This article defends the notion of downward causation, relating it to a notion of conditional independence.

We develop and apply a multi-dimensional account of explanatory depth towards a comparative analysis of inflationary and bouncing paradigms in primordial cosmology. Our analysis builds on earlier work due to Azhar and Loeb (2021) that establishes initial conditions fine-tuning as a dimension of explanatory depth relevant to debates in contemporary cosmology. We propose dynamical fine-tuning and autonomy as two further dimensions of depth in the context of problems with instability and trans-Planckian modes that afflict bouncing and inflationary approaches respectively. In (...) the context of the latter issue, we argue that the recently formulated trans-Planckian censorship conjecture leads to a trade-off for inflationary models between dynamical fine-tuning and autonomy. We conclude with the suggestion that explanatory preference with regard to the different dimensions of depth is best understood in terms of differing attitudes towards heuristics for future model building. (shrink)

If counterfactual dependence is sufficient for causation and if omissions can be causes, then all events have many more causes than common sense tends to recognize. This problem is standardly addressed by appeal to pragmatics. However, Carolina Sartorio [2010] has recently raised what I shall argue is a more interesting problem concerning omissions for counterfactual theories of causation—more interesting because it demands a more subtle pragmatic solution. I discuss the relationship between the idea that causes are proportional to their effects, (...) the idea that causation is contrastive, and the question of the dimensions along which causal explanations should be evaluated with respect to one another. (shrink)

We propose a formal framework for interpretable machine learning. Combining elements from statistical learning, causal interventionism, and decision theory, we design an idealised explanation game in which players collaborate to find the best explanation for a given algorithmic prediction. Through an iterative procedure of questions and answers, the players establish a three-dimensional Pareto frontier that describes the optimal trade-offs between explanatory accuracy, simplicity, and relevance. Multiple rounds are played at different levels of abstraction, allowing the players to explore overlapping causal (...) patterns of variable granularity and scope. We characterise the conditions under which such a game is almost surely guaranteed to converge on a optimal explanation surface in polynomial time, and highlight obstacles that will tend to prevent the players from advancing beyond certain explanatory thresholds. The game serves a descriptive and a normative function, establishing a conceptual space in which to analyse and compare existing proposals, as well as design new and improved solutions. (shrink)

We propose a formal framework for interpretable machine learning. Combining elements from statistical learning, causal interventionism, and decision theory, we design an idealisedexplanation gamein which players collaborate to find the best explanation(s) for a given algorithmic prediction. Through an iterative procedure of questions and answers, the players establish a three-dimensional Pareto frontier that describes the optimal trade-offs between explanatory accuracy, simplicity, and relevance. Multiple rounds are played at different levels of abstraction, allowing the players to explore overlapping causal patterns of (...) variable granularity and scope. We characterise the conditions under which such a game is almost surely guaranteed to converge on a (conditionally) optimal explanation surface in polynomial time, and highlight obstacles that will tend to prevent the players from advancing beyond certain explanatory thresholds. The game serves a descriptive and a normative function, establishing a conceptual space in which to analyse and compare existing proposals, as well as design new and improved solutions. (shrink)

Unsupervised learning algorithms are widely used for many important statistical tasks with numerous applications in science and industry. Yet despite their prevalence, they have attracted remarkably little philosophical scrutiny to date. This stands in stark contrast to supervised and reinforcement learning algorithms, which have been widely studied and critically evaluated, often with an emphasis on ethical concerns. In this article, I analyze three canonical unsupervised learning problems: clustering, abstraction, and generative modeling. I argue that these methods raise unique epistemological and (...) ontological questions, providing data-driven tools for discovering natural kinds and distinguishing essence from contingency. This analysis goes some way toward filling the lacuna in contemporary philosophical discourse on unsupervised learning, as well as bringing conceptual unity to a heterogeneous field more often described by what it isnot(i.e., supervised or reinforcement learning) than by what itis. I submit that unsupervised learning is not just a legitimate subject of philosophical inquiry but perhaps the most fundamental branch of all AI. However, an uncritical overreliance on unsupervised methods poses major epistemic and ethical risks. I conclude by advocating for a pragmatic, error-statistical approach that embraces the opportunities and mitigates the challenges posed by this powerful class of algorithms. (shrink)

