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)

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)

As machine learning has gradually entered into ever more sectors of public and private life, there has been a growing demand for algorithmic explainability. How can we make the predictions of complex statistical models more intelligible to end users? A subdiscipline of computer science known as interpretable machine learning (IML) has emerged to address this urgent question. Numerous influential methods have been proposed, from local linear approximations to rule lists and counterfactuals. In this article, I highlight (...) three conceptual challenges that are largely overlooked by authors in this area. I argue that the vast majority of IML algorithms are plagued by (1) ambiguity with respect to their true target; (2) a disregard for error rates and severe testing; and (3) an emphasis on product over process. Each point is developed at length, drawing on relevant debates in epistemology and philosophy of science. Examples and counterexamples from IML are considered, demonstrating how failure to acknowledge these problems can result in counterintuitive and potentially misleading explanations. Without greater care for the conceptual foundations of IML, future work in this area is doomed to repeat the same mistakes. (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)

We argue why interpretability should have primacy alongside empiricism for several reasons: first, if machine learning models are beginning to render some of the high-risk healthcare decisions instead of clinicians, these models pose a novel medicolegal and ethical frontier that is incompletely addressed by current methods of appraising medical interventions like pharmacological therapies; second, a number of judicial precedents underpinning medical liability and negligence are compromised when ‘autonomous’ ML recommendations are considered to be en par with human instruction in (...) specific contexts; third, explainable algorithms may be more amenable to the ascertainment and minimisation of biases, with repercussions for racial equity as well as scientific reproducibility and generalisability. We conclude with some reasons for the ineludible importance of interpretability, such as the establishment of trust, in overcoming perhaps the most difficult challenge ML will face in a high-stakes environment like healthcare: professional and public acceptance. (shrink)

Given the pervasiveness of AI systems and their potential negative effects on people’s lives (especially among already marginalised groups), it becomes imperative to comprehend what goes on when an AI system generates a result, and based on what reasons, it is achieved. There are consistent technical efforts for making systems more “explainable” by reducing their opaqueness and increasing their interpretability and explainability. In this paper, we explore an alternative non-technical approach towards explainability that complement existing ones. Leaving aside technical, statistical, (...) or data-related issues, we focus on the very conceptual underpinnings of the design decisions made by developers and other stakeholders during the lifecycle of a machine learning project. For instance, the design and development of an app to track snoring to detect possible health risks presuppose some picture or another of “health”, which is a key notion that conceptually underpins the project. We take it as a premise that these key concepts are necessarily present during design and development, albeit perhaps tacitly. We argue that by providing “justificatory explanations” about how the team understands the relevant key concepts behind its design decisions, interested parties could gain valuable insights and make better sense of the workings and outcomes of systems. Using the concept of “health”, we illustrate how a particular understanding of it might influence decisions during the design and development stages of a machine learning project, and how making this explicit by incorporating it into ex-post explanations might increase the explanatory and justificatory power of these explanations. We posit that a greater conceptual awareness of the key concepts that underpin design and development decisions may be beneficial to any attempt to develop explainability methods. We recommend that “justificatory explanations” are provided as technical documentation. These are declarative statements that contain at its simplest: (1) a high-level account of the understanding of the relevant key concepts a team possess related to a project’s main domain, (2) how these understandings drive decision-making during the life-cycle stages, and (3) it gives reasons (which could be implicit in the account) that the person or persons doing the explanation consider to have plausible justificatory power for the decisions that were made during the project. (shrink)

Decision-making assisted by algorithms developed by machine learning is increasingly determining our lives. Unfortunately, full opacity about the process is the norm. Would transparency contribute to restoring accountability for such systems as is often maintained? Several objections to full transparency are examined: the loss of privacy when datasets become public, the perverse effects of disclosure of the very algorithms themselves (“gaming the system” in particular), the potential loss of companies’ competitive edge, and the limited gains in answerability to be (...) expected since sophisticated algorithms usually are inherently opaque. It is concluded that, at least presently, full transparency for oversight bodies alone is the only feasible option; extending it to the public at large is normally not advisable. Moreover, it is argued that algorithmic decisions preferably should become more understandable; to that effect, the models of machine learning to be employed should either be interpreted ex post or be interpretable by design ex ante. (shrink)

Advances in intracranial electroencephalography and neurophysiology have enabled the study of previously inaccessible brain regions with high fidelity temporal and spatial resolution. Studies of iEEG have revealed a rich neural code subserving healthy brain function and which fails in disease states. Machine learning, a form of artificial intelligence, is a modern tool that may be able to better decode complex neural signals and enhance interpretation of these data. To date, a number of publications have applied ML to iEEG, but (...) clinician awareness of these techniques and their relevance to neurosurgery, has been limited. The present work presents a review of existing applications of ML techniques in iEEG data, discusses the relative merits and limitations of the various approaches, and examines potential avenues for clinical translation in neurosurgery. One-hundred-seven articles examining artificial intelligence applications to iEEG were identified from 3 databases. Clinical applications of ML from these articles were categorized into 4 domains: i) seizure analysis, ii) motor tasks, iii) cognitive assessment, and iv) sleep staging. The review revealed that supervised algorithms were most commonly used across studies and often leveraged publicly available timeseries datasets. We conclude with recommendations for future work and potential clinical applications. (shrink)

Machine learning algorithms may radically improve our ability to diagnose and treat disease. For moral, legal, and scientific reasons, it is essential that doctors and patients be able to understand and explain the predictions of these models. Scalable, customisable, and ethical solutions can be achieved by working together with relevant stakeholders, including patients, data scientists, and policy makers.

