms, often neural in nature, runs contrary to the enactivist ethos. Despite appearances, this paper argues that mechanistic explanations of cognition, being neither narrow nor reductive, and compatible with plausible iterations of ideas like (...) emergence and downward causation, are congruent with enactivism. Attention to enactivist ideas, moreover, may serve as a heuristic for mechanistic investigations of cognition. Nevertheless, I show how enactivism and approaches that prioritise mechanistic modelling may diverge in starting assumptions about the nature of cognitive phenomena, such as where the constitutive boundaries of cognition lie. (shrink)

A subarea of the debate over the nature of evolutionary theory addresses what the nature of the explanations yielded by evolutionary theory are. The statisticalist line is that the general principles of evolutionary theory are not only amenable to a mathematical interpretation but that they need not invoke causes to furnish explanations. Causalists object that construction of these general principles involves crucial causal assumptions. A recent view claims that some biological explanations are statistically autonomous explanations (SAEs) whereby phenomena are accounted (...) for statistically and which prescind from micro-causal details. I raise three major problems for this account and then advance a view which unifies SAEs as mathematical explanations: the MSAE view. The MSAE view not only resolves the issues bedeviling the original SAE account but serves to importantly broaden the class of non-causal explanations in population biology. (shrink)

Some philosophers argue that we should eschew cross-explanatory integrations of mechanistic, dynamicist, and psychological explanations in cognitive science, because, unlike integrations of mechanistic explanations, they do not deliver genuine, cognitive scientific explanations. Here I challenge this claim by comparing the theoretical virtues of both kinds of explanatory integrations. I first identify two theoretical virtues of integrations of mechanistic explanations—unification and greater qualitative parsimony—and argue that no cross-explanatory integration could have such virtues. However, I go on to argue that this is (...) only a problem for those who think that cognitive science aims to specify one fundamental structure responsible for cognition. For those who do not, cross-explanatory integration will have at least two theoretical virtues to a greater extent than integrations of mechanistic explanations: explanatory depth and applicability. I conclude that one’s views about explanatory integration in cognitive science cannot be segregated from one’s views about the explanatory task of cognitive science. (shrink)

This paper advances the view that some explanations in cognitive science are extra-mathematical explanations. Demonstrating the plausibility of this interpretation centers around certain efficient coding cases which ineliminably enlist information theoretic laws, facts and theorems to identify in-principle, mathematical constraints on neuronal information processing capacities. The explanatory structure in these cases is shown to parallel other putative instances of mathematical explanation. The upshot for cognitive mathematical explanations is thus two-fold: first, the view capably rebuts standard mechanistic objections to non-mechanistic explanation; (...) second, this view serves to importantly widen the possibility space for non-mechanistic forms of explanation in cognitive science. (shrink)

Cartwright et al. in European Journal for Philosophy of Science, 10 and the new mechanists agree that regular behaviors described in cp laws are generated by mechanisms. However, there is disagreement with regard to the two questions that Cartwright at al. ask: the epistemological question and the ontological question. Most importantly, Cartwright et al. argue that the explanation involved is a CL-explanation, while the new mechanists insist that mechanistic explanation and CL-explanation are competitors. In this reply, I will highlight some (...) worries regarding Cartwright et al.’s analysis of the relationship between mechanisms and cp laws and I will provide alternative answers to the two questions in line with the new mechanistic approach. (shrink)

Dynamical systems play a central role in explanations in cognitive neuroscience. The grounds for these explanations are hotly debated and generally fall under two approaches: non-mechanistic and mechanistic. In this paper, I first outline a neurodynamical explanatory schema that highlights the role of dynamical systems in cognitive phenomena. I next explore the mechanistic status of such neurodynamical explanations. I argue that these explanations satisfy only some of the constraints on mechanistic explanation and should be considered pseudomechanistic explanations. I defend this (...) argument against three alternative interpretations of the neurodynamical explanatory schema. The independent interpretation holds that neurodynamical explanations and mechanisms are independent. The constitutive interpretation holds that neurodynamical explanations are constitutive but otherwise non-mechanistic. Both the independent and constitutive interpretations fail to account for all the features of neurodynamical explanations. The partial interpretation assumes that the targets of dynamical systems models are mechanisms and so holds that neurodynamical explanations are incomplete because they lack mechanistic details. I contend instead that the targets of those models are dynamical systems distinct from mechanisms and defend this claim against several objections. I conclude with a defense of the pseudomechanistic interpretation and a discussion of the source of their explanatory power in relation to a causal-mechanical description of the world. (shrink)