William Whewell’s 19th century philosophy of science is sometimes glossed over as a footnote to Kant. There is however a key feature of Whewell’s account worth noting. This is his appeal to Aristotle’s form/matter hylomorphism as a metaphor to explain how mind and world merge in successful scientific inquiry. Whewell’s hylomorphism suggests a middle way between rationalism and empiricism reminiscent of experience pragmatists like Steven Levine’s view that mind and world are entwined in experience. I argue however that Levine does (...) not adequately explain exactly how mind and world entwine. He could nonetheless do so if he appealed to Whewell’s hylomorphic metaphor. We may prefer a reductive metaphysical explanation, but I suggest that pragmatists only have recourse to metaphor in this case. Both reductive and metaphorical explanations can enjoy great explanatory power if they exhibit a suitable measure of what I will call sematic distance. Semantic distance measures how close or how far apart explanandum and explanans are from each other in meaning. Metaphorical explanation - as evident in Whewell’s hylomorphism and as detailed via the notion of semantic distance - presents a valuable new explanatory tool to those who hold that mind and world are entwined sans recourse to metaphysics. (shrink)

In ‘Proportionality in Causation, Part I: Theories’, I presented various ways of understanding the idea that causes which are ‘proportional’ to their effects are in some sense preferable. In this companion article, I discuss the principal applications of the resulting theories of proportionality, and the challenges they face.

A much-discussed idea in the causation literature is that it is preferable to invoke causes which are proportional to—neither too general nor too specific for—the effect. This article presents various ways of understanding this idea. In what sense are such causal claims ‘preferable’? And what is it for one event to be ‘proportional’ to another? In a companion article, ‘Proportionality in Causation, Part II: Applications and Challenges’, I discuss the principal applications of the resulting theories of proportionality, and the challenges (...) they face. (shrink)

Proportionality—the idea that causes are neither too general nor too specific for their effects—seems to recommend implausibly disjunctive causes (McGrath, 1998 ; Shapiro & Sober, 2012 ; Franklin-Hall, 2016 ). I argue that this problem should be avoided by appeal to the notion of cohesion. I propose an account of cohesion in terms of the similarity structure of property-spaces, argue that it is not objectionably mysterious, and that alternative approaches—based on naturalness, interventionism, and contrastivism—are inadequate without appeal to it. In (...) an appendix, I show how my proposal can be perspicuously formalized by adapting structural equation models. (shrink)

When once-successful physical theories are abandoned, common wisdom has it that their characteristic theoretical entities are abandoned with them: examples include phlogiston, light rays, Newtonian forces, Euclidean space. But sometimes a theory sees ongoing use, despite being superseded. What should scientific realists say about the characteristic entities of the theories in such cases? The standard answer is that these ‘theoretical relicts’ are merely useful fictions. In this paper we offer a different answer. We start by distinguishing horizontal reduction (in which (...) a superseded theory approximates the successor theory) from vertical reduction (in which a higher-level theory abstracts away from the lower-level theory, but nonetheless can be constructed from it); these are usually regarded as having different ontological consequences. We describe a ‘verticalization’ procedure which transforms horizontal reductions into vertical reductions. The resulting verticalized theories are abstractions rather than approximations, with restricted domains. We identify a sense in which the higher-level theory describes distinct subject matters from the lower-level theory, enabling in certain cases the higher-level theory to retain distinctive explanatory power even in the presence of reduction. We suggest that theoretical entities from superseded theories should be retained in a scientific realist worldview just when, reinterpreted as higher-level abstractions, those theories and their characteristic entities continue to perform distinctive explanatory work in providing the best explanation for less-fundamental phenomena of interest. In slogan form: a good relict is an emergent relict. (shrink)

Recently philosophers of science have begun to pay more attention to the use of highly idealized mathematical models in scientific theorizing. An important example of this kind of highly idealized modeling is the widespread use of optimality models within evolutionary biology. One way to understand the explanations provided by these models is as a censored causal explanation: an explanation that omits certain causal factors in order to focus on a modular subset of the causal processes that led to the explanandum. (...) In this paper, I first argue that the censored causal model approach fails to establish a permanent explanatory role for optimality models in biology and mischaracterizes the explanatory virtues of biological optimality modeling. In addition, I argue that many biological optimality explanations cannot be characterized as censored causal explanations. In response, I propose an alternative approach that analyzes optimality models’ reliance on synchronically representing a system’s constraints and tradeoffs as well as their employment of various kinds of idealization in order to provide equilibrium explanations. (shrink)

