Epidemiological explanation often has a “black box” character, meaning the intermediate steps between cause and effect are unknown. Filling in black boxes is thought to improve causal inferences by making them intelligible. I argue that adding information about intermediate causes to a black box explanation is an unreliable guide to pragmatic intelligibility because it may mislead us about the stability of a cause. I diagnose a problem that I call wishful intelligibility, which occurs when scientists misjudge the (...) limitations of certain features of an explanation. Wishful intelligibility gives us a new reason to prefer black box explanations in some contexts. (shrink)

An oversized reliance on big data-driven algorithmic decision-making systems, coupled with a lack of critical inquiry regarding such systems, combine to create the paradoxical “black box” at work. The “black box” simultaneously demands a higher level of transparency from the worker in regard to data collection, while shrouding the decision-making in secrecy, making employer decisions even more opaque to the worker. To access employment, the worker is commanded to divulge highly personal information, and when hired, must submit further (...) still to algorithmic processes of evaluations which will make authoritative claims as to the workers’ productivity. Furthermore, in and out of the workplace, the worker is governed by an invisible data-created leash deploying wearable technology to collect intimate worker data. At all stages, the worker is confronted with a lack of transparency, accountability, or explanation as to the inner workings or even the logic of the “black box” at work. This data revolution of the workplace is alarming for several reasons: the “black box at work” not only serves to conceal disparities in hiring, but could also allow for a level of “data-laundering” that beggars any notion of equal opportunity in employment and there exists, the danger of a “mission creep” attitude to data collection that allows for pervasive surveillance, contributing to the erosion of both the personhood and autonomy of workers. Thus, the “black box at work” not only enables worker domination in the workplace, it deprives the worker of Rawlsian justice. (shrink)

Many of the computing systems programmed using Machine Learning are opaque: it is difficult to know why they do what they do or how they work. Explainable Artificial Intelligence aims to develop analytic techniques that render opaque computing systems transparent, but lacks a normative framework with which to evaluate these techniques’ explanatory successes. The aim of the present discussion is to develop such a framework, paying particular attention to different stakeholders’ distinct explanatory requirements. Building on an analysis of “opacity” from (...) philosophy of science, this framework is modeled after accounts of explanation in cognitive science. The framework distinguishes between the explanation-seeking questions that are likely to be asked by different stakeholders, and specifies the general ways in which these questions should be answered so as to allow these stakeholders to perform their roles in the Machine Learning ecosystem. By applying the normative framework to recently developed techniques such as input heatmapping, feature-detector visualization, and diagnostic classification, it is possible to determine whether and to what extent techniques from Explainable Artificial Intelligence can be used to render opaque computing systems transparent and, thus, whether they can be used to solve the Black Box Problem. (shrink)

Although decision‐making algorithms are not new to medicine, the availability of vast stores of medical data, gains in computing power, and breakthroughs in machine learning are accelerating the pace of their development, expanding the range of questions they can address, and increasing their predictive power. In many cases, however, the most powerful machine learning techniques purchase diagnostic or predictive accuracy at the expense of our ability to access “the knowledge within the machine.” Without an explanation in terms of reasons or (...) a rationale for particular decisions in individual cases, some commentators regard ceding medical decision‐making to black box systems as contravening the profound moral responsibilities of clinicians. I argue, however, that opaque decisions are more common in medicine than critics realize. Moreover, as Aristotle noted over two millennia ago, when our knowledge of causal systems is incomplete and precarious—as it often is in medicine—the ability to explain how results are produced can be less important than the ability to produce such results and empirically verify their accuracy. (shrink)

The article comments on Peetz’ concept of relational mechanisms. This concept is an alternative to mechanistical explanations of analytical sociology, conceptualized as based on human agents. Peetz criticises this foundation, juxtaposing it with the idea of the analytical primacy of relations. This perspective does not necessarily presuppose agents but can explain their emergence. To demonstrate the efficiency of his concept, he presents an explanation of a concrete mechanism. The analysis of this explanation shows that a crucial point is missing (...) from the concept of relational mechanisms: the steps that produce a social process are never spelt out. Peetz thus presents a black box explanation, which is contrary to the demands of mechanistical explanations. His preference for black box argumentation is owed to his concepts. Unlike an enlightened methodological individualist, he is not in a position to explain the productions necessary for the formation of mechanistical processes. (shrink)

