Recently, requirements for the explainability of software systems have gained prominence. One of ... more Recently, requirements for the explainability of software systems have gained prominence. One of the primary motivators for such requirements is that explainability is expected to facilitate stakeholders' trust in a system. Although this seems intuitively appealing, recent psychological studies indicate that explanations do not necessarily facilitate trust. Thus, explainability requirements might not be suitable for promoting trust. One way to accommodate this finding is, we suggest, to focus on trustworthiness instead of trust. While these two may come apart, we ideally want both: a trustworthy system and the stakeholder's trust. In this paper, we argue that even though trustworthiness does not automatically lead to trust, there are several reasons to engineer primarily for trustworthiness-and that a system's explainability can crucially contribute to its trustworthiness.
This book is about how scientists explain cognitive phenomena.¹ The account I develop is tailored... more This book is about how scientists explain cognitive phenomena.¹ The account I develop is tailored to scientific explanation in cognitive science, especially cognitive neuroscience. My project is primarily descriptive rather than normative in nature. I set out from the success stories of empirical research into memory and language processing, respectively. By systematic experimentation, scientists have made substantial progress on explaining these phenomena. My aim is to examine the research strategies behind this success and reflect them against the background of contemporary philosophy of science. I will draw two major conclusions. First, although often treated as a 'package deal' the most promising accounts of scientific explanation to date (mechanistic explanations and interventionism) are incompatible. Second, even if we succeed in rendering them compatible, they take into account only a very limited portion of the research practices scientists actually use to investigate cognitive phenomena. To understand the epistemology of scientific experimentation we must consider a wider range of experimental designs, however. This includes various observational and manipulative techniques scientists employ as well as different ways of data analysis. Taking a first step into this direction, I develop a catalog of experiments. This catalog characterizes different studies with respect to what kinds of manipulations they employ and which sets of research questions they can potentially answer. Along the way I will discuss how to best conceive of interlevel relations and how to modify interventionism such that it can be used to uncover different dependence relations. 1 This is not a book about metaphysics. When talking about "processes", "capacities", "phenomena", "events", and the like I use these notions in a non-technical sense unless otherwise indicated.
Modern computer systems are ubiquitous in contemporary life yet many of them remain opaque. This ... more Modern computer systems are ubiquitous in contemporary life yet many of them remain opaque. This poses significant challenges in domains where desiderata such as fairness or accountability are crucial. We suggest that the best strategy for achieving system transparency varies depending on the specific source of opacity prevalent in a given context. Synthesizing and extending existing discussions, we propose a taxonomy consisting of eight sources of opacity that fall into three main categories: architectural, analytical, and socio-technical. For each source, we provide initial suggestions as to how to address the resulting opacity in practice. The taxonomy provides a starting point for requirements engineers and other practitioners to understand contextually prevalent sources of opacity, and to select or develop appropriate strategies for overcoming them.
Why is it rational for scientists to pursue multiple models of a phenomenon at the same time? The... more 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.
To explain phenomena, scientists often use systematic manipulations. Philosophers of science have... more To explain phenomena, scientists often use systematic manipulations. Philosophers of science have recently focused on one particular kind of manipulation in this context: interventions into some factor X applied to see how they affect another factor Y. According to the mutual manipulability account, we can infer a constitutive relation between X and Y if we can affect Y by intervening into X and vice versa. Problematically though, we cannot employ interventions without knowing what to manipulate, how to target it, and where and how to assess the effects of our manipulations to begin with. And even if we know this, there are often practical limitations. Besides, observed manipulability alone usually underdetermines the precise dependence relation at issue. In this paper I suggest a way of handling these issues by thinking beyond interventions: empirical researchers often resort to non-interventionist manipulations to enable, prepare for, or supplement intervention studies. To capture this practice, I introduce the concept of mere interactions. Once we recognize the important role mere interactions play in scientific practice, we gain a better understanding of how scientists actually explain phenomena.