A prominent approach to scientific explanation and modeling claims that for a model to provide an explanation it must accurately represent at least some of the actual causes in the event's causal history. In this paper, I argue that many optimality explanations present a serious challenge to this causal approach. I contend that many optimality models provide highly idealized equilibrium explanations that do not accurately represent the causes of their target system. Furthermore, in many contexts, it is in virtue of (...) their independence of causes that optimality models are able to provide a better explanation than competing causal models. Consequently, our account of explanation and modeling must expand beyond the causal approach. (shrink)

In this article, network science is discussed from a methodological perspective, and two central theses are defended. The first is that network science exploits the very properties that make a system complex. Rather than using idealization techniques to strip those properties away, as is standard practice in other areas of science, network science brings them to the fore, and uses them to furnish new forms of explanation. The second thesis is that network representations are particularly helpful in explaining the properties (...) of non-decomposable systems. Where part-whole decomposition is not possible, network science provides a much-needed alternative method of compressing information about the behavior of complex systems, and does so without succumbing to problems associated with combinatorial explosion. The article concludes with a comparison between the uses of network representation analyzed in the main discussion, and an entirely distinct use of network representation that has recently been discussed in connection with mechanistic modeling. (shrink)

It is often argued that biological generalizations have a distinctive and special status by comparison with the generalizations of other natural sciences, such as that biological generalizations are riddled with exceptions defying systematic and simple treatment. This special status of biology is used as a premise in arguments that posit a deprived explanatory, nomological, or methodological status in the biological sciences. I will discuss the traditional and still almost universally held idea that the biological sciences cannot deal with exceptions and (...) application conditions of their generalizations with their own distinctive and proprietary explanantia, but need the help of lower-level sciences to carry out this task. The idea of lower-level explanations of exceptions is connected to the idea that the biological sciences need lower-level sciences to better themselves and to the idea that biological sciences cannot provide reliable and extrapolatable results or explanations by themselves. I present counterexamples to the idea of lower-level explanations of exceptions in biology. I also discuss and refute more general arguments why the idea of lower-level explanations of exceptions has been held to hold in the special sciences, such as the screening-off and openness arguments. This suggest that there might be nothing special about the biological sciences vis-à-vis the more fundamental natural sciences, such as physics insofar as explanations of exceptions are concerned, and that the biological sciences can provide reliable and extrapolatable results or explanations by themselves. (shrink)

In addition to their core explanatory and predictive assumptions, scientific models include simplifying assumptions, which function as idealizations, approximations, and abstractions. There are methods to investigate whether simplifying assumptions bias the results of models, such as robustness analyses. However, the equally important issue – the focus of this paper – has received less attention, namely, what are the methodological and epistemic strengths and limitations associated with different simplifying assumptions. I concentrate on one type of simplifying assumption, the use of mega (...) parameters as abstractions in ecological models. First, I argue that there are two kinds of mega parameters qua abstractions, sufficient parameters and aggregative parameters, which have gone unnoticed in the literature. The two are associated with different heuristics, holism and reductionism, which many view as incompatible. Second, I will provide a different analysis of abstractions and the associated heuristics than previous authors. Reductionism and holism and the accompanying abstractions have different methodological and epistemic functions, strengths, and limitations, and the heuristics should be viewed as providing complementary research perspectives of cognitively limited beings. This is then, third, used as a premise to argue for epistemic and methodological pluralism in theoretical ecology. Finally, the presented taxonomy of abstractions is used to comment on the current debate whether mechanistic accounts of explanation are compatible with the use of abstractions. This debate has suffered from an abstract discussion of abstractions. With a better taxonomy of abstractions the debate can be resolved. (shrink)

It is argued that explanations of shock waves display explanatory emergence in two different ways. Firstly, the use of discontinuities to model jumps in flow variables is an example of “physics avoidance”. This is where microphysical details can be ignored in an abstract model thus allowing us access to modal information which cannot be attained in principle in any other way. Secondly, Whitham’s interleaving criterion for continuous shock structure is an example of the way different characteristic scales interact in shock (...) dynamics. To fully explain the shock structure one must take account of these different scales, and by doing so explanations of shock structure have irreducible aspects. Lastly, the implications of this explanatory irreducibility are examined in the context of explanatory indispensability arguments used in realism debates elsewhere in the philosophy of science. It is concluded that explanatory emergence on its own only supports an epistemic form of emergence. Yet this epistemic emergence is fully objective. (shrink)

