We compute that the index set of PAC-learnable concept classes is m-complete Σ30\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Sigma^{0}_{3}}$$\end{document} within the set of indices for all concept classes of a reasonable form. All concept classes considered are computable enumerations of computable Π10\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\Pi^{0}_{1}}$$\end{document} classes, in a sense made precise here. This family of concept classes is sufficient to cover all standard examples, and also has the property that PAC learnability (...) is equivalent to finite VC dimension. (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)

Machine learning algorithms are increasingly used to support decision-making in the exercise of public authority. Here, we argue that an important consideration has been overlooked in previous discussions: whether the use of ML undermines the democratic legitimacy of public institutions. From the perspective of democratic legitimacy, it is not enough that ML contributes to efficiency and accuracy in the exercise of public authority, which has so far been the focus in the scholarly literature engaging with these developments. According to one (...) influential theory, exercises of administrative and judicial authority are democratically legitimate if and only if administrative and judicial decisions serve the ends of the democratic law maker, are based on reasons that align with these ends and are accessible to the public. These requirements are not satisfied by decisions determined through ML since such decisions are determined by statistical operations that are opaque in several respects. However, not all ML-based decision support systems pose the same risk, and we argue that a considered judgment on the democratic legitimacy of ML in exercises of public authority need take the complexity of the issue into account. This paper outlines considerations that help guide the assessment of whether a ML undermines democratic legitimacy when used to support public decisions. We argue that two main considerations are pertinent to such normative assessment. The first is the extent to which ML is practiced as intended and the extent to which it replaces decisions that were previously accessible and based on reasons. The second is that uses of ML in exercises of public authority should be embedded in an institutional infrastructure that secures reason giving and accessibility. (shrink)

Legal structures may form barriers to, or enablers of, adoption of precision agriculture management with small autonomous agricultural robots. This article develops a conceptual regulatory framework for small autonomous agricultural robots, from a practical, self-contained engineering guide perspective, sufficient to get working research and commercial agricultural roboticists quickly and easily up and running within the law. The article examines the liability framework, or rather lack of it, for agricultural robotics in EU, and their transpositions to UK law, as a case (...) study illustrating general international legal concepts and issues. It examines how the law may provide mitigating effects on the liability regime, and how contracts can be developed between agents within it to enable smooth operation. It covers other legal aspects of operation such as the use of shared communications resources and privacy in the reuse of robot-collected data. Where there are some grey areas in current law, it argues that new proposals could be developed to reform these to promote further innovation and investment in agricultural robots. (shrink)

The distinction between distributive and non-distributive profiles figures prominently in current evaluations of the ethical and epistemological risks that are associated with automated profiling practices. The diagnosis that non-distributive profiles may coincidentally situate an individual in the wrong category is often perceived as the central shortcoming of such profiles. According to this diagnosis, most risks can be retraced to the use of non-universal generalisations and various other statistical associations. This article develops a top-down analysis of non-distributive profiles in which this (...) fallibility of non-distributive profiles is no longer central. Instead, it focuses on how profiling creates various asymmetries between an individual data-subject and a profiler. The emergence of informational, interest, and perspectival asymmetries between data-subject and profiler explains how non-distributive profiles weaken the epistemic position of a profiled individual. This alternative analysis provides a more balanced assessment of the epistemic risks associated with non-distributive profiles. (shrink)

Various semantics for studying the square of opposition have been proposed recently. So far, only [14] studied a probabilistic version of the square where the sentences were interpreted by (negated) defaults. We extend this work by interpreting sentences by imprecise (set-valued) probability assessments on a sequence of conditional events. We introduce the acceptability of a sentence within coherence-based probability theory. We analyze the relations of the square in terms of acceptability and show how to construct probabilistic versions of the square (...) of opposition by forming suitable tripartitions. Finally, as an application, we present a new square involving generalized quantifiers. (shrink)