We introduce and discuss a knowledge-driven distillation approach to explaining black-box models by means of two kinds of interpretable models. The first is perceptron (or threshold) connectives, which enrich knowledge representation languages such as Description Logics with linear operators that serve as a bridge between statistical learning and logical reasoning. The second is Trepan Reloaded, an ap- proach that builds post-hoc explanations of black-box classifiers in the form of decision trees enhanced by domain knowledge. Our aim is, (...) firstly, to target a model-agnostic distillation approach exemplified with these two frameworks, secondly, to study how these two frameworks interact on a theoretical level, and, thirdly, to investigate use-cases in ML and AI in a comparative manner. Specifically, we envision that user-studies will help determine human understandability of explanations generated using these two frameworks. (shrink)

This paper argues that judges assessing the scientific validity and the legal admissibility of forensic science techniques ought to privilege testing over explanation. Their evaluation of reliability should be more concerned with whether the technique has been adequately validated by appropriate empirical testing than with whether the expert can offer an adequate description of the methods she uses, or satisfactorily explain her methodology or the theory from which her claims derive. This paper explores these issues within two specific contexts: latent (...) fingerprint examination and the use of breath tests for the detection of alcohol. Especially in the forensic science arena, I suggest courts have often been seduced by superficially plausible explanations and descriptions of a technique or method, and permitted these to serve as a substitute for empirical testing. Thinking through these two examples illustrates both why evaluating the extent of testing should be the most important method by which courts assess reliability, and why, when other forms of explanatory evidence are readily available, we may nonetheless elect to make use of them. This paper suggests that these descriptions and explanations may at times usefully supplement evidence of testing, but should not generally be substituted for it. Finally, this paper embraces a kind of evidentiary pragmatism, in which the quantum of evidence required to establish legal reliability is determined not in the abstract, but in relation to the evidence that is, or ought to be, available as a result of reasonable research and investigation. (shrink)

The use of opaque, uninterpretable artificial intelligence systems in health care can be medically beneficial, but it is often viewed as potentially morally problematic on account of this opacity—because the systems are black boxes. Alex John London has recently argued that opacity is not generally problematic, given that many standard therapies are explanatorily opaque and that we can rely on statistical validation of the systems in deciding whether to implement them. But is statistical validation sufficient to justify implementation of (...) these AI systems in health care, or is it merely one of the necessary criteria? I argue that accountability, which holds an important role in preserving the patient‐physician trust that allows the institution of medicine to function, contributes further to an account of AI system justification. Hence, I endorse the vanishing accountability principle: accountability in medicine, in addition to statistical validation, must be preserved. AI systems that introduce problematic gaps in accountability should not be implemented. (shrink)

This article argues that analytical sociology an approach that attempts to study social mechanisms ‘without, black boxes’ can benefit from the use of explanatory mixed methods. Analytical sociologists mainly relate their theoretical and agent-based models to representative surveys and experiments. While their central claim is to find and test the actual mechanisms that have produced the explanandum. the mechanisms they postulate often remain speculative. Neither agent-based models, nor experiments or mainstream quantitative methods, give access to some of the central (...) elements of the causal mechanisms and the relevant subjective and objective contextual parameters. One of the most important reasons for this lies in the fact that social reality is changing fast., characterized by strong diversity and complexified by the phenomenon of cultural meanings. I argue that by creating and testing the models of analytical sociology with explanatory mixed methods, researchers have the possibility of getting closer to their object of research and therefore of having the chance to create more valid explanations. (shrink)

In the first part of this article, I argue that Christopher Megone's natural-kind interpretation of Aristotle's argument that "the function of a human being is reason" does not resolve major puzzles about the argument, specifically the puzzles of why a human being has a function and why reason is that function. I attempt to resolve these puzzles by supplementing the natural-kind account with the doctrine that reason is the master regulatory natural function by which individuals enter into social life. In (...) the second half, I critique Megone's value account of function and argue for a nonevaluative, causal-explanatory account of function and teleological explanation. (shrink)

In the past few years, machine learning (ML) tools have been implemented with success in the medical context. However, several practitioners have raised concerns about the lack of transparency—at the algorithmic level—of many of these tools; and solutions from the field of explainable AI (XAI) have been seen as a way to open the ‘black box’ and make the tools more trustworthy. Recently, Alex London has argued that in the medical context we do not need machine learning tools to (...) be interpretable at the algorithmic level to make them trustworthy, as long as they meet some strict empirical desiderata. In this paper, we analyse and develop London’s position. In particular, we make two claims. First, we claim that London’s solution to the problem of trust can potentially address another problem, which is how to evaluate the reliability of ML tools in medicine for regulatory purposes. Second, we claim that to deal with this problem, we need to develop London’s views by shifting the focus from the opacity of algorithmic details to the opacity of the way in which ML tools are trained and built. We claim that to regulate AI tools and evaluate their reliability, agencies need an explanation of how ML tools have been built, which requires documenting and justifying the technical choices that practitioners have made in designing such tools. This is because different algorithmic designs may lead to different outcomes, and to the realization of different purposes. However, given that technical choices underlying algorithmic design are shaped by value-laden considerations, opening the black box of the design process means also making transparent and motivating (technical and ethical) values and preferences behind such choices. Using tools from philosophy of technology and philosophy of science, we elaborate a framework showing how an explanation of the training processes of ML tools in medicine should look like. (shrink)