Dimensional analysis can offer us explanations by allowing us to answer What-if–things-had-been-different? questions rather than in virtue of, say, unifying diverse phenomena, important as that is. Additionally, it is argued that dimensional analysis is a form of modelling as it involves several of the aspects crucial in modelling, such as misrepresenting aspects of a target system. By highlighting the continuities dimensional analysis has with forms of modelling we are able to describe more precisely what makes dimensional analysis explanatory and understand (...) otherwise puzzling aspects of dimensional reasoning, such as introducing fictitious dimensions and excluding dimensionally relevant information to characterise some systems. Finally, thinking of dimensional arguments as a form of modelling allows an explication of the role abstraction and multiple realisability; not as compatibility with other possible worlds but as compatibility with different fictional descriptions of our own world. (shrink)

It is commonly assumed that causation is transitive and in this paper I aim to reconcile this widely-held assumption with apparent evidence to the contrary. I will discuss a familiar approach to certain well-known counterexamples, before introducing a more resistant sort of case of my own. I will then offer a novel solution, based on Yablo’s proportionality principle, that succeeds in even these more resistant cases. There is a catch, however. Either proportionality is a constraint on which causal claims are (...) true, and the solution works, or it is not and causation is not transitive after all. I will argue that the first horn has unacceptable consequences and should be rejected, but that the second horn is less costly than it might initially appear. (shrink)

_ Source: _Page Count 14 Jonathan Vogel has argued in support of an explanationist response to skepticism in several works. Central to this explanationist response is the fact that our non-skeptical view of the world is a better explanation of our experiences than its skeptical rivals. Despite the plausibility of this response to skepticism, it is not without its critics. Recently, Matthew Gifford has argued that Vogel’s response has problems on two fronts. First, Gifford argues that Vogel’s strategy for showing (...) that the rwh is a better explanation than its skeptical rivals fails. Second, he argues that if Vogel’s strategy does succeed, then it accomplishes too much—it removes skeptical doubts when it should not. I argue that Gifford is mistaken on both accounts. (shrink)

As a discipline of its own, the philosophy of science can be traced back to the founding of its academic journals, some of which go back to the first half of the twentieth century. While the discipline has been the object of many historical studies, notably focusing on specific schools or major figures of the field, little work has focused on the journals themselves. Here, we investigate contemporary philosophy of science by means of computational text-mining approaches: we apply topic-modeling algorithms (...) to eight major philosophy of science journals, from the 1930s up until 2017. Based on the full-text content of some 15,897 articles, we identified 25 research themes and 8 thematic clusters that show how the research agenda of the philosophy of science has changed in its content over the course of the last eight decades, up to the philosophy of science we now know. We also show how each one of the journals contributed in its own way to this thematic evolution. (shrink)

Our ordinary causal concept seems to fit poorly with how our best physics describes the world. We think of causation as a time-asymmetric dependence relation between relatively local events. Yet fundamental physics describes the world in terms of dynamical laws that are, possible small exceptions aside, time symmetric and that relate global time slices. My goal in this paper is to show why we are successful at using local, time-asymmetric models in causal explanations despite this apparent mismatch with fundamental physics. (...) In particular, I will argue that there is an important connection between time asymmetry and locality, namely: understanding the locality of our causal models is the key to understanding why the physical time asymmetries in our universe give rise to time asymmetry in causal explanation. My theory thus provides a unified account of why causation is local and time asymmetric and thereby enables a reply to Russell’s famous attack on causation. (shrink)

Scientific models are often said to be more or less general depending on how many cases they cover. In this paper we argue that the cardinality of cases is insufficient as a metric of generality, and we present a novel account based on measure theory. This account overcomes several problems with the cardinality approach, and additionally provides some insight into the nature of assessments of generality. Specifically, measure theory affords a natural and quantitative way of describing local spaces of possibility. (...) The generality of models can be understood as the measure of possibilities to which the models apply. In order to illustrate our view, we consider the example of structural genomics, the ongoing project of building three-dimensional models of biological macromolecules like proteins. Using measure theory, we interpret the practice of homology modeling, where such models are understood to apply widely but imperfectly to the space of possible proteins. (shrink)

In the last couple of years a few seemingly independent debates on scientific explanation have emerged, with several key questions that take different forms in different areas. For example, the questions what makes an explanation distinctly mathematical and are there any non-causal explanations in sciences sometimes take a form of the question what makes mathematical models explanatory, especially whether highly idealized models in science can be explanatory and in virtue of what they are explanatory. These questions raise further issues about (...) counterfactuals, modality, and explanatory asymmetries: i.e., do mathematical and non-causal explanations support... (shrink)