Accounts of mental disorders focusing either on the brain as neurophysiological substrate or on s... more Accounts of mental disorders focusing either on the brain as neurophysiological substrate or on systematic connections between symptoms are insufficient to account for the multifactorial nature of mental illnesses. Recently, multiplexes have been suggested to provide a holistic view of psychopathology that integrates data from different factors, at different scales, or across time. Intuitively, these multi-layered network structures present quite appealing models of mental disorders that can be constructed by powerful computational machinery based on increasing amounts of real-world data. In this paper, I systematically examine what challenges psychopathology models face and to what extent different species of psychopathology models can address them. My analysis highlights that while multiplexes, as they are usually conceived, appear promising, they suffer from the same problems as other approaches. To remedy this, I suggest, we must go a step further and combine different kinds of ...
Recently, requirements for the explainability of software systems have gained prominence. One of ... more Recently, requirements for the explainability of software systems have gained prominence. One of the primary motivators for such requirements is that explainability is expected to facilitate stakeholders' trust in a system. Although this seems intuitively appealing, recent psychological studies indicate that explanations do not necessarily facilitate trust. Thus, explainability requirements might not be suitable for promoting trust. One way to accommodate this finding is, we suggest, to focus on trustworthiness instead of trust. While these two may come apart, we ideally want both: a trustworthy system and the stakeholder's trust. In this paper, we argue that even though trustworthiness does not automatically lead to trust, there are several reasons to engineer primarily for trustworthiness -- and that a system's explainability can crucially contribute to its trustworthiness.
Non-signs compared to signs from an unknown sign language differentially engage the parietal cort... more Non-signs compared to signs from an unknown sign language differentially engage the parietal cortex of deaf native signers
The lexicons of British Sign Language (BSL) and Swedish Sign Language (SSL) only partially overla... more The lexicons of British Sign Language (BSL) and Swedish Sign Language (SSL) only partially overlap. This offers the opportunity of studying the neurocognitive systems that support the processing of ...
Explanations build on insights from scientific discoveries. But what does it take to make these d... more Explanations build on insights from scientific discoveries. But what does it take to make these discoveries and which norms guide scientific inquiry and explanation? Contemporary philosophy of science prominently features a mechanistic view according to which scientists discover mechanisms responsible for phenomena in the world by manipulation studies. However, there is disagreement about the primary norms for mechanistic explanations. According to the ontic conception explanations aim to get the causal structure of the world right. According to the epistemic conception explanations seek to render phenomena intelligible. We argue that both these are important norms but neither is fundamental. Drawing on John Haugeland's work and construing mechanisms as patterns we show that ontic and epistemic aspects of explanation are mutually implicating. Framing mechanistic discoveries in terms of pattern recognition, which forms part of a lager system of scientific practice, we also illustrate with what skills individual scientist contribute to scientific progress and the construction of mechanistic explanations spanning multiple levels. Squaring the mechanistic view with patterns not only illuminates the mechanistic picture, it also sheds light on Haugeland's pattern view.
Recently, requirements for the explainability of software systems have gained prominence. One of ... more Recently, requirements for the explainability of software systems have gained prominence. One of the primary motivators for such requirements is that explainability is expected to facilitate stakeholders' trust in a system. Although this seems intuitively appealing, recent psychological studies indicate that explanations do not necessarily facilitate trust. Thus, explainability requirements might not be suitable for promoting trust. One way to accommodate this finding is, we suggest, to focus on trustworthiness instead of trust. While these two may come apart, we ideally want both: a trustworthy system and the stakeholder's trust. In this paper, we argue that even though trustworthiness does not automatically lead to trust, there are several reasons to engineer primarily for trustworthiness-and that a system's explainability can crucially contribute to its trustworthiness.
This book is about how scientists explain cognitive phenomena.¹ The account I develop is tailored... more This book is about how scientists explain cognitive phenomena.¹ The account I develop is tailored to scientific explanation in cognitive science, especially cognitive neuroscience. My project is primarily descriptive rather than normative in nature. I set out from the success stories of empirical research into memory and language processing, respectively. By systematic experimentation, scientists have made substantial progress on explaining these phenomena. My aim is to examine the research strategies behind this success and reflect them against the background of contemporary philosophy of science. I will draw two major conclusions. First, although often treated as a 'package deal' the most promising accounts of scientific explanation to date (mechanistic explanations and interventionism) are incompatible. Second, even if we succeed in rendering them compatible, they take into account only a very limited portion of the research practices scientists actually use to investigate cognitive phenomena. To understand the epistemology of scientific experimentation we must consider a wider range of experimental designs, however. This includes various observational and manipulative techniques scientists employ as well as different ways of data analysis. Taking a first step into this direction, I develop a catalog of experiments. This catalog characterizes different studies with respect to what kinds of manipulations they employ and which sets of research questions they can potentially answer. Along the way I will discuss how to best conceive of interlevel relations and how to modify interventionism such that it can be used to uncover different dependence relations. 1 This is not a book about metaphysics. When talking about "processes", "capacities", "phenomena", "events", and the like I use these notions in a non-technical sense unless otherwise indicated.