Giving the impression that perichoresis solves the “threeness-oneness problem” or the “two natures–one person problem” without an explanation of how perichoresis works is problematic; as such, an explanation of perichoresis ought to be provided. I provide one way to address this problem by drawing upon the work of Eleonore Stump. In contrast to approaches that avoid the metaphysics of perichoresis I provide an account of the metaphysics of perichoresis and suggest that a Stump-inspired account of perichoresis—that is, an account that (...) places an emphasis on the notion of sharing some aspect of the mental life—deserves serious attention by those who feel the weight of the problematic use of perichoresis. (shrink)

It has long been a truism of the history of philosophy that intentionality is an invention of the medieval period, and within this standard narrative, the central place of Arabic philosophy has always been acknowledged. Yet there are many misconceptions surrounding the theories of intentionality advanced by the two main Arabic thinkers whose works were available to the West, Avicenna and Averroes. In the first part of this paper I offer an overview of the general accounts of intentionality and intentional (...) being found in the linguistic, psychological, and metaphysical writings of Avicenna and Averroes, and I trace the terminology of “intentions” to a neglected passage from Avicenna’s logic. In the second part of the paper I examine the way that Avicenna and Averroes apply their general theories of intentionality to the realm of sense perception. I offer an explanation of why Avicenna might have chosen to denominate the objects of the internal sense faculty of estimation as “intentions”, and I explore the implications of Averroes’s decision to attribute intentionality to the external senses and the media of perception. (shrink)

Many contemporary philosophers argue that assertion is governed by an epistemic norm. In particular, many defend the knowledge account of assertion, which says that one should assert only what one knows. Here, I defend a non‐normative alternative to the knowledge account that I call the repK account of assertion. According to the repK account, assertion represents knowledge, but it is not governed by a constitutive epistemic rule. I show that the repK account offers a more straightforward interpretation of the conversational (...) patterns and intuitions that motivate the knowledge account. It does so in terms of ordinary normative principles that philosophers already accept, none of which are constitutive to assertion. I then contend that the repK account is preferable to the knowledge account because it is simpler, its implications are less contentious, and it avoids a problem for normative accounts of assertion recently raised by Peter Pagin. I also argue that the repK account offers a satisfying explanation of selfless assertion, a counterexample to the knowledge account posed by Jennifer Lackey. (shrink)

According to a Moorean response to skepticism, the standards for knowledge are invariantly comparatively low, and we can know across contexts all that we ordinarily take ourselves to know. It is incumbent upon the Moorean to defend his position by explaining how, in contexts in which S seems to lack knowledge, S can nevertheless have knowledge. The explanation proposed here relies on a warranted-assertability maneuver: Because we are warranted in asserting that S doesn’t know that p, it can seem that (...) S does in fact lack that piece of knowledge. Moreover, this warranted-assertability maneuver is unique and better than similar maneuvers because it makes use of H. P. Grice’s general conversational rule of Quantity—“Do not make your contribution more informative than is required”—in explaining why we are warranted in asserting that S doesn’t know that p. (shrink)

Contextualist responses to skepticism about the external world are inadequate, and we should prefer an invariantist response to skepticism. There are two kinds of contextualism---anti-theoretical and theoretical. Anti-theoretical contextualists argue that the principles on which skepticism depends are absent from our ordinary epistemic ways of thinking. So anti-theoretical contextualists conclude that the burden of proof is on the skeptic. But some argue that the principles on which skepticism depends are not absent from our ordinary ways of thinking. The existence of (...) this unresolved dispute suggests that anti-theoretical contextualists have not shifted the burden of proof to the skeptic. Anti-theoretical contextualists also fail to refute skepticism outright. Their primary argument here is one in which the notion of certainty plays a key role. But their argument is unsound on all significant uses of the word 'certain.' Thus, anti-theoretical contextualists have neither refuted skepticism nor shifted the burden of proof to the skeptic. ;Theoretical contextualism claims that the solution to skeptical puzzles lies in the fact that the standards for knowledge shift from context to context. Yet if theoretical contextualist solutions are to be adequate, they must explain how those standards shift. Unfortunately, theoretical contextualism fails to provide this explanation. This explanatory failure, along with the problems that face individual theoretical contextualist accounts, shows that theoretical contextualist responses to skepticism are inadequate. ;Even though contextualism fails, a Moorean invariantism---according to which the standards for knowledge are always low---allows us to know across contexts the things we ordinarily take ourselves to know. We should prefer this Moorean invariantism over contextualism. It stands on the same theoretical ground as the dominant contextualisms while still allowing us to explain how we know that certain skeptical hypotheses are false. Also, contextualism raises questions---for example, questions concerning how the standards for knowledge shift---that receive only inadequate answers. The Moorean invariantist account does not raise such questions. All of this suggests that the Moorean invariantist response to skepticism is better than any of its contextualist counterparts. (shrink)