In the last couple of years a few seemingly independent debates on scientific explanation have emerged, with several key questions that take different forms in different areas. For example, the question what makes an explanation distinctly mathematical and are there any non-causal explanations in sciences (i.e. explanations that don’t cite causes in the explanans) sometimes take a form of the question what makes mathematical models explanatory, especially whether highly idealized models in science can be explanatory and in virtue of what (...) they are explanatory. These questions raise further issues about counterfactuals, modality and explanatory asymmetries, i.e. do mathematical and non-causal explanations support counterfactuals, and how to understand explanatory asymmetries in non-causal explanations. Even though these are very common issues in the philosophy of physics and mathematics, they can be found in different guises in the philosophy of biology, where there is the statistical interpretation of the Modern Synthesis theory of evolution, according to which the post-Darwinian theory of natural selection explains evolutionary change by citing statistical properties of populations and not the causes of changes. These questions also arise in philosophy of ecology or neuroscience in regard to the nature of topological explanations. The question here is whether in network models in biology, ecology, neuroscience and computer science the mathematical or more precisely topological properties can be explanatory of physical phenomena, or they are just different ways to represent causal structures. The aim of the special issue is to unify all these debates around several overlapping questions. (shrink)

In this paper, I outline a heuristic for thinking about the relation between explanation and understanding that can be used to capture various levels of “intimacy”, between them. I argue that the level of complexity in the structure of explanation is inversely proportional to the level of intimacy between explanation and understanding, i.e. the more complexity the less intimacy. I further argue that the level of complexity in the structure of explanation also affects the explanatory depth in a similar way (...) to intimacy between explanation and understanding, i.e. the less complexity the greater explanatory depth and vice versa. (shrink)

This article considers the popular thesis that a more proportional relationship between a cause and its effect yields a more abstract causal explanation of that effect, which in turn produces a deeper explanation. This thesis is taken to have important implications for choosing the optimal granularity of explanation for a given explanandum. In this article, I argue that this thesis is not generally true of probabilistic causal relationships. In light of this finding, I propose a pragmatic, interest-relative measure of explanatory (...) depth. This measure uses a decision-theoretic model of information pricing to determine the optimal granularity of explanation for a given explanandum, agent, and decision problem. (shrink)

Proponents of various causal exclusion arguments claim that for any given event, there is often a unique level of granularity at which that event is caused. Against these causal exclusion arguments, causal ecumenists argue that the same event or phenomenon can be caused at multiple levels of granularity. This paper argues that the Bayesian network approach to representing the causal structure of target systems is consistent with causal ecumenism. Given the ubiquity of Bayesian networks as a tool for representing causal (...) structure in both philosophy of science and science itself, this result speaks in favor of the ecumenical view, and against rival exclusionary accounts. Gebharter’s (Philos Phenomenol Res 95(2):353–375, 2017) argument that the Bayes nets formalism is consistent with causal exclusion is considered and rebutted. (shrink)

The focus in the literature on scientific explanation has shifted in recent years towards model-based approaches. In recent work, Alisa Bokulich has argued that idealization has a central role to play in explanation. Bokulich claims that certain highly-idealized, structural models can be explanatory, even though they are not considered explanatory by causal, mechanistic, or covering law accounts of explanation. This paper focuses on Bokulich’s account in order to make the more general claim that there are problems with maintaining that a (...) structural criterion can capture the way that highly-idealized models explain. This paper examines Bokulich’s claim that the structural model explanation of quantum wavefunction scarring, featuring semiclassical mechanics, is deeper than the explanation provided by the local quantum model. The challenge for Bokulich is to show that the semiclassical model answers a wider range of w-questions, as this is her method of assessing structural information. I look at two reasonable approaches employing w-questions, and I argue that neither approach is ultimately satisfactory. Because structural similarity has preferences for more fundamental models, I argue that the local quantum model provides explanations that at least as deep as the semiclassical ones. The criterion either wrongly identifies all models as explanatory, or prefers models from fundamental theory. Either way, it cannot capture the way that highly-idealized models explain. (shrink)