Modern computer systems are ubiquitous in contemporary life yet many of them remain opaque. This ... more Modern computer systems are ubiquitous in contemporary life yet many of them remain opaque. This poses significant challenges in domains where desiderata such as fairness or accountability are crucial. We suggest that the best strategy for achieving system transparency varies depending on the specific source of opacity prevalent in a given context. Synthesizing and extending existing discussions, we propose a taxonomy consisting of eight sources of opacity that fall into three main categories: architectural, analytical, and socio-technical. For each source, we provide initial suggestions as to how to address the resulting opacity in practice. The taxonomy provides a starting point for requirements engineers and other practitioners to understand contextually prevalent sources of opacity, and to select or develop appropriate strategies for overcoming them.
Why is it rational for scientists to pursue multiple models of a phenomenon at the same time? The... more 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.
To explain phenomena, scientists often use systematic manipulations. Philosophers of science have... more To explain phenomena, scientists often use systematic manipulations. Philosophers of science have recently focused on one particular kind of manipulation in this context: interventions into some factor X applied to see how they affect another factor Y. According to the mutual manipulability account, we can infer a constitutive relation between X and Y if we can affect Y by intervening into X and vice versa. Problematically though, we cannot employ interventions without knowing what to manipulate, how to target it, and where and how to assess the effects of our manipulations to begin with. And even if we know this, there are often practical limitations. Besides, observed manipulability alone usually underdetermines the precise dependence relation at issue. In this paper I suggest a way of handling these issues by thinking beyond interventions: empirical researchers often resort to non-interventionist manipulations to enable, prepare for, or supplement intervention studies. To capture this practice, I introduce the concept of mere interactions. Once we recognize the important role mere interactions play in scientific practice, we gain a better understanding of how scientists actually explain phenomena.
Accounts of mental disorders focusing either on the brain as neurophysiological substrate or on s... more Accounts of mental disorders focusing either on the brain as neurophysiological substrate or on systematic connections between symptoms are insufficient to account for the multifactorial nature of mental illnesses. Recently, multiplexes have been suggested to provide a holistic view of psychopathology that integrates data from different factors, at different scales, or across time. Intuitively, these multi-layered network structures present quite appealing models of mental disorders that can be constructed by powerful computational machinery based on increasing amounts of real-world data. In this paper, I systematically examine what challenges psychopathology models face and to what extent different species of psychopathology models can address them. My analysis highlights that while multiplexes, as they are usually conceived, appear promising, they suffer from the same problems as other approaches. To remedy this, I suggest, we must go a step further and combine different kinds of ...
Recently, requirements for the explainability of software systems have gained prominence. One of ... more Recently, requirements for the explainability of software systems have gained prominence. One of the primary motivators for such requirements is that explainability is expected to facilitate stakeholders' trust in a system. Although this seems intuitively appealing, recent psychological studies indicate that explanations do not necessarily facilitate trust. Thus, explainability requirements might not be suitable for promoting trust. One way to accommodate this finding is, we suggest, to focus on trustworthiness instead of trust. While these two may come apart, we ideally want both: a trustworthy system and the stakeholder's trust. In this paper, we argue that even though trustworthiness does not automatically lead to trust, there are several reasons to engineer primarily for trustworthiness -- and that a system's explainability can crucially contribute to its trustworthiness.
Non-signs compared to signs from an unknown sign language differentially engage the parietal cort... more Non-signs compared to signs from an unknown sign language differentially engage the parietal cortex of deaf native signers
The lexicons of British Sign Language (BSL) and Swedish Sign Language (SSL) only partially overla... more The lexicons of British Sign Language (BSL) and Swedish Sign Language (SSL) only partially overlap. This offers the opportunity of studying the neurocognitive systems that support the processing of ...