Whether collective agency is a coherent concept depends on the theory of agency that we choose to adopt. We argue that the enactive theory of agency developed by Barandiaran, Di Paolo and Rohde (2009) provides a principled way of grounding agency in biological organisms. However the importance of biological embodiment for the enactive approach might lead one to be skeptical as to whether artificial systems or collectives of individuals could instantiate genuine agency. To explore this issue we contrast the concept (...) of collective agency with multi-agent systems and multi-system agents, and argue that genuinely collective agents instantiate agency at both the collective level and at the level of the component parts. Developing the enactive model, we propose understanding agency – both at the level of the individual and of the collective – as spectra that are constituted by dimensions that vary across time. Finally, we consider whether collectives that are not merely metaphorically ‘agents’ but rather are genuinely agentive also instantiate subjectivity at the collective level. We propose that investigations using the perceptual crossing paradigm suggest that a shared lived perspective can indeed emerge but this should not be conflated with a collective first-person perspective, for which material integration in a living body may be required. (shrink)

In a recently published paper, we argued that theories of cultural evolution can gain explanatory power by being more pluralistic. In his reply to it, Dennett agreed that more pluralism is needed. Our paper’s main point was to urge cultural evolutionists to get their hands dirty by describing the fine details of cultural products and by striving to offer detailed and, when explanatory, varied algorithms or mechanisms to account for them. While Dennett’s latest work on cultural evolution does marvelously well (...) on the first point, it has only whet our appetite on the second. Accordingly, the present paper aims to show what an evolutionary explanation of culture that takes the variety of cultural evolution mechanisms seriously would look like. We will focus on the cultural evolution of social epistemic mechanisms and we will propose that Darwinian algorithms should be complemented with a cultural analogue of the error reduction mechanism proposed to account for human cognition, with a particular emphasis on the necessity to build independencies between different sub-systems in charge of tracking states of the world. To illustrate our point, we will present how the evolution of the legal system as epistemic systems can be understood as a process of building increasingly better independencies and how various criticisms of the actual legal system calls for building even more of them. (shrink)

A sense of agency (SoA) is the experience of subjective awareness regarding the control of one’s actions. Humans have a natural tendency to generate prediction models of the environment and adapt their models according to changes in the environment. The SoA is associated with the degree of the adaptation of the prediction models, e.g., insufficient adaptation causes low predictability and lowers the SoA over the environment. Thus, identifying the mechanisms behind the adaptation process of a prediction model related to the (...) SoA is essential for understanding the generative process of the SoA. In the first half of the current study, we constructed a mathematical model in which the SoA represents a likelihood value for a given observation (sensory feedback) in a prediction model of the environment and in which the prediction model is updated according to the likelihood value. From our mathematical model, we theoretically derived a testable hypothesis that the prediction model is updated according to a Bayesian rule or a stochastic gradient. In the second half of our study, we focused on the experimental examination of this hypothesis. In our experiment, human subjects were repeatedly asked to observe a moving square on a computer screen and press a button after a beep sound. The button press resulted in an abrupt jump of the moving square on the screen. Experiencing the various stochastic time intervals between the action execution button-press) and the consequent event square jumping caused gradual changes in the subjects' degree of their SoA. By comparing the above theoretical hypothesis with the experimental results, we concluded that the update (adaptation) rule of the prediction model based on the SoA is better described by a Bayesian update than by a stochastic gradient descent. (shrink)

Two striking claims are advanced on behalf of the free energy principle in cognitive science and philosophy: that it identifies a condition of the possibility of existence for self-organising systems; and that it has important implications for our understanding of how the brain works, defining a set of process theories—roughly, theories of the structure and functions of neural mechanisms—consistent with the free energy minimising imperative that it derives as a necessary feature of all self-organising systems. I argue that the conjunction (...) of claims and rests on a fallacy of equivocation. The FEP can be interpreted in two ways: as a claim about how it is possible to redescribe the existence of self-organising systems, and as a claim about how such systems maintain their existence. Although the Descriptive FEP plausibly does identify a condition of the possibility of existence for self-organising systems, it has no important implications for our understanding of how the brain works. Although the Explanatory FEP would have such implications if it were true, it does not identify a condition of the possibility of existence for self-organising systems. I consider various ways of responding to this conclusion, and I explore its implications for the role and importance of the FEP in cognitive science and philosophy. (shrink)