The debate over the explanatory nature of cognitive models has been waged mostly between two factions: the mechanists and the dynamical systems theorists. The former hold that cognitive models are explanatory only if they satisfy a set of mapping criteria, particularly the 3M/3m* requirement. The latter have argued, pace the mechanists, that some cognitive models are both dynamical and constitute covering-law explanations. In this paper, I provide a minimal model interpretation of dynamical cognitive models, arguing that this both provides needed (...) clarity to the mechanist versus dynamicist divide in cognitive science and also paves the way towards further insights about scientific explanation generally. (shrink)

It is widely believed that mathematics carries a substantial part of the explanatory burden in science. However, mathematics can also play important heuristic roles of a different kind, being a source of new ideas and approaches, allowing us to build toy models, enhancing expressive power and providing fruitful conceptualizations. In this paper, we focus on the application of dynamical systems theory (DST) within the extended cognition (EC) field of cognitive science, considering this case study to be a good illustration of (...) a general phenomenon. In the paper, we justify both a negative and a positive claim. The negative claim is that dynamical systems theory hardly plays any explanatory role in EC research. We justify our claim by analyzing several accounts of the explanatory role of mathematics and stressing the way mathematical arguments are used in explanations. Our positive claim is that even though, for now, DST has no explanatory power in many of the EC approaches, it still plays an important heuristic role there. In particular, using mathematical notions improves the expressive power of the language and gives a sense of understanding of the phenomena under investigation. The case study of EC allows us to identify and analyze this important role of mathematics, which seems to be neglected in contemporary discussions. (shrink)

Although there is a substantial philosophical literature on dynamical systems theory in the cognitive sciences, the same is not the case for neuroscience. This paper attempts to motivate increased discussion via a set of overlapping issues. The first aim is primarily historical and is to demonstrate that dynamical systems theory is currently experiencing a renaissance in neuroscience. Although dynamical concepts and methods are becoming increasingly popular in contemporary neuroscience, the general approach should not be viewed as something entirely new to (...) neuroscience. Instead, it is more appropriate to view the current developments as making central again approaches that facilitated some of neuroscience’s most significant early achievements, namely, the Hodgkin–Huxley and FitzHugh–Nagumo models. The second aim is primarily critical and defends a version of the “dynamical hypothesis” in neuroscience. Whereas the original version centered on defending a noncomputational and nonrepresentational account of cognition, the version I have in mind is broader and includes both cognition and the neural systems that realize it as well. In view of that, I discuss research on motor control as a paradigmatic example demonstrating that the concepts and methods of dynamical systems theory are increasingly and successfully being applied to neural systems in contemporary neuroscience. More significantly, such applications are motivating a stronger metaphysical claim, that is, understanding neural systems as being dynamical systems, which includes not requiring appeal to representations to explain or understand those phenomena. Taken together, the historical claim and the critical claim demonstrate that the dynamical hypothesis is undergoing a renaissance in contemporary neuroscience. (shrink)

The advancement of computing technology makes it possible to build extremely accurate digital reconstructions of brain circuits. Are such unprecedented levels of biological accuracy essential for brain simulations to play the roles they are expected to play in neuroscientific research? The main goal of this paper is to clarify this question by distinguishing between various roles played by large-scale simulations in contemporary neuroscience, and by reflecting about what makes a simulation biologically accurate. It is argued that large-scale simulations may play (...) model-oriented and prediction-oriented roles in brain research, and that the concept of biological accuracy can be interpreted as related to the plausibility of the theoretical model implemented in the simulation system, to the accuracy of the computer implementation, and to the level of details of the implemented model. Building on these observations and distinctions, it is argued that biological accuracy is not essential for a computer simulation to play the epistemic roles it is expected to play in brain research. (shrink)

A tension exists between those who do—e.g. Meyer (The British Journal for the Philosophy of Science 71:959–985, 2020 ) and Chemero ( 2011 )—and those who do not—e.g. Kaplan and Craver (Philosophy of Science 78:601–627, 2011 ) Piccinini and Craver (Synthese 183:283–311, 2011 )—afford nonmechanistic explanations a role in cognitive science. Here, I argue that one’s perspective on this matter will cohere with one’s interpretation of the tasks of cognitive science; that is, of the actions for which cognitive scientists are (...) specially fitted. To make this point concrete, I consider two tasks of cognitive science—establishing causal claims and unifying—and show how these tasks are interpreted differently by mechanists and nonmechanists respectively. I then argue that, as a result of these different interpretations, we cannot expect to resolve the tension between mechanists and nonmechanists by appeal to cognitive scientific practice. I conclude that we should give up on the debate between mechanists and nonmechanists, and search for progress elsewhere. (shrink)