The operations of deep networks are widely acknowledged to be inscrutable. The growing field of Explainable AI has emerged in direct response to this problem. However, owing to the nature of the opacity in question, XAI has been forced to prioritise interpretability at the expense of completeness, and even realism, so that its explanations are frequently interpretable without being underpinned by more comprehensive explanations faithful to the way a network computes its predictions. While this has been taken to be (...) a shortcoming of the field of XAI, I argue that it is broadly the right approach to the problem. (shrink)

In this paper we argue that transparency of machine learning algorithms, just as explanation, can be defined at different levels of abstraction. We criticize recent attempts to identify the explanation of black box algorithms with making their decisions (post-hoc) interpretable, focusing our discussion on counterfactual explanations. These approaches to explanation simplify the real nature of the black boxes and risk misleading the public about the normative features of a model. We propose a new form of (...) algorithmic transparency, that consists in explaining algorithms as an intentional product, that serves a particular goal, or multiple goals (Daniel Dennet’s design stance) in a given domain of applicability, and that provides a measure of the extent to which such a goal is achieved, and evidence about the way that measure has been reached. We call such idea of algorithmic transparency “design publicity.” We argue that design publicity can be more easily linked with the justification of the use and of the design of the algorithm, and of each individual decision following from it. In comparison to post-hoc explanations of individual algorithmic decisions, design publicity meets a different demand (the demand for impersonal justification) of the explainee. Finally, we argue that when models that pursue justifiable goals (which may include fairness as avoidance of bias towards specific groups) to a justifiable degree are used consistently, the resulting decisions are all justified even if some of them are (unavoidably) based on incorrect predictions. For this argument, we rely on John Rawls’s idea of procedural justice applied to algorithms conceived as institutions. (shrink)

Recent work on interpretability in machine learning and AI has focused on the building of simplified models that approximate the true criteria used to make decisions. These models are a useful pedagogical device for teaching trained professionals how to predict what decisions will be made by the complex system, and most importantly how the system might break. However, when considering any such model it’s important to remember Box’s maxim that "All models are wrong but some are useful." We focus (...) on the distinction between these models and explanations in philosophy and sociology. These models can be understood as a "do it yourself kit" for explanations, allowing a practitioner to directly answer "what if questions" or generate contrastive explanations without external assistance. Although a valuable ability, giving these models as explanations appears more difficult than necessary, and other forms of explanation may not have the same trade-offs. We contrast the different schools of thought on what makes an explanation, and suggest that machine learning might benefit from viewing the problem more broadly. (shrink)

According to Clifford Geertz, the purpose of anthropology is not to explain culture but to explicate it. That should cause us to rethink our relationship with machine learning. It is, we contend, perfectly possible that machine learning algorithms, which are unable to explain, and could even be unexplainable themselves, can still be of critical use in a process of explication. Thus, we report on an experiment with anthropological AI. From a dataset of 175K Facebook comments, we trained a (...) neural network to predict the emoji reaction associated with a comment and asked a group of human players to compete against the machine. We show that a) the machine can reach the same accuracy as the players, b) it fails in roughly the same ways as the players, and c) easily predictable emoji reactions tend to reflect unambiguous situations where interpretation is easy. We therefore repurpose the failures of the neural network to point us to deeper and more ambiguous situations where interpretation is hard and explication becomes both necessary and interesting. We use this experiment as a point of departure for discussing how experiences from anthropology, and in particular the tension between formalist ethnoscience and interpretive thick description, might contribute to debates about explainable AI. (shrink)

This study investigates the performance of 21 monolingual and 56 bilingual children aged 5;6-9;0 on German-LITMUS-sentence-repetition (SRT; Hamann et al., 2013) and nonword-repetition-tasks (NWRT; Grimm et al., 2014), which were constructed according to the LITMUS-principles (Language Impairment Testing in Multilingual Settings; Armon-Lotem et al., 2015). Both tasks incorporate complex structures shown to be cross-linguistically challenging for children with Specific Language Impairment (SLI) and aim at minimizing bias against bilingual children while still being indicative of the presence of language impairment across (...) language combinations. Given the great variability in bilingual language exposure and the potential effect of language experience on language performance in bilingual children, we examined whether background variables related to bilingualism, particularly, the degree language dominance as measured by relative amount of use and exposure, could compromise the diagnostic accuracy of the German-LITMUS-SRT and NWRT. We further investigated whether a combination of the two tasks provides better diagnostic accurracy. To address this, we used an unsupervised machine learning algorithm, the Partitioning-Around-Medoids (PAM), for deriving a clinical category for the children as ±language-impaired based on performance scores on SRT and/or NWRT while withholding information about their clinical status based on standardized assessment in L1/L2. Subsequently, we calculated diagnostic accuracy and investigated which background variables (age of onset, length of exposure, degree of language-dominance, socio-economic-status, and risk factors for SLI) best explained clinical-group-membership yielded from the PAM-analysis based on NWRT and/or SRT scores. Results show that although language-dominance influences the performance of bilingual typically-developing children, especially in the SRT, the diagnostic accuracy of the tools is not compromised by language dominance: while risk factors for SLI were significant predictors for clinical group membership in all models, language dominance did not contribute at all to explaining clinical cluster membership as typically-developing or SLI based on any of the combinations of the SRT and NWRT variables. Additionally, results confirm that a combination of SRT scored by correct target structure and the structurally more complex language-dependent part of the NWRT yields better diagnostic accuracy than single measures and is only sensitive for risk factors for SLI and not to dominance levels or SES. (shrink)

Rapid advancements in machine learning techniques allow mass surveillance to be applied on larger scales and utilize more and more personal data. These developments demand reconsideration of the privacy-security dilemma, which describes the tradeoffs between national security interests and individual privacy concerns. By investigating mass surveillance techniques that use bulk data collection and machine learning algorithms, we show why these methods are unlikely to pinpoint terrorists in order to prevent attacks. The diverse characteristics of terrorist attacks—especially when considering (...) lone-wolf terrorism—lead to irregular and isolated footprints. The irregularity of data affects the accuracy of machine learning algorithms and the mass surveillance that depends on them which can be explained by three kinds of known problems encountered in machine learning theory: class imbalance, the curse of dimensionality, and spurious correlations. Proponents of mass surveillance often invoke the distinction between collecting data and metadata, in which the latter is understood as a lesser breach of privacy. Their arguments commonly overlook the ambiguity in the definitions of data and metadata and ignore the ability of machine learning techniques to infer the former from the latter. Given the sparsity of datasets used for machine learning in counterterrorism and the privacy risks attendant with bulk data collection, policymakers and other relevant stakeholders should critically re-evaluate the likelihood of success of the algorithms and the collection of data on which they depend. (shrink)