Enculturation is known to shape the perception of meter in music but this is not explicitly accounted for by current cognitive models of meter perception. We hypothesize that meter perception is a strategy for increasing the predictability of rhythmic patterns and that the way in which it is shaped by the cultural environment can be understood in terms of probabilistic predictive coding. Based on this hypothesis, we present a probabilistic model of meter perception that uses statistical properties of the relation (...) between rhythm and meter to infer meter from quantized rhythms. We show that our model can successfully predict annotated time signatures from quantized rhythmic patterns derived from folk melodies. Furthermore, we show that by inferring meter, our model can better predict the onsets of future events than a similar probabilistic model that does not infer meter. Finally, we demonstrate how our model can be used in a proof-of-concept simulation of enculturation. From this simulation, we derive a class of rhythms that are likely to be interpreted differently by enculturated listeners with different histories of exposure to rhythms. (shrink)

This article argues how one problem of computing lies in realizing a significant instance given a class or type. Analysis of a case study on digital narrative suggests two general processes for instantiating significant instances: interaction and optimization. The article then explains how the problem of universals needs to be deconstructed when trying to understand what type of entities significant instances are and what the process for obtaining them is.

The no-free-lunch theorems promote a skeptical conclusion that all possible machine learning algorithms equally lack justification. But how could this leave room for a learning theory, that shows that some algorithms are better than others? Drawing parallels to the philosophy of induction, we point out that the no-free-lunch results presuppose a conception of learning algorithms as purely data-driven. On this conception, every algorithm must have an inherent inductive bias, that wants justification. We argue that many standard learning algorithms should rather (...) be understood as model-dependent: in each application they also require for input a model, representing a bias. Generic algorithms themselves, they can be given a model-relative justification. (shrink)

One of the paradigms in judgment and decision-making involves decision-makers sample information before making a final consequential choice. In the sampling paradigm, certain computational models have been proposed where a set of single or distribution parameters is calibrated to the choice proportions of a group of participants. However, currently little is known on how aggregate and hierarchical models would account for choices made by individual participants in the sampling paradigm. In this paper, we test the ability of aggregate and hierarchical (...) models to explain choices made by individual participants. Several models, Ensemble, Cumulative Prospect Theory, Best Estimation and Simulation Techniques, Natural-Mean Heuristic, and Instance-Based Learning, had their parameters calibrated to individual choices in a large dataset involving the sampling paradigm. Later, these models were generalized to two large datasets in the sampling paradigm. Results revealed that the aggregate models accounted for individual choices better than hierarchical models upon generalization to problems that were like those encountered during calibration. Furthermore, the CPT model, which relies on differential valuing of gains and losses, respectively, performed better than other models during calibration and generalization on datasets with similar set of problems. The IBL model, relying on recency and frequency of sampled information, and the NMH model, relying on frequency of sampled information, performed better than other models during generalization to a challenging dataset. Sequential analyses of results from different models showed how these models accounted for transitions from the last sample to final choice in human data. We highlight the implications of using aggregate and hierarchical models in explaining individual choices from experience. (shrink)

This paper presents a computational model of the way humans inductively identify and aggregate concepts from the low-level stimuli they are exposed to. Based on the idea that humans tend to select the simplest structures, it implements a dynamic hierarchical chunking mechanism in which the decision whether to create a new chunk is based on an information-theoretic criterion, the Minimum Description Length (MDL) principle. We present theoretical justifications for this approach together with results of an experiment in which participants, exposed (...) to meaningless symbols, have been implicitly encouraged to create high-level concepts by grouping them. Results show that the designed model, called hereafter MDLChunker, makes precise quantitative predictions both on the kind of chunks created by the participants and also on the moment at which these creations occur. They suggest that the simplicity principle used to design MDLChunker is particularly efficient to model chunking mechanisms. The main interest of this model over existing ones is that it does not require any adjustable parameter. (shrink)

This commentary gives a personal perspective on modeling and modeling developments in cognitive science, starting in the 1950s, but focusing on the author’s personal views of modeling since training in the late 1960s, and particularly focusing on advances since the official founding of the Cognitive Science Society. The range and variety of modeling approaches in use today are remarkable, and for many, bewildering. Yet to come to anything approaching adequate insights into the infinitely complex fields of mind, brain, and intelligent (...) systems, an extremely wide array of modeling approaches is vital and necessary. (shrink)