This book is outstandingly important for two reasons: first, because it offers a new explanation for the uniqueness of the West, and secondly because it develops in a radical way the notion of hist...

Museum displays tend to black-box science, by displaying scientific facts without explanations of how those facts were made. A recent trend in exhibit design upends this omission by putting scientists, technicians, and volunteers to work in glass-walled laboratories, just a window away from visitors. How is science conceived, portrayed, and performed in glass-walled laboratories? Interviews and participant observation in several “fishbowl” paleontology laboratories reveal that glass walls alter lab workers’ typical tasks and behavior. However, despite glass-walled labs’ incomplete (...) and edited enactment of scientific work, displaying an active workplace valuably challenges visitors’ assumptions that science is passive and that museums are home only to facts and dead things. Thus, glass-walled labs do not destroy the black box that obscures scientific practice for nonscientists. Instead, they exemplify a glass box, a kind of black box that contains a performance of scientific work. Glass-boxing describes a common way in which museums present scientific practice—that is, by making it observable but incompletely so—by inviting the public to construct a rich understanding of science as human work. (shrink)

This paper discusses various problems of explanations by mechanisms. Two positions are distinguished: the narrow position claims that only explanations by mechanisms are acceptable. It is argued that this position leads to an infinite regress because the discovery of a mechanism must entail the search for other mechanisms etc. Another paradoxical consequence of this postulate is that every successful explanation by mechanisms is unsatisfactory because it generates new ``black box'' explanations. The second – liberal – position (...) that is advanced in this paper regards, besides explanations by mechanisms, also the discovery of bivariate correlations as a first step of an explanation by mechanisms as meaningful. It is further argued that there is no contradiction between causal analysis and the explanation by mechanisms. Instead, explanations by mechanisms always presuppose the analysis of causal structures (but not vice versa). The final point is that an explanation by mechanisms is not inconsistent with the Hempel-Oppenheim scheme of explanation. (shrink)

In economics, models are built to answer specific questions. Each type of question requires its own type of models; in other words, it defines the requirements that a model should meet and thereby instructs how the models should be built. An explanation is an answer to a ‘why’-question. In economics, this answer is provided by a white -box model. To answer a ‘how much’-question, which is asking for a measurement, economists can make use of black-box models. Economic phenomena are (...) often so complex that white -box models that should function as explanations for them appear not to be intelligible. So, for questions concerning understanding - e.g. ‘Can you clarify this phenomenon‘’ ‘Can you tell me what would happen if‘’ - it would appear that grey-box models are more adequate. This implies a twofold revision of de Regt and Dieks’s criteria for understanding phenomena. First, whenever the term ‘theory’ is used it should be substituted by ‘model’. Secondly, the Criterion for the Intelligibility of Models is now simplified as: ‘A model M is intelligible for economists if they have built M as a grey-box construction’. A grey-box construction implies that they ‘can recognize qualitatively characteristic consequences of M without performing exact calculations’. (shrink)