As scientists start to adopt machine learning (ML) as one research tool, the security of ML and the knowledge generated become a concern. In this paper, I explain how supervised ML can be improved with better data ontology, or the way we make categories and turn information into data. More specifically, we should design data ontology in such a way that is consistent with the knowledge that we have about the target phenomenon so that such ontology can help us (...) make the inductive leap. I do so by thinking through a thought experiment, Goodman’s New Riddle of Induction (Fact, fiction, and forecast, Harvard University Press, 1955). Goodman’s riddle helps flesh out three problems of induction: (1) the problem of equal goodies, that there are often too many equally good inductive results given the same data; (2) the problem of diverging performance, that these equally good results can give opposite predictions in the future; and (3) the problem of mediocrity, that when averaged across all equally possible datasets and tasks, no inductive algorithm outperforms any other. I show that all these three problems are manifested as real obstacles in ML practice, namely, the Rashomon effect (Breiman in Stat Sci 16(3):199–231, 2001), the problem of underspecification (D’Amour et al. in J Mach Learn Res, 2020, https://doi.org/10.48550/arXiv.2011.03395 ), and the No Free Lunch theorem (Wolpert in Neural Comput 8(7):1341–90, 1996, https://doi.org/10.1162/neco.1996.8.7.1341 ). Lastly, I argue that proper data ontology can help mitigate these problems and I demonstrate how using concrete examples from climate science. This research highlights the links between philosophers’ discussions of induction and implications in ML practice. (shrink)

Machine learning-based AI algorithms lack transparency. In this article, I offer an interpretation of AI’s explainability problem and highlight its ethical saliency. I try to make the case for the legal enforcement of a strong explainability requirement: human organizations which decide to automate decision-making should be legally obliged to demonstrate the capacity to explain and justify the algorithmic decisions that have an impact on the wellbeing, rights, and opportunities of those affected by the decisions. This legal duty can (...) be derived from the demands of Rawlsian public reason. In the second part of the paper, I try to show that the argument from the limitations of human cognition fails to get AI off the hook of public reason. Against a growing trend in AI ethics, my main argument is that the analogy between human minds and artificial neural networks fails because it suffers from an atomistic bias which makes it blind to the social and institutional dimension of human reasoning processes. I suggest that developing interpretive AI algorithms is not the only possible answer to the explainability problem; social and institutional answers are also available and in many cases more trustworthy than techno-scientific ones. (shrink)

IntroductionThe assessments of the motor symptoms in Parkinson’s disease are usually limited to clinical rating scales, and it depends on the clinician’s experience. This study aims to propose a machine learning technique algorithm using the variables from upper and lower limbs, to classify people with PD from healthy people, using data from a portable low-cost device. And can be used to support the diagnosis and follow-up of patients in developing countries and remote areas.MethodsWe used Kinect®eMotion system to capture the (...) spatiotemporal gait data from 30 patients with PD and 30 healthy age-matched controls in three walking trials. First, a correlation matrix was made using the variables of upper and lower limbs. After this, we applied a backward feature selection model using R and Python to determine the most relevant variables. Three further analyses were done using variables selected from backward feature selection model, movement disorders specialist, and all the variables from the dataset. We ran seven machine learning models for each model. Dataset was divided 80% for algorithm training and 20% for evaluation. Finally, a causal inference model using the DoWhy library was performed on Dataset B due to its accuracy and simplicity.ResultsThe Random Forest model is the most accurate for all three variable Datasets followed by the support vector machine. The CIM shows a relation between leg variables and the arms swing asymmetry and a proportional relationship between ASA and the diagnosis of PD with a robust estimator.ConclusionsMachine learning techniques based on objective measures using portable low-cost devices are useful and accurate to classify patients with Parkinson’s disease. This method can be used to evaluate patients remotely and help clinicians make decisions regarding follow-up and treatment. (shrink)

Decision-making assisted by algorithms developed by machine learning is increasingly determining our lives. Unfortunately, full opacity about the process is the norm. Would transparency contribute to restoring accountability for such systems as is often maintained? Several objections to full transparency are examined: the loss of privacy when datasets become public, the perverse effects of disclosure of the very algorithms themselves, the potential loss of companies’ competitive edge, and the limited gains in answerability to be expected since sophisticated algorithms usually (...) are inherently opaque. It is concluded that, at least presently, full transparency for oversight bodies alone is the only feasible option; extending it to the public at large is normally not advisable. Moreover, it is argued that algorithmic decisions preferably should become more understandable; to that effect, the models of machine learning to be employed should either be interpreted ex post or be interpretable by design ex ante. (shrink)

Decision-making assisted by algorithms developed by machine learning is increasingly determining our lives. Unfortunately, full opacity about the process is the norm. Would transparency contribute to restoring accountability for such systems as is often maintained? Several objections to full transparency are examined: the loss of privacy when datasets become public, the perverse effects of disclosure of the very algorithms themselves, the potential loss of companies’ competitive edge, and the limited gains in answerability to be expected since sophisticated algorithms usually (...) are inherently opaque. It is concluded that, at least presently, full transparency for oversight bodies alone is the only feasible option; extending it to the public at large is normally not advisable. Moreover, it is argued that algorithmic decisions preferably should become more understandable; to that effect, the models of machine learning to be employed should either be interpreted ex post or be interpretable by design ex ante. (shrink)