This contribution aims at proposing a framework for articulating different kinds of “normativities” that are and can be attributed to “algorithmic systems.” The technical normativity manifests itself through the lineage of technical objects. The norm expresses a technical scheme’s becoming as it mutates through, but also resists, inventions. The genealogy of neural networks shall provide a powerful illustration of this dynamic by engaging with their concrete functioning as well as their unsuspected potentialities. The socio-technical normativity accounts for the manners in (...) which engineers, as actors folded into socio-technical networks, willingly or unwittingly, infuse technical objects with values materialized in the system. Surveillance systems’ design will serve here to instantiate the ongoing mediation through which algorithmic systems are endowed with specific capacities. The behavioral normativity is the normative activity, in which both organic and mechanical behaviors are actively participating, undoing the identification of machines with “norm following,” and organisms with “norminstitution”. This proposition productively accounts for the singularity of machine learning algorithms, explored here through the case of recommender systems. The paper will provide substantial discussions of the notions of “normative” by cutting across history and philosophy of science, legal, and critical theory, as well as “algorithmics,” and by confronting our studies led in engineering laboratories with critical algorithm studies. (shrink)

Data mining’s primary clustering method has several uses, including gene analysis. A set of unlabeled data is divided into clusters using data features in a clustering study, which is an unsupervised learning problem. Data in a cluster are more comparable to one another than to those in other groups. However, the number of clusters has a direct impact on how well the K-means algorithm performs. In order to find the best solutions for these real-world optimization issues, it is necessary to (...) use techniques that properly explore the search spaces. In this research, an enhancement of K-means clustering is proposed by applying an equilibrium optimization approach. The suggested approach adjusts the number of clusters while simultaneously choosing the best attributes to find the optimal answer. The findings establish the usefulness of the suggested method in comparison to existing algorithms in terms of intra-cluster distances and Rand index based on five datasets. Through the results shown and a comparison of the proposed method with the rest of the traditional methods, it was found that the proposal is better in terms of the internal dimension of the elements within the same cluster, as well as the Rand index. In conclusion, the suggested technique can be successfully employed for data clustering and can offer significant support. (shrink)

Emotion research will leap forward when its focus changes from comparing averaged statistics of self-report data across people experiencing emotion in laboratories to characterizing patterns of data from individuals and clusters of similar individuals experiencing emotion in real life. Such an advance will come about through engineers and psychologists collaborating to create new ways for people to measure, share, analyze, and learn from objective emotional responses in situations that truly matter to people. This approach has the power to greatly advance (...) the science of emotion while also providing personalized help to participants in the research. (shrink)

Why is interaction so simple? This article presents a theory of interaction based on the use of shared representations as “coordination tools” (e.g., roundabouts that facilitate coordination of drivers). By aligning their representations (intentionally or unintentionally), interacting agents help one another to solve interaction problems in that they remain predictable, and offer cues for action selection and goal monitoring. We illustrate how this strategy works in a joint task (building together a tower of bricks) and discuss its requirements from a (...) computational viewpoint. (shrink)

An intervention may have an effect on units other than those to which it was administered. This phenomenon is called interference and it usually goes unmodeled. In this paper, we propose to combine Lauritzen-Wermuth-Frydenberg and Andersson-Madigan-Perlman chain graphs to create a new class of causal models that can represent both interference and non-interference relationships for Gaussian distributions. Specifically, we define the new class of models, introduce global and local and pairwise Markov properties for them, and prove their equivalence. We also (...) propose an algorithm for maximum likelihood parameter estimation for the new models, and report experimental results. Finally, we show how to compute the effects of interventions in the new models. (shrink)

Suppose that we are interested in the average causal effect of a binary treatment on an outcome when this relationship is confounded by a binary confounder. Suppose that the confounder is unobserved but a nondifferential proxy of it is observed. We show that, under certain monotonicity assumption that is empirically verifiable, adjusting for the proxy produces a measure of the effect that is between the unadjusted and the true measures.