In lieu of an abstract, here is a brief excerpt of the content:Reviewed by:The Dharma of Justice in the Sanskrit Epics: Debates on Gender, Varna, and Species by Ruth VanitaBrian Black (bio)The Dharma of Justice in the Sanskrit Epics: Debates on Gender, Varna, and Species. By Ruth Vanita. Oxford: Oxford Unity Press, 2021. Pp. 298. Hardcover £70.00, isbn 978-0-19-285982-2. Ruth Vanita's The Dharma of Justice in the Sanskrit Epics: Debates on Gender, Varna, and Species examines how the Mahābhārata and (...) Rāmāyaṇa discuss questions of justice. She reads the epics as bhakti texts to focus on their depictions of discrimination based on social categories such as gender, varṇa, species, age, and disability. The book claims that the epics criticize oppression in two main ways: by articulating philosophical ideas about social justice and by celebrating characters of a low social status. Six of the twelve chapters are devoted to themes related to gender, including marriage, independence, consent, gender-crossing, parenting, and masculinity. Focusing on the words and actions of characters in the narrative, Vanita does not directly address the question of whether we can take the marginal voices in the text as representing authentic voices of its social context. She also sidesteps the issue of patriarchy: "Every written text ever composed is, in the widest sense, patriarchal, since it was composed in a male-dominated society. Patriarchy is the air we breathe. Therefore, explaining the meaning of a text as patriarchal is as useful as explaining people's actions by the fact that they breathe" (p. 22). I was left wondering, however, if this explanation serves the interests of her study. In a book about oppressed characters, it seems crucial to say more about how systems of oppression operate. Even if we assume that all societies are patriarchal, this does not mean that they are all patriarchal in the same ways or to the same degree. Vanita is an imaginative reader, whose often speculative interpretations offer some valuable perspectives and raise interesting questions. One of the most refreshing features of the book is its extensive discussions of a number of characters and sub-stories rarely analyzed in scholarly literature. In particular, I enjoyed reading Vanita's take on the Mahābhārata's story of Diśā and Aṣṭāvakra (Mbh 13.19–22), where she shows her sensitivity to the ambiguities of this intriguing episode. This story, which appears in the Anuśāsana Parvan, begins when the ascetic Aṣṭāvakra wants to marry the sage Vadānya's daughter, Suprabhā. Before offering his consent, Vadānya sends Aṣṭāvakra to visit the aged female sage Diśā. As Vanita observes, this episode "maps a landscape of desire" [End Page 1] (p. 77) and raises a number of interesting questions about female sexuality as well as female independence. Vanita sees Diśā's motives, rather than merely testing Aṣṭāvakra's desires, as combining her role as a teacher with her own sexual yearnings. Vanita concludes that Diśā's message is that "desire cannot be entirely overcome" (p. 82). From this, she sees Diśā as advocating the conversion of "feverish agitation into happiness within marriage" (p. 83). Vanita's analysis of this story is both engaging and compelling, yet her claim that this episode is an integral part of Yudhiṣṭhira's education as a king remains undeveloped. I wanted to hear more about how this one story fits into the context of the countless others that Bhīṣma recounts to Yudhiṣṭhira. Moreover, her conclusion at the end of the chapter (ch. 4), that the epics "celebrate single women" (p. 85), seems a bit of a stretch considering the few examples she discusses. In chapter 6, Vanita revisits the encounter between Sulabhā and King Janaka (Mbh 12.308), which she discussed in detail twenty years ago.1 Her return to this episode examines the sexual power dynamics between the female sage and the legendary king from Upaniṣadic lore. As Vanita observes, Sulabhā takes the form of a beautiful young woman to test Janaka's claim to be liberated. Vanita explains:We may expect ordinary people to notice a person's looks first of all, but one who claims to be... (shrink)

The evolution of complex life forms, such as multicellular organisms, is the result of a number of evolutionary transitions in individuality (ETIs). Several attempts have been made to explain their origins, many of which have been internalist (i.e., based largely on internal properties of these life form's ancestors). Here, we show how an externalist perspective, via the ecological scaffolding model in which properties of complex life forms arise from an external scaffold, can shed new light on the question of ETIs. (...) Ultimately, we anticipate progress in the field will occur by recognizing the importance of both the internalist and externalist modes of explanation for ETIs. We illustrate this by considering an extension of the ecological scaffolding model by niche construction in which particles modify the environment which later becomes the scaffold giving rise to collective-level individuality. (shrink)

Autonomous artificial intelligence (AI) systems can lead to unpredictable behavior causing loss or damage to individuals. Intricate questions must be resolved to establish how courts determine liability. Until recently, understanding the inner workings of “black boxes” has been exceedingly difficult; however, the use of Explainable Artificial Intelligence (XAI) would help simplify the complex problems that can occur with autonomous AI systems. In this context, this article seeks to provide technical explanations that can be given by XAI, and to (...) show how suitable explanations for liability can be reached in court. It provides an analysis of whether existing liability frameworks, in both civil and common law tort systems, with the support of XAI, can address legal concerns related to AI. Lastly, it claims their further development and adoption should allow AI liability cases to be decided under current legal and regulatory rules until new liability regimes for AI are enacted. (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)