In lieu of an abstract, here is a brief excerpt of the content:NARRATIVE AND EXPLANATION: EXPLAINING ANNA KARENINA IN THE LIGHT OF ITS EPIGRAPH Marina Ludwigs University ofCalifornia, Irvine In this paper, I will be examining the relation of explanation to narrative, looking briefly at the theoretical side ofthe problematic and in more detail at specific explanatory issues that arise in Tolstoy's novel Anna Karenina. Although the use itselfofthe term "explanation" is not as visible in the humanities (...) as it is in the sciences, the explanatory enterprise by other names is just as prominent. For indeed, an animating impetus behind textual interpretation is an explanation ofhowatextcan be brought into a correspondence with another text, while contemporary critical-theoretical activities of demystifying aesthetic and cultural assumptions, contextualizing historical knowledge, or laying bare ideological presuppositions are, in one way or another, engagements in the explanation ofthe workings of language. I approach humanistic explanation via a brief detour of scientific explanation, which constitutes an already established discourse with a set of defined issues and standard arguments, limiting myself to touching upon issues relevant to the narrativization of explanation. In scientific debates, explanation is defined as an answer to the question of "why?," in contradistinction to the question of "what?," which prompts a description of a phenomenon in question. Descriptive and explanatory conceptual frameworks, however, are interdependent. A physical description involves an "in-depth" understanding of underlying regularities. An explanatory answer to a question is, in its turn, an argument that logically deduces the explanandum (the description ofa phenomenon to be explained) from the explanans (well-understood regularities). This Marina Ludwigs125 formal definition of an explanation necessitates a symmetry between explanation and prediction, because the reconstructed logical sequence can also be read in the other direction, demonstrating the inevitability of what has occurred. Thus the sight of shattered glass and a baseball lying about allows me to explain the accident by invoking the law ofthe conservation ofmomentum and the low shock-resistance ofglass. But the knowledge of these regularities would have also allowed me to predict the same outcome had I seen the baseball flying in its direction ofthe window. The ontological status of "regularities" has elicited a lively debate in the scientific community pertaining to the problem ofthe "real." The standard view of regularities defines them as general laws (the kind we learn about in physics classes) that could be deductive or inductive, deterministic or statistical. A general law may be subsumed under another, even more general law, thus forming a unifying hierarchy of laws. A rival conception appeals to causal mechanisms and defines regularities as established chains oftemporal succession. Both models, the covering-law and the causal one, have been criticized: the former—for its failure to distinguish a law from an accidental generalization; the latter—for its resurrection ofthe metaphysical notion ofcausal influence and its inability to deal with probabilistic events. One of the later contenders is Nancy Cartwright's model, that proposes to conceive of regularities as so-called "nomological machines"—"fixed (enough) arrangements] ofcomponents, or factors, with stable (enough) capacities that in the right sort of stable (enough) environment will, with repeated operation, give rise to the kind of regular behavior that we represent in our scientific laws" (Cartwright 50). Cartwright claims that scientific models work on the ceteris paribus (everything else being equal) basis. That is to say they "hold only in circumscribed conditions or so long as no factors relevant to the effect besides those specified occur" (Cartwright 28), such as inside a battery, refrigerator, or a rocket—insofar as special shielding conditions are in effect (as a counterexample, imagine usingNewton's laws ofmechanics for describing the motion of a dropped crumpled dollar). Cartwright's explanatory framework is thus pragmatic and local, avoiding the twin pitfalls of the metaphysical assumptions underlying the two older models—the totalizing and unifying idea of a covering law, on the one hand, and the influence of causal "force," on the other. What is instead emphasized by her is a procedural aspect of scientific knowledge that has been historically prevalent and reflects the practical exigencies of human existence—the experience of dangers and natural obstacles to human 126Explaining 'Anna Karenina ' desires. When ill-defined impediments are analyzed and reduced to concrete difficulties, specific... (shrink)

The use of counterfactuals for considerations of algorithmic fairness and explainability is gaining prominence within the machine learning community and industry. This paper argues for more caution with the use of counterfactuals when the facts to be considered are social categories such as race or gender. We review a broad body of papers from philosophy and social sciences on social ontology and the semantics of counterfactuals, and we conclude that the counterfactual approach in machine learning fairness and (...) social explainability can require an incoherent theory of what social categories are. Our findings suggest that most often the social categories may not admit counterfactual manipulation, and hence may not appropriately satisfy the demands for evaluating the truth or falsity of counterfactuals. This is important because the wide- spread use of counterfactuals in machine learning can lead to misleading results when applied in high-stakes domains. Accordingly, we argue that even though counterfactuals play an essential part in some causal inferences, their use for questions of algorithmic fairness and social explanations can create more problems than they resolve. Our positive result is a set of tenets about using counterfactuals for fairness and explanations in machine learning. (shrink)

Increasing student involvement in classes has always been a challenge for teachers and school managers. In online learning, some interactivity mechanisms like quizzes are increasingly used to engage students during classes and tasks. However, there is a high demand for tools that evaluate the efficiency of these mechanisms. In order to distinguish between high and low levels of engagement in tasks, it is possible to monitor brain activity through functional near-infrared spectroscopy. The main advantages of this technique are portability, low (...) cost, and a comfortable way for students to concentrate and perform their tasks. This setup provides more natural conditions for the experiments if compared to the other acquisition tools. In this study, we investigated levels of task involvement through the identification of correct and wrong answers of typical quizzes used in virtual environments. We collected data from the prefrontal cortex region of 18 students while watching a video lecture. This data was modeled with supervised learning algorithms. We used random forests and penalized logistic regression to classify correct answers as a function of oxyhemoglobin and deoxyhemoglobin concentration. These models identify which regions best predict student performance. The random forest and penalized logistic regression obtained, respectively, 0.67 and 0.65 area of the ROC curve. Both models indicate that channels F4-F6 and AF3-AFz are the most relevant for the prediction. The statistical significance of these models was confirmed through cross-validation and a permutation test. This methodology can be useful to better understand the teaching and learning processes in a video lecture and also provide improvements in the methodologies used in order to better adapt the presentation content. (shrink)

Machine behavior that is based on learning algorithms can be significantly influenced by the exposure to data of different qualities. Up to now, those qualities are solely measured in technical terms, but not in ethical ones, despite the significant role of training and annotation data in supervised machine learning. This is the first study to fill this gap by describing new dimensions of data quality for supervised machine learning applications. Based on the rationale that different social and (...) psychological backgrounds of individuals correlate in practice with different modes of human–computer-interaction, the paper describes from an ethical perspective how varying qualities of behavioral data that individuals leave behind while using digital technologies have socially relevant ramification for the development of machine learning applications. The specific objective of this study is to describe how training data can be selected according to ethical assessments of the behavior it originates from, establishing an innovative filter regime to transition from the big data rationale n = all to a more selective way of processing data for training sets in machine learning. The overarching aim of this research is to promote methods for achieving beneficial machine learning applications that could be widely useful for industry as well as academia. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory (...) strategies will lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