Racial disparities in health outcomes have recently become a flashpoint in the debate about the value of race as a biological concept. What role, if any, race has in the etiology of disease is a philosophically and scientifically contested topic. In this article, I expand on the insights of the new mechanistic philosophy of science to defend a mechanism discovery approach to investigating epidemiological racial disparities. The mechanism discovery approach has explanatory virtues lacking in the populational approach typically employed in (...) the study of race and biomedicine. The explanatory constraints that form an integral part of the new mechanistic approach enable mechanism discovery to avoid the epistemic and normative shortcomings of the populational approach. The methodology of mechanism discovery can fruitfully be extended to the treatment and reversal of epidemiological racial disparities. (shrink)

A number of philosophers have recently suggested that some abstract, plausibly non-causal and/or mathematical, explanations explain in a way that is radically dif- ferent from the way causal explanation explain. Namely, while causal explanations explain by providing information about causal dependence, allegedly some abstract explanations explain in a way tied to the independence of the explanandum from the microdetails, or causal laws, for example. We oppose this recent trend to regard abstractions as explanatory in some sui generis way, and argue (...) that a prominent ac- count of causal explanation can be naturally extended to capture explanations that radically abstract away from microphysical and causal-nomological details. To this end, we distinguish di erent senses in which an explanation can be more or less abstract, and analyse the connection between explanations’ abstractness and their explanatory power. According to our analysis abstract explanations have much in common with counterfactual causal explanations. (shrink)

Scientific understanding is a fundamental goal of science. However, there is currently no good way to measure the scientific understanding of agents, whether these be humans or Artificial Intelligence systems. Without a clear benchmark, it is challenging to evaluate and compare different levels of scientific understanding. In this paper, we propose a framework to create a benchmark for scientific understanding, utilizing tools from philosophy of science. We adopt a behavioral conception of understanding, according to which genuine understanding should be recognized (...) as an ability to perform certain tasks. We extend this notion of scientific understanding by considering a set of questions that gauge different levels of scientific understanding, covering information retrieval, the capability to arrange information to produce an explanation, and the ability to infer how things would be different under different circumstances. We suggest building a Scientific Understanding Benchmark (SUB), formed by a set of these tests, allowing for the evaluation and comparison of scientific understanding. Benchmarking plays a crucial role in establishing trust, ensuring quality control, and providing a basis for performance evaluation. By aligning machine and human scientific understanding we can improve their utility, ultimately advancing scientific understanding and helping to discover new insights within machines. (shrink)

The interventionist account of causal explanation, in the version presented by Jim Woodward, has been recently claimed capable of buttressing the widely felt—though poorly understood—hunch that high-level, relatively abstract explanations, of the sort provided by sciences like biology, psychology and economics, are in some cases explanatorily optimal. It is the aim of this paper to show that this is mistaken. Due to a lack of effective constraints on the causal variables at the heart of the interventionist causal-explanatory scheme, as presently (...) formulated it is either unable to prefer high-level explanations to low, or systematically overshoots, recommending explanations at so high of a level as to be virtually vacuous. (shrink)

This essay develops an inferential account of model explanation, based on Mauricio Suárez’s inferential conception of scientific representation and Alisa Bokulich’s counterfactual account of model explanation. It is suggested that the fact that a scientific model can explain is essentially linked to how a modeler uses an established model to make various inferences about the target system on the basis of results derived from the model. The inference practice is understood as a two-step activity, with the first step involving making (...) counterfactual statements about the model itself and the second step involving making hypothetical statements transferring over claims derived from the model onto the target. To illustrate how this two-step activity proceeds, an agent-based simulation model is discussed. (shrink)

Recent years have seen increasing interest in interventionist analyses of metaphysical explanation. One area where interventionism traditionally shines, is in providing an account of explanatory depth; the sense in which explanation comes in degrees. However, the literature on metaphysical explanation has left the notion of depth almost entirely unexplored. In this paper I shall attempt to rectify this oversight by motivating an interventionist analysis of metaphysical explanatory depth (MED), in terms of the range of interventions under which a metaphysically explanatory (...) generalization remains invariant. After elucidating the notion through a toy-example, I demonstrate the important work which MED can perform in characterizing debate within contemporary metaphysics. Focusing upon rival approaches to explaining the identity and distinctness of concrete objects, I argue that the progress achieved in this debate can be characterized in terms of increasing explanatory depth. Having made an initial case for the utility of MED, I then turn this analysis to the metaphysics of explanation itself. By adopting an interventionist framework with respect to MED, I will show that we can assess the depth of competing theories of explanation. This application has two interesting results: first, it suggests that an interventionist analysis of explanation provides deeper explanations of the connection between explanans and explanandum than rival accounts; and second, it suggests that explanations provided by interventionism become deeper still, if one accepts that this methodology ranges over metaphysical, as well as causal, instances. (shrink)