Recent works in Explainable AI mostly address the transparency issue of black-box models or create explanations for any kind of models (i.e., they are model-agnostic), while leaving explanations of interpretable models largely underexplored. In this paper, we fill this gap by focusing on explanations for a specific interpretable model, namely pattern-based logistic regression (PLR) for binary text classification. We do so because, albeit interpretable, PLR is challenging when it comes to explanations. In particular, we found (...) that a standard way to extract explanations from this model does not consider relations among the features, making the explanations hardly plausible to humans. Hence, we propose AXPLR, a novel explanation method using (forms of) computational argumentation to generate explanations (for outputs computed by PLR) which unearth model agreements and disagreements among the features. Specifically, we use computational argumentation as follows: we see features (patterns) in PLR as arguments in a form of quantified bipolar argumentation frameworks (QBAFs) and extract attacks and supports between arguments based on specificity of the arguments; we understand logistic regression as a gradual semantics for these QBAFs, used to determine the arguments’ dialectic strength; and we study standard properties of gradual semantics for QBAFs in the context of our argumentative re-interpretation of PLR, sanctioning its suitability for explanatory purposes. We then show how to extract intuitive explanations (for outputs computed by PLR) from the constructed QBAFs. Finally, we conduct an empirical evaluation and two experiments in the context of human-AI collaboration to demonstrate the advantages of our resulting AXPLR method. (shrink)

Explainable AI (xAI) methods are important for establishing trust in using black-box models. However, recent criticism has mounted against current xAI methods that they disagree, are necessarily false, and can be manipulated, which has started to undermine the deployment of black-box models. Rudin (2019) goes so far as to say that we should stop using black-box models altogether in high-stakes cases because xAI explanations ‘must be wrong’. However, strict fidelity to the truth is historically not a (...) desideratum in science. Idealizations – the intentional distortions introduced to scientific theories and models – are commonplace in the natural sciences and are seen as a successful scientific tool. Thus, it is not falsehood qua falsehood that is the issue. In this paper, I outline the need for xAI research to engage in idealization evaluation. Drawing on the use of idealizations in the natural sciences and philosophy of science, I introduce a novel framework for evaluating whether xAI methods engage in successful idealizations or deceptive explanations (SIDEs). SIDEs evaluates whether the limitations of xAI methods, and the distortions that they introduce, can be part of a successful idealization or are indeed deceptive distortions as critics suggest. I discuss the role that existing research can play in idealization evaluation and where innovation is necessary. Through a qualitative analysis we find that leading feature importance methods and counterfactual explanations are subject to idealization failure and suggest remedies for ameliorating idealization failure. (shrink)

The artificial intelligence models with machine learning that exhibit the best predictive accuracy, and therefore, the most powerful ones, are, paradoxically, those with the most opaque black-box architectures. At the same time, the unstoppable computerization of advanced industrial societies demands the use of these machines in a growing number of domains. The conjunction of both phenomena gives rise to a control problem on AI that in this paper we analyze by dividing the issue into two. First, we carry out (...) an epistemological examination of the AI’s opacity in light of the latest techniques to remedy it. And second, we evaluate the rationality of delegating tasks in opaque agents. (shrink)

In light of the widespread use of algorithmic (intelligent) systems across numerous domains, there is an increasing awareness about the need to explain their underlying decision-making process and resulting outcomes. Since oftentimes these systems are being considered as black boxes, adding explanations to their outcomes may contribute to the perception of their transparency and, as a result, increase users’ trust and fairness perception towards the system, regardless of its actual fairness, which can be measured using various fairness tests (...) and measurements. Different explanation styles may have a different impact on users’ perception of fairness towards the system and on their understanding of the outcome of the system. Hence, there is a need to understand how various explanation styles may impact non-expert users’ perceptions of fairness and understanding of the system’s outcome. In this study we aimed at fulfilling this need. We performed a between-subject user study in order to examine the effect of various explanation styles on users’ fairness perception and understanding of the outcome. In the experiment we examined four known styles of textual explanations (case-based, demographic-based, input influence-based and sensitivity-based) along with a new style (certification-based) that reflect the results of an auditing process of the system. The results suggest that providing some kind of explanation contributes to users’ understanding of the outcome and that some explanation styles are more beneficial than others. Moreover, while explanations provided by the system are important and can indeed enhance users’ perception of fairness, their perception mainly depends on the outcome of the system. The results may shed light on one of the main problems in explainability of algorithmic systems, which is choosing the best explanation to promote users’ fairness perception towards a particular system, with respect to the outcome of the system. The contribution of this study is reflected in the new and realistic case study that was examined, in the creation and evaluation of a new explanation style that can be used as the link between the actual (computational) fairness of the system and users’ fairness perception and in the need of analyzing and evaluating explanations while taking into account the outcome of the system. (shrink)

Without doctors being able to explain medical decisions to patients, I argue their use of black box AIs would erode the effective and respectful care they provide patients. In addition, I argue that physicians should use AI black boxes only for patients in dire straits, or when physicians use AI as a “co-pilot” (analogous to a spellchecker) but can independently confirm its accuracy. I respond to A.J. London’s objection that physicians already prescribe some drugs without knowing why they (...) work. (shrink)