Background and ObjectivesFocal cortical dysplasia is a type of malformations of cortical development and one of the leading causes of drug-resistant epilepsy. Postoperative results improve the diagnosis of lesions on structural MRIs. Advances in quantitative algorithms have increased the identification of FCD lesions. However, due to significant differences in size, shape, and location of the lesion in different patients and a big deal of time for the objective diagnosis of lesion as well as the dependence of individual interpretation, sensitive approaches (...) are required to address the challenge of lesion diagnosis. In this research, a FCD computer-aided diagnostic system to improve existing methods is presented.MethodsMagnetic resonance imaging data were collected from 58 participants. Morphological and intensity-based features were calculated for each cortical surface and inserted into an artificial neural network. Statistical examinations evaluated classifier efficiency.ResultsNeural network evaluation metrics—sensitivity, specificity, and accuracy—were 96.7, 100, and 98.6%, respectively. Furthermore, the accuracy of the classifier for the detection of the lobe and hemisphere of the brain, where the FCD lesion is located, was 84.2 and 77.3%, respectively.ConclusionAnalyzing surface-based features by automated machine learning can give a quantitative and objective diagnosis of FCD lesions in presurgical assessment and improve postsurgical outcomes. (shrink)

Alcohol use disorder is a burden to society creating social and health problems. Detection of AUD and its effects on the brain are difficult to assess. This problem is enhanced by the comorbid use of other substances such as nicotine that has been present in previous studies. Recent machine learning algorithms have raised the attention of researchers as a useful tool in studying and detecting AUD. This work uses AUD and controls samples free of any other substance use to (...) assess the performance of a set of commonly used machine learning classifiers detecting AUD from resting state functional network connectivity derived from independent component analysis. The cohort used included 51 alcohol dependent subjects and 51 control subjects. Despite alcohol, none of the 102 subjects reported use of nicotine, cannabis or any other dependence or habit formation substance. Classification features consisted of whole brain rsFNC estimates undergoing a feature selection process using a random forest approach. Features were then fed to 10 different machine learning classifiers to be evaluated based on their classification performance. A neural network classifier showed the highest performance with an area under the curve of 0.79. Other good performers with similar AUC scores were logistic regression, nearest neighbor, and support vector machine classifiers. The worst results were obtained with Gaussian process and quadratic discriminant analysis. The feature selection outcome pointed to functional connections between visual, sensorimotor, executive control, reward, and salience networks as the most relevant for classification. We conclude that AUD can be identified using machine learning classifiers in the absence of nicotine comorbidity. (shrink)

The workshop “Machine Learning: Prediction Without Explanation?” brought together philosophers of science and scholars from various fields who study and employ Machine Learning (ML) techniques, in order to discuss the changing face of science in the light of ML's constantly growing use. One major focus of the workshop was on the impact of ML on the concept and value of scientific explanation. One may speculate whether ML’s increased use in science exemplifies a paradigmatic turn towards mere (...) pattern recognition and prediction and away from science’s traditional aim of explanation. In contrast, certain conceptions of explanation, such as statistical explanation, could turn out to fit well with the achievements of present-day ML and concede an explanatory value to these achievements after all. It is an open question how to explain ML successes themselves, and this question was in the focus of several talks in the workshop. Based on the topics raised, we will discuss the talks’ contents in more detail as organized into (i) practitioners’ perspectives, (ii) explanations from ML, (iii) explanations of ML, (iv) societal implications, and (v) global and historical perspectives. (shrink)

The food security of more than half of the world’s population depends on rice production which is one of the key objectives of precision agriculture. The traditional rice almanac used astronomical and climate factors to estimate yield response. However, this research integrated meteorological, agro-chemical, and soil physiographic factors for yield response prediction. Besides, the impact of those factors on the production of three major rice ecotypes has also been studied in this research. Moreover, this study found a different set of (...) those factors with respect to the yield response of different rice ecotypes. Machine learning algorithms named Extreme Gradient Boosting and Support Vector Regression have been used for predicting the yield response. The SVR shows better results than XGBoost for predicting the yield of the Aus rice ecotype, whereas XGBoost performs better for forecasting the yield of the Aman and Boro rice ecotypes. The result shows that the root mean squared error of three different ecotypes are in between 9.38% and 24.37% and that of R-squared values are between 89.74% and 99.13% on two different machine learning algorithms. Moreover, the explainability of the models is also shown in this study with the help of the explainable artificial intelligence model called Local Interpretable Model-Agnostic Explanations. (shrink)

The paper presents a semiotic interpretation of the phenomenological debate on the notion of person, focusing in particular on Edmund Husserl, Max Scheler, and Edith Stein. The semiotic interpretation lets us identify the categories that orient the debate: collective/individual and subject/object. As we will see, the phenomenological analysis of the relation between person and social units such as the community, the association, and the mass shows similarities to contemporary socio-semiotic models. The difference between community, association, and mass provides an (...) class='Hi'>explanation for the establishment of legal systems. The notion of person we inherit from phenomenology can also be useful in facing juridical problems raised by the use of non-human decision-makers such as machine learning algorithms and artificial intelligence applications. (shrink)

GoalThis paper presents an immersive Virtual Reality system to analyze and train Executive Functions of soccer players. EFs are important cognitive functions for athletes. They are a relevant quality that distinguishes amateurs from professionals.MethodThe system is based on immersive technology, hence, the user interacts naturally and experiences a training session in a virtual world. The proposed system has a modular design supporting the extension of various so-called game modes. Game modes combine selected game mechanics with specific simulation content to target (...) particular training aspects. The system architecture decouples selection/parameterization and analysis of training sessions via a coaching app from an Unity3D-based VR simulation core. Monitoring of user performance and progress is recorded by a database that sends the necessary feedback to the coaching app for analysis.ResultsThe system is tested for VR-critical performance criteria to reveal the usefulness of a new interaction paradigm in the cognitive training and analysis of EFs. Subjective ratings for overall usability show that the design as VR application enhances the user experience compared to a traditional desktop app; whereas the new, unfamiliar interaction paradigm does not negatively impact the effort for using the application.ConclusionThe system can provide immersive training of EF in a fully virtual environment, eliminating potential distraction. It further provides an easy-to-use analyzes tool to compare user but also an automatic, adaptive training mode. (shrink)