“Decisions from experience” (DFE) refers to a body of work that emerged in research on behavioral decision making over the last decade. One of the major experimental paradigms employed to study experienced-based choice is the “sampling paradigm”, which serves as a model of decision making under limited knowledge about the statistical structure of the world. In this paradigm respondents are presented with two payoff distributions, which, in contrast to standard approaches in behavioral economics, are specified not in terms of explicit (...) outcome-probability information, but by the opportunity to sample outcomes from each distribution without economic consequences. Participants are encouraged to explore the distributions until they feel confident enough to decide from which they would prefer to draw from in a final trial involving real monetary payoffs. One commonly employed measure to characterize the behavior of participants in the sampling paradigm is the sample size, that is, the number of outcome draws which participants choose to obtain from each distribution prior to terminating sampling. A natural question that arises in this context concerns the “optimal” sample size, which could be used as a normative benchmark to evaluate human sampling behavior in DFE. In this theoretical manuscript, we relate the DFE sampling paradigm to the classical statistical decision theoretic literature and, under a probabilistic inference assumption, evaluate optimal sample sizes for decisions from experience. In our treatment we go beyond analytically established results by showing how the classical statistical decision theoretic framework can be used to derive optimal sample sizes under arbitrary, but numerically evaluable, constraints. Finally, we critically evaluate the value of deriving optimal sample sizes under this framework as testable predictions for the experimental study of sampling behavior in DFE. (shrink)

The use of cognitive systems like pattern recognition or video tracking technology in security applications is becoming ever more common. The paper considers cases in which the cognitive systems are meant to assist human tasks by providing information, but the final decision is left to the human. All these systems and their various applications have a common feature: an intrinsic difference in how a situation or an event is assessed by a human being and a cognitive system. This difference, which (...) here is named “the model gap,” is analyzed pertaining to its epistemic role and its ethical consequences. The main results are as follows: (1) The model gap is not a problem, which might be solved by future research, but the central feature of cognitive systems. (2) The model gap appears on two levels: the aspects of the world, which are evaluated, and the way they are processed. This leads to changes in central concepts. While differences on the first level often are the very reason for the deployment of cognitive systems, the latter is hard to notice and often goes unreflected. (3) Such a missing reflection is ethically problematic because the human is meant to give the final judgment. It is particularly problematic in security applications where it might lead to a conflation of descriptive and normative concepts. (4) The idea of the human operator having the last word is based on an assumption of independent judgment. This assumption is flawed for two reasons: The cognitive system and the human operators form a “hybrid system” the components of which cannot be assessed independently; and additional modes of judgment might pose new ethical problems. (shrink)

Over the last fifteen years, an ambitious explanatory framework has been proposed to unify explanations across biology and cognitive science. Active inference, whose most famous tenet is the free energy principle, has inspired excitement and confusion in equal measure. Here, we lay the ground for proper critical analysis of active inference, in three ways. First, we give simplified versions of its core mathematical models. Second, we outline the historical development of active inference and its relationship to other theoretical approaches. Third, (...) we describe three different kinds of claim -- labelled mathematical, empirical and general -- routinely made by proponents of the framework, and suggest dialectical links between them. Overall, we aim to increase philosophical understanding of active inference so that it may be more readily evaluated. -/- This the final submitted version of the Introduction to the Topical Collection "The Free Energy Principle: From Biology to Cognition", forthcoming in Biology & Philosophy. (shrink)

Classic conceptions of model simplicity for machine learning are mainly based on the analysis of the structure of the model. Bayesian, Frequentist, information theoretic and expressive power concepts are the best known of them, which are reviewed in this work, along with their underlying assumptions and weaknesses. These approaches were developed before the advent of the Big Data deluge, which has overturned the importance of structural simplicity. The computational simplicity concept is presented, and it is argued that it is more (...) encompassing and closer to actual machine learning practices than the classic ones. In order to process the huge datasets which are commonplace nowadays, the computational complexity of the learning algorithm is the decisive factor to assess the viability of a machine learning strategy, while the classic accounts of simplicity play a surrogate role. Some of the desirable features of computational simplicity derive from its reliance on the learning system concept, which integrates key aspects of machine learning that are ignored by the classic concepts. Moreover, computational simplicity is directly associated with energy efficiency. In particular, the question of whether the maximum possibly achievable predictive accuracy should be attained, no matter the economic cost of the associated energy consumption pattern, is considered. (shrink)