The last 5 years have seen a series of remarkable achievements in deep-neural-network-based artificial intelligence research, and some modellers have argued that their performance compares favourably to human cognition. Critics, however, have argued that processing in deep neural networks is unlike human cognition for four reasons: they are (i) data-hungry, (ii) brittle, and (iii) inscrutable black boxes that merely (iv) reward-hack rather than learn real solutions to problems. This article rebuts these criticisms by exposing comparative bias within them, in (...) the process extracting some more general lessons that may also be useful for future debates. (shrink)

The use of black box algorithms in medicine has raised scholarly concerns due to their opaqueness and lack of trustworthiness. Concerns about potential bias, accountability and responsibility, patient autonomy and compromised trust transpire with black box algorithms. These worries connect epistemic concerns with normative issues. In this paper, we outline that black box algorithms are less problematic for epistemic reasons than many scholars seem to believe. By outlining that more transparency in algorithms is not always necessary, and (...) by explaining that computational processes are indeed methodologically opaque to humans, we argue that the reliability of algorithms provides reasons for trusting the outcomes of medical artificial intelligence. To this end, we explain how computational reliabilism, which does not require transparency and supports the reliability of algorithms, justifies the belief that results of medical AI are to be trusted. We also argue that several ethical concerns remain with black box algorithms, even when the results are trustworthy. Having justified knowledge from reliable indicators is, therefore, necessary but not sufficient for normatively justifying physicians to act. This means that deliberation about the results of reliable algorithms is required to find out what is a desirable action. Thus understood, we argue that such challenges should not dismiss the use of black box algorithms altogether but should inform the way in which these algorithms are designed and implemented. When physicians are trained to acquire the necessary skills and expertise, and collaborate with medical informatics and data scientists, black box algorithms can contribute to improving medical care. (shrink)

Bricks and boxes -- Between Scylla and Charybdis -- Lessons from the history of science -- Placeholders -- Black-boxing 101 -- History of science 'black-boxing style' -- Diet mechanistic philosophy -- Emergence reframed -- The fuel of scientific progress -- Sailing through the strait.

Gilroy and Wade wrote 20 years ago that every policy presupposes an underlying moral argument that justifies it. This claim is now rarely contested: policy making is an inescapably moral enterprise...

AI is considered to be very abstract to a range of critics. In this regard, algorithms are referred to regularly as black boxes and divorced from human intervention. A particular philosophical maneuver supports this outcome. The aim of this article is to (1) bring the philosophy to the surface that has contributed to this distance between AI and people and (2) offer an alternative philosophical position that can bring this technology closer to individuals and communities. The overall goal of (...) the analysis in this paper is the humanising of AI by addressing the shortcomings of conceptualising algorithms as black boxes. (shrink)

One staple of living in our information society is having access to the web. Web-connected devices interpret our queries and retrieve information from the web in response. Today’s web devices even purport to answer our queries directly without requiring us to comb through search results in order to find the information we want. How do we know whether a web device is trustworthy? One way to know is to learn why the device is trustworthy by inspecting its inner workings, 156–170 (...) 1995; Humphreys 2004, Episteme, 6, 221–229 2009). But ordinary users of web devices cannot inspect their inner workings because of their scale, complexity, and the corporate secrecy which enshrouds both the procedures by which the devices operate and the companies that make them. Further piling on this predicament, authors have criticized web technology on the grounds that the invisibility of the web devices’ inner workings prevents users from critically assessing the procedures that produce a given output, in some cases, barring users from fulfilling their epistemic responsibilities, 343–355 2010; Miller and Record Episteme, 10, 117–134 2013). I consider four broad kinds of reasons which we can acquire without inspecting the inner workings of black-boxed technology: individual understanding, expert testimony, testing through experience, and social vetting; and show how each is a viable method of appraising black-boxed technology. By deploying these methods, we can remain responsible inquirers while nonetheless benefitting from today’s epistemic resources on the web. (shrink)

Once upon a time, there were two large black boxes, A and B, connected by a long insulated copper wire. On box A there were two buttons, marked *a* and *b*, and on box B there were three lights, red, green, and amber. Scientists studying the behavior of the boxes had observed that whenever you pushed the *a* button on box A, the red light flashed briefly on box B, and whenever you pushed the *b* button on box A, (...) the green light flashed briefly. The amber light never seemed to flash. They performed a few billion trials, under a very wide variety of conditions, and found no exceptions. There seemed to them to be a causal regularity, which they conveniently summarized thus. (shrink)