This article examines what the rise in machine learning systems might mean for social theory. Focusing on financial markets, in which algorithmic securities trading founded on ML-based decision-making is gaining traction, I discuss the extent to which established sociological notions remain relevant or demand a reconsideration when applied to an ML context. I argue that ML systems have some capacity for agency and for engaging in forms of collective machine behaviour, in which ML systems interact with other (...) machines. However, ML-based collective machine behaviour is irreducible to human decision-making and thereby challenges established sociological notions of financial markets. I argue that such behaviour can nonetheless be analysed through an adaptation of sociological theories of interaction and collective behaviour. (shrink)

Objective: To examine the role of explainability in machine learning for healthcare (MLHC), and its necessity and significance with respect to effective and ethical MLHC application. Study Design and Setting: This commentary engages with the growing and dynamic corpus of literature on the use of MLHC and artificial intelligence (AI) in medicine, which provide the context for a focused narrative review of arguments presented in favour of and opposition to explainability in MLHC. Results: We find that concerns regarding explainability (...) are not limited to MLHC, but rather extend to numerous well-validated treatment interventions as well as to human clinical judgment itself. We examine the role of evidence-based medicine in evaluating unexplainable treatments and technologies, and highlight the analogy between the concept of explainability in MLHC and the related concept of mechanistic reasoning in evidence-based medicine. Conclusion: Ultimately, we conclude that the value of explainability in MLHC is not intrinsic, but is instead instrumental to achieving greater imperatives such as performance and trust. We caution against the uncompromising pursuit of explainability, and advocate instead for the development of robust empirical methods to successfully evaluate increasingly inexplicable algorithmic systems. (shrink)

In a recent article, Erasmus, Brunet, and Fisher (2021) argue that Artificial Neural Networks (ANNs) are explainable. They survey four influential accounts of explanation: the Deductive-Nomological model, the Inductive-Statistical model, the Causal-Mechanical model, and the New-Mechanist model. They argue that, on each of these accounts, the features that make something an explanation is invariant with regard to the complexity of the explanans and the explanandum. Therefore, they conclude, the complexity of ANNs (and other Machine Learning models) does (...) not make them less explainable. In this reply, it is argued that Erasmus et al. left out one influential account of explanation from their discussion: the Unificationist model. It is argued that, on the Unificationist model, the features that makes something an explanation is sensitive to complexity. Therefore, on the Unificationist model, ANNs (and other Machine Learning models) are not explainable. It is emphasized that Erasmus et al.’s general strategy is correct. The literature on explainable Artificial Intelligence can benefit by drawing from philosophical accounts of explanation. However, philosophical accounts of explanation do not settle the problem of whether ANNs are explainable because they do not unanimously declare that explanation is invariant with regard to complexity. (shrink)

Machine learning (ML) models recently led to major breakthroughs in predictive tasks in the natural sciences. Yet their benefits for the social sciences are less evident, as even high-profile studies on the prediction of life trajectories have shown to be largely unsuccessful – at least when measured in traditional criteria of scientific success. This paper tries to shed light on this remarkable performance gap. Comparing two social science case studies to a paradigm example from the natural sciences, we argue (...) that, in addition to explanation, prediction is an important goal of social science – and we identify constraints that impede pure ML prediction from being successful in that field. As a remedy, we outline elements of an integrative modelling approach that combines explanatory models and predictive ML models. (shrink)

Machine learning is used more and more in scientific contexts, from the recent breakthroughs with AlphaFold2 in protein fold prediction to the use of ML in parametrization for large climate/astronomy models. Yet it is unclear whether we can obtain scientific explanations from such models. I argue that when machine learning is used to conduct causal inference we can give a new positive answer to this question. However, these ML models are purpose-built models and there are technical results showing (...) that standard machine learning models cannot be used for the same type of causal inference. Instead, there is a pathway to causal explanations from predictive ML models through new explainability techniques; specifically, new methods to extract structural equation models from such ML models. The extracted models are likely to suffer from issues though: they will often fail to account for confounders and colliders, as well as deliver simply incorrect causal graphs due to ML models tendency to violate physical laws such as the conservation of energy. In this case, extracted graphs are a starting point for new explanations, but predictive accuracy is no guarantee for good explanations. (shrink)

Modern explainable AI methods remain far from providing human-like answers to ‘why’ questions, let alone those that satisfactorily agree with human-level understanding. Instead, the results that such methods provide boil down to sets of causal attributions. Currently, the choice of accepted attributions rests largely, if not solely, on the explainee’s understanding of the quality of explanations. The paper argues that such decisions may be transferred from a human to an XAI agent, provided that its machine-learning algorithms perform genuinely abductive (...) inferences. The paper outlines the key predicament in the current inductive paradigm of ML and the associated XAI techniques, and sketches the desiderata for a truly participatory, second-generation XAI, which is endowed with abduction. (shrink)

Businesses increasingly rely on algorithms that are data-trained sets of decision rules (i.e., the output of the processes often called “machine learning”) and implement decisions with little or no human intermediation. In this article, we provide a philosophical foundation for the claim that algorithmic decision-making gives rise to a “right to explanation.” It is often said that, in the digital era, informed consent is dead. This negative view originates from a rigid understanding that presumes informed consent is (...) a static and complete transaction. Such a view is insufficient, especially when data are used in a secondary, noncontextual, and unpredictable manner—which is the inescapable nature of advanced artificial intelligence systems. We submit that an alternative view of informed consent—as an assurance of trust for incomplete transactions—allows for an understanding of why the rationale of informed consent already entails a right to ex post explanation. (shrink)

Floods are one of the most common natural disasters in the world that affect all aspects of life, including human beings, agriculture, industry, and education. Research for developing models of flood predictions has been ongoing for the past few years. These models are proposed and built-in proportion for risk reduction, policy proposition, loss of human lives, and property damages associated with floods. However, flood status prediction is a complex process and demands extensive analyses on the factors leading to the occurrence (...) of flooding. Consequently, this research proposes an Internet of Things-based flood status prediction (IoT-FSP) model that is used to facilitate the prediction of the rivers flood situation. The IoT-FSP model applies the Internet of Things architecture to facilitate the flood data acquisition process and three machine learning (ML) algorithms, which are Decision Tree (DT), Decision Jungle, and Random Forest, for the flood prediction process. The IoT-FSP model is implemented in MATLAB and Simulink as development platforms. The results show that the IoT-FSP model successfully performs the data acquisition and prediction tasks and achieves an average accuracy of 85.72% for the three-fold cross-validation results. The research finding shows that the DT scores the highest accuracy of 93.22%, precision of 92.85, and recall of 92.81 among the three ML algorithms. The ability of the ML algorithm to handle multivariate outputs of 13 different flood textual statuses provides the means of manifesting explainable artificial intelligence and enables the IoT-FSP model to act as an early warning and flood monitoring system. (shrink)