Explanations build on insights from scientific discoveries. But what does it take to make these d... more Explanations build on insights from scientific discoveries. But what does it take to make these discoveries and which norms guide scientific inquiry and explanation? Contemporary philosophy of science prominently features a mechanistic view according to which scientists discover mechanisms responsible for phenomena in the world by manipulation studies. However, there is disagreement about the primary norms for mechanistic explanations. According to the ontic conception explanations aim to get the causal structure of the world right. According to the epistemic conception explanations seek to render phenomena intelligible. We argue that both these are important norms but neither is fundamental. Drawing on John Haugeland's work and construing mechanisms as patterns we show that ontic and epistemic aspects of explanation are mutually implicating. Framing mechanistic discoveries in terms of pattern recognition, which forms part of a lager system of scientific practice, we also illustrate with what skills individual scientist contribute to scientific progress and the construction of mechanistic explanations spanning multiple levels. Squaring the mechanistic view with patterns not only illuminates the mechanistic picture, it also sheds light on Haugeland's pattern view.
To explain phenomena, scientists systematically manipulate them. Recently , philosophers of scien... more To explain phenomena, scientists systematically manipulate them. Recently , philosophers of science have focused on interventions in this context. The basic idea behind interventions is that we can wiggle something to see what else will wiggle along with it. However, not all of the manipulations scientists use qualify as interventions in this sense. The current paper introduces the concept of mere interactions to capture non-interventionist manipulations. It will illustrate their use in scientific practice by looking at dierent examples, discuss the contrast to interventions, and illustrate how recognizing mere interactions can help contemporary philosophy of science meet some of the challenges it faces.
According to the mechanistic view, scientists explain phenomena by uncovering the mechanisms resp... more According to the mechanistic view, scientists explain phenomena by uncovering the mechanisms responsible for them (e.g. Machamer, Darden, and Craver 2000. What precisely this "being responsible for" means, however, is a matter of heated debate in contemporary philosophy of science (see e.g. Couch 2011, Harbecke 2010, Leuridan 2012, Kaiser and Krickel forthcoming). The suggestions vary significantly with the exact conceptions of "mechanism" different authors suggest. Most prominently perhaps, we are told to conceive of the relation between mechanisms and their phenomena in causal or constitutive terms. Strikingly, discussions about causal and constitutive relations in this context usually take the notion of "phenomenon" as a primitive. But even if we simply conceive of phenomena as the explananda of mechanistic explanations, we must realize that phenomena do not form a unified category. Mechanisms are indeed supposed to explain a whole collection of different things: processes with multiple phases, end products or final stages of processes, stables states, continuous or repetitive behaviors, even properties.
Explanations build on insights from scientific discoveries. But what does it take to make these d... more Explanations build on insights from scientific discoveries. But what does it take to make these discoveries and which norms guide scientific inquiry and explanation? Contemporary philosophy of science prominently features a mechanistic view according to which scientists discover mechanisms responsible for phenomena in the world by manipulation studies. However, there is disagreement about the primary norms for mechanistic explanations. According to the ontic conception explanations aim to get the causal structure of the world right. According to the epistemic conception explanations seek to render phenomena intelligible. We argue that both these are important norms but neither is fundamental. Drawing on John Haugeland's work and construing mechanisms as patterns we show that ontic and epistemic aspects of explanation are mutually implicating. Framing mechanistic discoveries in terms of pattern recognition, which forms part of a lager system of scientific practice, we also illustrate with what skills individual scientist contribute to scientific progress and the construction of mechanistic explanations spanning multiple levels. Squaring the mechanistic view with patterns not only illuminates the mechanistic picture, it also sheds light on Haugeland's pattern view.
Craver and Darden (2013) distinguish mechanisms that produce, underlie, and maintain phenomena. H... more Craver and Darden (2013) distinguish mechanisms that produce, underlie, and maintain phenomena. However, the status of and relations between them remain unclear. I argue that the explanations we seek are conceptually tied to our research questions. To explain how an end product or result is generated, we will usually search for the mechanism that produced it; to explain a process, we will typically search for the mechanism underlying it; and to explain how a system's stable state or continuous behavior is maintained, we search for the mechanism maintaining it. I highlight that each of these explanatory projects offers different benefits and show how they are systematically connected with one another.
Uploads
Papers by Lena Kästner