In the last few years, biologists and computer scientists have claimed that the introduction of data science techniques in molecular biology has changed the characteristics and the aims of typical outputs (i.e. models) of such a discipline. In this paper we will critically examine this claim. First, we identify the received view on models and their aims in molecular biology. Models in molecular biology are mechanistic and explanatory. Next, we identify the scope and aims of data science (machine learning in (...) particular). These lie mainly in the creation of predictive models which performances increase as data set increases. Next, we will identify a tradeoff between predictive and explanatory performances by comparing the features of mechanistic and predictive models. Finally, we show how this a priori analysis of machine learning and mechanistic research applies to actual biological practice. This will be done by analyzing the publications of a consortium—The Cancer Genome Atlas—which stands at the forefront in integrating data science and molecular biology. The result will be that biologists have to deal with the tradeoff between explaining and predicting that we have identified, and hence the explanatory force of the ‘new’ biology is substantially diminished if compared to the ‘old’ biology. However, this aspect also emphasizes the existence of other research goals which make predictive force independent from explanation. (shrink)

In this paper, I argue that enactivism and computationalism—two seemingly incompatible research traditions in modern cognitive science—can be fruitfully reconciled under the framework of the free energy principle. FEP holds that cognitive systems encode generative models of their niches and cognition can be understood in terms of minimizing the free energy of these models. There are two philosophical interpretations of this picture. A computationalist will argue that as FEP claims that Bayesian inference underpins both perception and action, it entails a (...) concept of cognition as a computational process. An enactivist, on the other hand, will point out that FEP explains cognitive systems as constantly self-organizing to non-equilibrium steady-state. My claim is that these two interpretations are both true at the same time and that they enlighten each other. (shrink)

In this paper, we consider how certain longstanding philosophical questions about mental representation may be answered on the assumption that cognitive and perceptual systems implement hierarchical generative models, such as those discussed within the prediction error minimization framework. We build on existing treatments of representation via structural resemblance, such as those in Gładziejewski :559–582, 2016) and Gładziejewski and Miłkowski, to argue for a representationalist interpretation of the PEM framework. We further motivate the proposed approach to content by arguing that it (...) is consistent with approaches implicit in theories of unsupervised learning in neural networks. In the course of this discussion, we argue that the structural representation proposal, properly understood, has more in common with functional-role than with causal/informational or teleosemantic theories. In the remainder of the paper, we describe the PEM framework for approximate Bayesian inference in some detail, and discuss how structural representations might arise within the proposed Bayesian hierarchies. After explicating the notion of variational inference, we define a subjectively accessible measure of misrepresentation for hierarchical Bayesian networks by appeal to the Kullbach–Leibler divergence between posterior generative and approximate recognition densities, and discuss a related measure of objective misrepresentation in terms of correspondence with the facts. (shrink)

I examine the construction and evaluation of machine learning binary classification models. These models are increasingly used for societal applications such as classifying patients into two categories according to the presence or absence of a certain disease like cancer and heart disease. I argue that the construction of ML classification models involves an optimisation process aiming at the minimization of the inductive risk associated with the intended uses of these models. I also argue that the construction of these models is (...) underdetermined by the available data, and that this makes it necessary for ML modellers to make social value judgments in determining the error costs used in ML optimization. I thus suggest that the assessment of the inductive risk with respect to the social values of the intended users is an integral part of the construction and evaluation of ML classification models. I also discuss the implications of this conclusion for the philosophical debate concerning inductive risk. (shrink)