In this paper I argue that the search for explainable models and interpretable decisions in AI must be reformulated in terms of the broader project of offering a pragmatic and naturalistic account of understanding in AI. Intuitively, the purpose of providing an explanation of a model or a decision is to make it understandable to its stakeholders. But without a previous grasp of what it means to say that an agent understands a model or a decision, the explanatory (...) strategies will lack a well-defined goal. Aside from providing a clearer objective for XAI, focusing on understanding also allows us to relax the factivity condition on explanation, which is impossible to fulfill in many machine learning models, and to focus instead on the pragmatic conditions that determine the best fit between a model and the methods and devices deployed to understand it. After an examination of the different types of understanding discussed in the philosophical and psychological literature, I conclude that interpretative or approximation models not only provide the best way to achieve the objectual understanding of a machine learning model, but are also a necessary condition to achieve post hoc interpretability. This conclusion is partly based on the shortcomings of the purely functionalist approach to post hoc interpretability that seems to be predominant in most recent literature. (shrink)

This paper analyzes the rapid and unexpected rise of deep learning within Artificial Intelligence and its applications. It tackles the possible reasons for this remarkable success, providing candidate paths towards a satisfactory explanation of why it works so well, at least in some domains. A historical account is given for the ups and downs, which have characterized neural networks research and its evolution from “shallow” to “deep” learning architectures. A precise account of “success” is given, in order to sieve (...) out aspects pertaining to marketing or sociology of research, and the remaining aspects seem to certify a genuine value of deep learning, calling for explanation. The alleged two main propelling factors for deep learning, namely computing hardware performance and neuroscience findings, are scrutinized, and evaluated as relevant but insufficient for a comprehensive explanation. We review various attempts that have been made to provide mathematical foundations able to justify the efficiency of deep learning, and we deem this is the most promising road to follow, even if the current achievements are too scattered and relevant for very limited classes of deep neural models. The authors’ take is that most of what can explain the very nature of why deep learning works at all and even very well across so many domains of application is still to be understood and further research, which addresses the theoretical foundation of artificial learning, is still very much needed. (shrink)

Machine learning models are becoming increasingly prevalent in algorithmic trading and investment management. The spread of machine learning in finance challenges existing practices of modelling and model use and creates a demand for practical solutions for how to manage the complexity pertaining to these techniques. Drawing on interviews with quants applying machine learning techniques to financial problems, the article examines how these people manage model complexity in the process of devising machine learning-powered trading algorithms. The (...) analysis shows that machine learning quants use Ockham’s razor – things should not be multiplied without necessity – as a heuristic tool to prevent excess model complexity and secure a certain level of human control and interpretability in the modelling process. I argue that understanding the way quants handle the complexity of learning models is a key to grasping the transformation of the human’s role in contemporary data and model-driven finance. The study contributes to social studies of finance research on the human–model interplay by exploring it in the context of machine learning model use. (shrink)

The usefulness of machine learning algorithms has led to their widespread adoption prior to the development of a conceptual framework for making sense of them. One common response to this situation is to say that machine learning suffers from a “black box problem.” That is, machine learning algorithms are “opaque” to human users, failing to be “interpretable” or “explicable” in terms that would render categorization procedures “understandable.” The purpose of this paper is to challenge the widespread (...) agreement about the existence and importance of a black box problem. The first section argues that “interpretability” and cognates lack precise meanings when applied to algorithms. This makes the concepts difficult to use when trying to solve the problems that have motivated the call for interpretability. Furthermore, since there is no adequate account of the concepts themselves, it is not possible to assess whether particular technical features supply formal definitions of those concepts. The second section argues that there are ways of being a responsible user of these algorithms that do not require interpretability. In many cases in which a black box problem is cited, interpretability is a means to a further end such as justification or non-discrimination. Since addressing these problems need not involve something that looks like an “interpretation” of an algorithm, the focus on interpretability artificially constrains the solution space by characterizing one possible solution as the problem itself. Where possible, discussion should be reformulated in terms of the ends of interpretability. (shrink)

Decisions based on algorithmic, machine learning models can be unfair, reproducing biases in historical data used to train them. While computational techniques are emerging to address aspects of these concerns through communities such as discrimination-aware data mining and fairness, accountability and transparency machine learning, their practical implementation faces real-world challenges. For legal, institutional or commercial reasons, organisations might not hold the data on sensitive attributes such as gender, ethnicity, sexuality or disability needed to diagnose and mitigate emergent (...) indirect discrimination-by-proxy, such as redlining. Such organisations might also lack the knowledge and capacity to identify and manage fairness issues that are emergent properties of complex sociotechnical systems. This paper presents and discusses three potential approaches to deal with such knowledge and information deficits in the context of fairer machine learning. Trusted third parties could selectively store data necessary for performing discrimination discovery and incorporating fairness constraints into model-building in a privacy-preserving manner. Collaborative online platforms would allow diverse organisations to record, share and access contextual and experiential knowledge to promote fairness in machine learning systems. Finally, unsupervised learning and pedagogically interpretable algorithms might allow fairness hypotheses to be built for further selective testing and exploration. Real-world fairness challenges in machine learning are not abstract, constrained optimisation problems, but are institutionally and contextually grounded. Computational fairness tools are useful, but must be researched and developed in and with the messy contexts that will shape their deployment, rather than just for imagined situations. Not doing so risks real, near-term algorithmic harm. (shrink)