I argue against the common and influential view that non-trivial chances arise only when the fundamental laws are indeterministic. The problem with this view, I claim, is not that it conflicts with some antecedently plausible metaphysics of chance or that it fails to capture our everyday use of ‘chance’ and related terms, but rather that it is unstable. Any reason for adopting the position that non-trivial chances arise only when the fundamental laws are indeterministic is also a reason for adopting (...) a much stronger, and far less attractive, position. I suggest an alternative account, according to which chances are probabilities that play a certain explanatory role: they are probabilities that explain associated frequencies. (shrink)

According to the reward-prediction error hypothesis of dopamine, the phasic activity of dopaminergic neurons in the midbrain signals a discrepancy between the predicted and currently experienced reward of a particular event. It can be claimed that this hypothesis is deep, elegant and beautiful, representing one of the largest successes of computational neuroscience. This paper examines this claim, making two contributions to existing literature. First, it draws a comprehensive historical account of the main steps that led to the formulation and subsequent (...) success of the RPEH. Second, in light of this historical account, it explains in which sense the RPEH is explanatory and under which conditions it can be justifiably deemed deeper than the incentive salience hypothesis of dopamine, which is arguably the most prominent contemporary alternative to the RPEH. (shrink)

In this paper, I examine Reinforcement Learning modelling practice in psychiatry, in the context of alcohol use disorders. I argue that the epistemic roles RL currently plays in the development of psychiatric classification and search for explanations of clinically relevant phenomena are best appreciated in terms of Chang’s account of epistemic iteration, and by distinguishing mechanistic and aetiological modes of computational explanation.

Some explanations are relatively abstract: they abstract away from the idiosyncratic or messy details of the case in hand. The received wisdom in philosophy is that this is a virtue for any explanation to possess. I argue that the apparent consensus on this point is illusory. When philosophers make this claim, they differ on which of four alternative varieties of abstractness they have in mind. What’s more, for each variety of abstractness there are several alternative reasons to think that the (...) variety of abstractness in question is a virtue. I identify the most promising reasons, and dismiss some others. The paper concludes by relating this discussion to the idea that explanations in biology, psychology and social science cannot be replaced by relatively micro explanations without loss of understanding. (shrink)

Some explanations in social science, psychology and biology belong to a higher level than other explanations. And higher explanations possess the virtue of abstracting away from the details of lower explanations, many philosophers argue. As a result, these higher explanations are irreplaceable. And this suggests that there are genuine higher laws or patterns involving social, psychological and biological states. I show that this ‘abstractness argument’ is really an argument schema, not a single argument. This is because the argument uses the (...) ‘is lower than’ relation, and this relation admits of different readings. I then suggest four rigorous definitions of the ‘is lower than’ relation, and show that the abstractness argument’s prospects are much brighter for some of these definitions than for others. To show this, I evaluate the so-called ‘disjunctive threat’ to the abstractness argument. (shrink)

Explainable artificial intelligence (XAI) aims to help people understand black box algorithms, particularly of their outputs. But what are these explanations and when is one explanation better than another? The manipulationist definition of explanation from the philosophy of science offers good answers to these questions, holding that an explanation consists of a generalization that shows what happens in counterfactual cases. Furthermore, when it comes to explanatory depth this account holds that a generalization that has more abstract variables, is broader in (...) scope and/or more accurate is better. By applying these definitions and contrasting them with alternative definitions in the XAI literature I hope to help clarify what a good explanation is for AI. (shrink)

There is a widely shared belief that the higher-level sciences can provide better explanations than lower-level sciences. But there is little agreement about exactly why this is so. It is often suggested that higher-level explanations are better because they omit details. I will argue instead that the preference for higher-level explanations is just a special case of our general preference for informative, logically strong, beliefs. I argue that our preference for informative beliefs entirely accounts for why higher-level explanations are sometimes (...) better—and sometimes worse—than lower-level explanations. The result is a step in the direction of the unity of science hypothesis. 1Introduction2Background: Is Omitting Details an Explanatory Virtue? 2.1Anti-reductionist arguments2.2Reductionist argument2.3Logical strength3Bases, Links and Logical Strength4Functionalism and Fodor’s Argument5Two Generalizations6Should the Base Really Be Maximally Strong?7Anti-reductionist Arguments Regarding the Base8Should the Antecedent of the Link Really Be Maximally Weak? (shrink)