Scientists routinely solve the problem of supplementing one’s store of variables with new theoretical posits that can explain the previously inexplicable. The banality of success at this task obscures a remarkable fact. Generating hypotheses that contain novel variables and accurately project over a limited amount of additional data is so difficult—the space of possibilities so vast—that succeeding through guesswork is overwhelmingly unlikely despite a very large number of attempts. And yet scientists do generate hypotheses of this sort in very few (...) tries. I argue that this poses a dilemma: either the long history of scientific success is a miracle, or there exists at least one method or algorithm for generating novel hypotheses with at least limited projectibility on the basis of what’s available to the scientist at a time, namely a set of observations, the history of past conjectures, and some prior theoretical commitments. In other words, either ordinary scientific success is miraculous or there exists a logic of discovery at the heart of actual scientific method. (shrink)

We present a generalization of the problem of pattern recognition to arbitrary probabilistic models. This version deals with the problem of recognizing an individual pattern among a family of different species or classes of objects which obey probabilistic laws which do not comply with Kolmogorov’s axioms. We show that such a scenario accommodates many important examples, and in particular, we provide a rigorous definition of the classical and the quantum pattern recognition problems, respectively. Our framework allows for the introduction of (...) non-trivial correlations between the different species involved, opening the door to a new way of harnessing these physical resources for solving pattern recognition problems. Finally, we present some examples and discuss the computational complexity of the quantum pattern recognition problem, showing that the most important quantum computation algorithms can be described as non-Kolmogorovian pattern recognition problems. (shrink)

The free energy principle says that any self-organising system that is at nonequilibrium steady-state with its environment must minimize its free energy. It is proposed as a grand unifying principle for cognitive science and biology. The principle can appear cryptic, esoteric, too ambitious, and unfalsifiable—suggesting it would be best to suspend any belief in the principle, and instead focus on individual, more concrete and falsifiable ‘process theories’ for particular biological processes and phenomena like perception, decision and action. Here, I explain (...) the free energy principle, and I argue that it is best understood as offering a conceptual and mathematical analysis of the concept of existence of self-organising systems. This analysis offers a new type of solution to long-standing problems in neurobiology, cognitive science, machine learning and philosophy concerning the possibility of normatively constrained, self-supervised learning and inference. The principle can therefore uniquely serve as a regulatory principle for process theories, to ensure that process theories conforming to it enable self-supervision. This is, at least for those who believe self-supervision is a foundational explanatory task, good reason to believe the free energy principle. (shrink)

Machine learning systems have shown great potential for performing or supporting inferential reasoning through analyzing large data sets, thereby potentially facilitating more informed decision-making. However, a hindrance to such use of ML systems is that the predictive models created through ML are often complex, opaque, and poorly understood, even if the programs “learning” the models are simple, transparent, and well understood. ML models become difficult to trust, since lay-people, specialists, and even researchers have difficulties gauging the reasonableness, correctness, and reliability (...) of the inferences performed. In this article, we argue that bridging this gap in the understanding of ML models and their reasonableness requires a focus on developing an improved methodology for their creation. This process has been likened to “alchemy” and criticized for involving a large degree of “black art,” owing to its reliance on poorly understood “best practices”. We soften this critique and argue that the seeming arbitrariness often is the result of a lack of explicit hypothesizing stemming from an empiricist and myopic focus on optimizing for predictive performance rather than from an occult or mystical process. We present some of the problems resulting from the excessive focus on optimizing generalization performance at the cost of hypothesizing about the selection of data and biases. We suggest embedding ML in a general logic of scientific discovery similar to the one presented by Charles Sanders Peirce, and present a recontextualized version of Peirce’s scientific hypothesis adjusted to ML. (shrink)

Despite its centrality in the philosophy of cognitive science, there has been little prior philosophical work engaging with the notion of representation in contemporary NLP practice. This paper attempts to fill that lacuna: drawing on ideas from cognitive science, I introduce a framework for evaluating the representational claims made about components of neural NLP models, proposing three criteria with which to evaluate whether a component of a model represents a property and operationalising these criteria using probing classifiers, a popular analysis (...) technique in NLP (and deep learning more broadly). The project of operationalising a philosophically-informed notion of representation should be of interest to both philosophers of science and NLP practitioners. It affords philosophers a novel testing-ground for claims about the nature of representation, and helps NLPers organise the large literature on probing experiments, suggesting novel avenues for empirical research. (shrink)