In this paper we analyse the conditions for attributing to AI autonomous systems the ontological status of “artificial moral agents”, in the context of the “distributed responsibility” between humans and machines in Machine Ethics (ME). In order to address the fundamental issue in ME of the unavoidable “opacity” of their decisions with ethical/legal relevance, we start from the neuroethical evidence in cognitive science. In humans, the “transparency” and then the “ethical accountability” of their actions as responsible moral agents (...) is not in contradiction with the unavoidable “opacity” (unawareness) of the brain process by which they perform their moral judgements on the right action to execute. In fact, the moral accountability of our actions depends on what is immediately before and after our “moral judgements” on the right action to execute (formally, deontic first order logic (FOL) decisions). I.e., our moral accountability depends on the “ethical constraints” we imposed to our judgement before performing it in an opaque way. Anyway, our moral accountability depends overall on the “ethical assessment” or explicit “moral reasoning” after and over the moral judgement before executing our actions (deontic higher order logic (HOL) assessment). In this way, in the light of the AI “imitation game”, the consistent attribution of the status of ethically accountable artificial moral agents to autonomous AI systems depends on two similar conditions. Firstly, it depends on the presence of “ethical constraints” to be satisfied in their Machine Learning (ML) supervised optimization algorithm during its training phase, to give the system ethical skills (“competences”) in its decisions. Secondly – and definitely—, it depends on the presence in an AI autonomous system of a deontic HOL “ethical reasoner” to perform an automatic, and fully transparent assessment (metalogical deontic valuation) about the decisions taken by the ethically skilled ML algorithm about the right action to execute, before executing it. Finally, we show that the proper deontic FOL and HOL for this class of artificial moral agents is Kripke’s modal relational logic, in its algebraic topological formalization. This is naturally implemented in the dissipative QFT unsupervised deep learning of our brains, based on the “doubling of the degrees of freedom” (DDF), and then in the so-called “deep-belief” artificial neural networks for the statistical data pre-processing. This unsupervised learning procedure is also compliant with the usage of the “maximin fairness principle”, used as a balancing aggregation principle of the statistical variables in Sen’s formal theory of fairness. (shrink)

Machine learning has become a key component of contemporary information systems. Unlike prior information systems explicitly programmed in formal languages, ML systems infer rules from data. This paper shows what this difference means for the critical analysis of socio-technical systems based on machine learning. To provide a foundation for future critical analysis of machine learning-based systems, we engage with how the term is framed and constructed in self-education resources. For this, we analyze machine learning tutorials, an (...) important information source for self-learners and a key tool for the formation of the practices of the machine learning community. Our analysis identifies canonical examples of machine learning as well as important misconceptions and problematic framings. Our results show that machine learning is presented as being universally applicable and that the application of machine learning without special expertise is actively encouraged. Explanations of machine learning algorithms are missing or strongly limited. Meanwhile, the importance of data is vastly understated. This has implications for the manifestation of social inequalities through machine learning-based systems. (shrink)

Deep learning and other black-box models are becoming more and more popular today. Despite their high performance, they may not be accepted ethically or legally because of their lack of explainability. This paper presents the increasing number of legal requirements on machine learning model interpretability and explainability in the context of private and public decision making. It then explains how those legal requirements can be implemented into machine-learning models and concludes with a call for more inter-disciplinary research on (...) explainability. (shrink)

Artificial Intelligence (AI) pervades humanity in 2022, and it is notoriously difficult to understand how certain aspects of it work. There is a movement—_Explainable_ Artificial Intelligence (XAI)—to develop new methods for explaining the behaviours of AI systems. We aim to highlight one important philosophical significance of XAI—it has a role to play in the elimination of vagueness. To show this, consider that the use of AI in what has been labeled _surveillance capitalism_ has resulted in humans quickly gaining the capability (...) to identify and classify most of the occasions in which languages are used. We show that the knowability of this information is incompatible with what a certain theory of vagueness—_epistemicism_—says about vagueness. We argue that one way the epistemicist could respond to this threat is to claim that this process brought about the end of vagueness. However, we suggest an alternative interpretation, namely that epistemicism is false, but there is a weaker doctrine we dub _technological epistemicism_, which is the view that vagueness is due to ignorance of linguistic usage, but the ignorance can be overcome. The idea is that knowing more of the relevant data and how to process it enables us to know the semantic values of our words and sentences with higher confidence and precision. Finally, we argue that humans are probably not going to believe what future AI algorithms tell us about the sharp boundaries of our vague words unless the AI involved can be explained in terms understandable by humans. That is, if people are going to accept that AI can tell them about the sharp boundaries of the meanings of their words, then it is going to have to be XAI. (shrink)

This book provides a novel framework for understanding and revising labor markets and education policies in an era of machine learning. It posits that while learning and knowing both require thinking, learning is fundamentally different than knowing because it results in cognitive processes that change over time. Learning, in contrast to knowing, requires time and agency. Therefore, “learning algorithms”—that enable machines to modify their actions based on real-world experiences—are a fundamentally new form of artificial intelligence that have potential to (...) be even more disruptive to labor markets than prior introductions of digital technology. To explore the difference between knowing and learning, Turing’s “Imitation Game,”—that he proposed as a test for machine thinking—is expanded to include time dependence. The arguments presented in the book introduce three novel concepts: (1) Comparative learning advantage: This is a concept analogous to comparative labor advantage but arises from the disparate times required to learn new knowledge bases/skillsets. It is argued that in the future, comparative learning advantages between humans and machines will determine their division of labor. (2) Two dimensions of job performance—expertise and interpersonal: Job tasks can be sorted into two broad categories. Tasks that require expertise have stable endpoints, which makes these tasks inherently repetitive and subject to automation. Tasks that are interpersonal are highly context-dependent and lack stable endpoints, which makes these tasks inherently non-routine. Humans compared to machines have a comparative learning advantage along the interpersonal dimension, which is increasing in value economically. (3) The Learning Game is a time-dependent version of Turing’s “Imitation Game.” It is more than a thought experiment. The “Learning Game” provides a mathematical framework with quantitative criteria for training and assessing comparative learning advantages. The book is highly interdisciplinary—presenting philosophical arguments in economics, artificial intelligence, and education. It also provides data, mathematical analysis, and testable criteria that researchers in these fields will find of practical use. The book calls for a rethinking of how labor markets operate and how the education system should prepare students for future jobs. It concludes with a list of counterintuitive recommendations for future education and labor policies that all stakeholders—employers, employees, educators, students, and political leaders—should heed. (shrink)