“Black box argument” is a metaphor for modular components of argumentative discussion that are, within a particular discussion, not open to expansion. In public policy debate such as the controversy over abstinence-only sex education, scientific conclusions enter the discourse as black boxes consisting of a result returned from an external and largely impenetrable process. In one way of looking at black box arguments, there is nothing fundamentally new for the argumentation theorist: A black box argument is (...) very like any other appeal to authority, and what might be said about any particular form of black box will turn out to be a particularized version of what might be said about evaluating arguments based on authority. But in another way of looking at black box arguments, they are a constantly evolving technology for coming to conclusions and making these conclusions broadly acceptable. Black boxes are to argumentation what material inventions are to engineering and related sciences. They are anchored in and constrained by fundamental natural processes, but they are also new things that require theoretical explication and practical assessment. (shrink)

The ‘black box’ in Stoic axiology refers to the mysterious connection between the input of Stoic deliberation (reasons generated by the value of indifferents) and the output (appropriate actions). In this paper, I peer into the black box by drawing an analogy between Stoic and Kantian axiology. The value and disvalue of indifferents is intrinsic, but conditional. An extrinsic condition on the value of a token indifferent is that one's selection of that indifferent is sanctioned by context-relative ethical (...) principles. The value of an indifferent that does not meet this condition is normatively silenced, such that it fails to constitute a reason for action. (shrink)

Mental health apps bring unprecedented benefits and risks to individual and public health. A thorough evaluation of these apps involves considering two aspects that are often neglected: the algorithms they deploy and the functions they perform. We focus on mental health apps based on black box algorithms, explore their forms of opacity, discuss the implications derived from their opacity, and propose how to use their outcomes in mental healthcare, self‐care practices, and research. We argue that there is a relevant (...) distinction between functions performed by algorithms in mental health apps, and we focus on the functions of analysis and generation of advice. When performing analytic functions, such as identifying patterns and making predictions concerning people's emotions, thoughts, and behaviors, black box algorithms can be better than other algorithms to provide information to identify early signs of relapse, support diagnostic processes, and improve research by generating outcomes that lead to a better understanding of mental health. However, when carrying out the function of providing mental health advice, black box algorithms have the potential to deliver unforeseen advice that may harm users. We argue that the outcomes of these apps may be trustworthy as a complementary source of information, but express caution about black box algorithms that give advice directly to users. To reap the benefits of mental health apps based on black box algorithms and avoid unintended consequences, we critically need to know whether these algorithms are fulfilling the function of providing mental health advice. (shrink)

In “Artificial Intelligence, Social Media and Depression: A New Concept of Health-Related Digital Autonomy,” Laacke and colleagues posit a revised model of autonomy when using digital algori...

An empirical procedure is suggested for testing a model that postulates variables that intervene between observed causes and abserved effects against a model that includes no such postulate. The procedure is applied to two experiments in psychology. One involves a conditioning regimen that leads to response generalization; the other concerns the question of whether chimpanzees have a theory of mind.

With automation of routine decisions coupled with more intricate and complex information architecture operating this automation, concerns are increasing about the trustworthiness of these systems. These concerns are exacerbated by a class of artificial intelligence that uses deep learning, an algorithmic system of deep neural networks, which on the whole remain opaque or hidden from human comprehension. This situation is commonly referred to as the black box problem in AI. Without understanding how AI reaches its conclusions, it is an (...) open question to what extent we can trust these systems. The question of trust becomes more urgent as we delegate more and more decision-making to and increasingly rely on AI to safeguard significant human goods, such as security, healthcare, and safety. Models that “open the black box” by making the non-linear and complex decision process understandable by human observers are promising solutions to the black box problem in AI but are limited, at least in their current state, in their ability to make these processes less opaque to most observers. A philosophical analysis of trust will show why transparency is a necessary condition for trust and eventually for judging AI to be trustworthy. A more fruitful route for establishing trust in AI is to acknowledge that AI is situated within a socio-technical system that mediates trust, and by increasing the trustworthiness of these systems, we thereby increase trust in AI. (shrink)

Discipline and Punish analyzes the role of collecting, managing, and operationalizing data in disciplinary institutions. Foucault’s discussion is compared to contemporary forms of surveillance and security practices using algorithmic data processing. The article highlights important similarities and differences regarding the way data processing plays a part in subjectivation. This is also compared to Deleuzian accounts and Foucault’s later discussion in Security, Territory, Population. Using these results, the article argues that the prevailing focus on transparency and accountability in the discussion of (...) algorithmic applications needs to be amended with a perspective on different forms of subjectivity and ensuing power relations. (shrink)