Much recent work on explanation in the interventionist tradition emphasizes the explanatory value of stable causal generalizations—i.e., causal generalizations that remain true in a wide range of background circumstances. We argue that two separate explanatory virtues are lumped together under the heading of `stability’. We call these two virtues breadth and guidance respectively. In our view, these two virtues are importantly distinct, but this fact is neglected or at least under-appreciated in the literature on stability. We argue that an adequate (...) theory of explanatory goodness should recognize breadth and guidance as distinct virtues, as breadth and guidance track different ideals of explanation, satisfy different cognitive and pragmatic ends, and play different theoretical roles in helping us understand the explanatory value of mechanisms. Thus keeping track of the distinction between these two forms of stability yields a more accurate and perspicuous picture of the role that stability considerations play in explanation. (shrink)

Theories of explanation need to account for a puzzling feature of our explanatory practices: the fact that we prefer explanations that are relatively abstract but only moderately so. Contra Franklin-Hall ([2016]), I argue that the interventionist account of explanation provides a natural and elegant explanation of this fact. By striking the right balance between specificity and generality, moderately abstract explanations optimally subserve what interventionists regard as the goal of explanation, namely identifying possible interventions that would have changed the explanandum.

Explaining the supervenience of the moral on the natural is, perhaps, the central metaphysical challenge for the non-naturalist. However, Scanlon (2014) and Fogal and Risberg (2020) have developed a strategy which purports to explain supervenience rather simply. Fogal and Risberg call it the 'Divide and Conquer' strategy. The key idea is to postulate explanatory moral principles linking the natural and the moral. The moral principles are metaphysically necessary, so trivially supervene on the natural. All other moral facts are determined by (...) the necessary moral principles and the natural facts, so they supervene on the natural too. -/- I argue that there are two versions of the Divide and Conquer strategy and both fail. The first strategy doesn’t give an adequate explanation of supervenience because it doesn't properly identify what makes a difference to the holding of supervenience. The second perhaps does give an adequate explanation of supervenience, but only by introducing something similarly puzzling that needs explanation. -/- Consequently, non-naturalists need additional metaphysical machinery to develop their view -- just appealing to moral principles or metaphysical laws won't do the job. (shrink)

I give an account of what makes an event a coincidence. -/- I start by critically discussing a couple of other approaches to the notion of coincidence -- particularly that of Lando (2017) -- before developing my own view. The central idea of my view is that the correct understanding of coincidences is closely related to our understanding of the correct 'level' or 'grain' of explanation. Coincidences have a kind of explanatory deficiency — if they did not have this deficiency (...) they would not be coincidences. This deficiency, I claim, is the same explanatory deficiency as when we give low-level explanations of special science phenomena. Such explanations are typically too specific and not robust enough. I claim that there is this same badness in purported explanations of coincidences. -/- I cash out this idea sketching an account of explanatory goodness — an account of what makes explanations better or worse -- and using that to give a more precise account of coincidences. (shrink)

Why do we value higher-level scientific explanations if, ultimately, the world is physical? An attractive answer is that physical explanations often cite facts that don’t make a difference to the event in question. I claim that to properly develop this view we need to commit to a type of deterministic chance. And in doing so, we see the theoretical utility of deterministic chance, giving us reason to accept a package of views including deterministic chance.

Scientific understanding is a fundamental goal of science. However, there is currently no good way to measure the scientific understanding of agents, whether these be humans or Artificial Intelligence systems. Without a clear benchmark, it is challenging to evaluate and compare different levels of scientific understanding. In this paper, we propose a framework to create a benchmark for scientific understanding, utilizing tools from philosophy of science. We adopt a behavioral conception of understanding, according to which genuine understanding should be recognized (...) as an ability to perform certain tasks. We extend this notion of scientific understanding by considering a set of questions that gauge different levels of scientific understanding, covering information retrieval, the capability to arrange information to produce an explanation, and the ability to infer how things would be different under different circumstances. We suggest building a Scientific Understanding Benchmark (SUB), formed by a set of these tests, allowing for the evaluation and comparison of scientific understanding. Benchmarking plays a crucial role in establishing trust, ensuring quality control, and providing a basis for performance evaluation. By aligning machine and human scientific understanding we can improve their utility, ultimately advancing scientific understanding and helping to discover new insights within machines. (shrink)