Machine learning operates at the intersection of statistics and computer science. This raises the question as to its underlying methodology. While much emphasis has been put on the close link between the process of learning from data and induction, the falsificationist component of machine learning has received minor attention. In this paper, we argue that the idea of falsification is central to the methodology of machine learning. It is commonly thought that machine learning algorithms infer general prediction rules from past (...) observations. This is akin to a statistical procedure by which estimates are obtained from a sample of data. But machine learning algorithms can also be described as choosing one prediction rule from an entire class of functions. In particular, the algorithm that determines the weights of an artificial neural network operates by empirical risk minimization and rejects prediction rules that lack empirical adequacy. It also exhibits a behavior of implicit regularization that pushes hypothesis choice toward simpler prediction rules. We argue that taking both aspects together gives rise to a falsificationist account of artificial neural networks. (shrink)

The paper proposes a synthesis between human scientists and artificial representation learning models as a way of augmenting epistemic warrants of realist theories against various anti-realist attempts. Towards this end, the paper fleshes out unconceived alternatives not as a critique of scientific realism but rather a reinforcement, as it rejects the retrospective interpretations of scientific progress, which brought about the problem of alternatives in the first place. By utilising adversarial machine learning, the synthesis explores possibility spaces of available evidence for (...) unconceived alternatives providing modal knowledge of what is possible therein. As a result, the epistemic warrant of synthesised realist theories should emerge bolstered as the underdetermination by available evidence gets reduced. While shifting the realist commitment away from theoretical artefacts towards modalities of the possibility spaces, the synthesis comes out as a kind of perspectival modelling. (shrink)

I argue for the causal character of modeling in data-intensive science, contrary to widespread claims that big data is only concerned with the search for correlations. After discussing the concept of data-intensive science and introducing two examples as illustration, several algorithms are examined. It is shown how they are able to identify causal relevance on the basis of eliminative induction and a related difference-making account of causation. I then situate data-intensive modeling within a broader framework of an epistemology of scientific (...) knowledge. In particular, it is shown to lack a pronounced hierarchical, nested structure. The significance of the transition to such “horizontal” modeling is underlined by the concurrent emergence of novel inductive methodology in statistics such as non-parametric statistics. Data-intensive modeling is well equipped to deal with various aspects of causal complexity arising especially in the higher level and applied sciences. (shrink)

This paper responds to Achinstein’s criticism of the thesis that the only empirical fact that can affect the truth of an objective evidence claim such as ‘e is evidence for h’ (or ‘e confirms h to degree r’) is the truth of e. It shows that cases involving evidential flaws, which form the basis for Achinstein’s objections to the thesis, can satisfactorily be accounted for by appeal to changes in background information and working assumptions. The paper also argues that the (...) a priori and empirical accounts of evidence are on a par when we consider scientific practice, but that a study of artificial intelligence might serve to differentiate them. (shrink)

Epistemologists have debated at length whether scientific discovery is a rational and logical process. If it is, according to the Artificial Intelligence hypothesis, it should be possible to write computer programs able to discover laws or theories; and if such programs were written, this would definitely prove the existence of a logic of discovery. Attempts in this direction, however, have been unsuccessful: the programs written by Simon's group, indeed, infer famous laws of physics and chemistry; but having found no new (...) law, they cannot properly be considered discovery machines. The programs written in the Turing tradition, instead, produced new and useful empirical generalization, but no theoretical discovery, thus failing to prove the logical character of the most significant kind of discoveries. A new cognitivist and connectionist approach by Holland, Holyoak, Nisbett and Thagard, looks more promising. Reflection on their proposals helps to understand the complex character of discovery processes, the abandonment of belief in the logic of discovery by logical positivists, and the necessity of a realist interpretation of scientific research. (shrink)

Philosophy Compass, Volume 17, Issue 1, January 2022.

It is considered in some quarters that hypothesis‐driven methods are the only valuable, reliable or significant means of scientific advance. Data‐driven or ‘inductive’ advances in scientific knowledge are then seen as marginal, irrelevant, insecure or wrong‐headed, while the development of technology—which is not of itself ‘hypothesis‐led’ (beyond the recognition that such tools might be of value)—must be seen as equally irrelevant to the hypothetico‐deductive scientific agenda. We argue here that data‐ and technology‐driven programmes are not alternatives to hypothesis‐led studies in (...) scientific knowledge discovery but are complementary and iterative partners with them. Many fields are data‐rich but hypothesis‐poor. Here, computational methods of data analysis, which may be automated, provide the means of generating novel hypotheses, especially in the post‐genomic era. BioEssays 26:99–105, 2004. © 2003 Wiley Periodicals, Inc. (shrink)

Philosophy Compass, Volume 17, Issue 1, January 2022.

Philosophical analyses of scientific methodology have long understood intuition to be incompatible with a rule based reasoning that is often considered necessary for a rational scientific method. This paper seeks to challenge this contention by highlighting the indispensable role that intuition plays in the application of methodologies for scientific discovery. In particular, it seeks to outline a positive role for intuition and personal judgment in scientific discovery by exploring a comparison between the use of heuristic reasoning in scientific practice and (...) engineering design. While these discussions share many features, it will also be shown that the successful use of heuristics in engineering design is often considered to depend on a crucial factor that is markedly absent from accounts of the use of heuristics in scientific discovery; experienced judgment. In the final sections of this paper, I will compare attitudes to the role of computer analysis in scientific and engineering practices, with the aim of showing how the limitations of scientific discovery machines reveal the need for including intuition in philosophical accounts of heuristic reasoning in scientific discovery. (shrink)

In this paper I argue for a place for logic inscientific methodology, at the same level asthat of computational and historicalapproaches. While it is well known that a awhole generation of philosophers dismissedLogical Positivism (not just for the logicthough), there are at least two reasons toreconsider logical approaches in the philosophyof science. On the one hand, the presentsituation in logical research has gone farbeyond the formal developments that deductivelogic reached last century, and new researchincludes the formalization of several othertypes of (...) reasoning, like induction andabduction. On the other hand, we call for abalanced Philosophy of Science, one inwhich both methods, the formal and thehistorical may be complementary, togetherproviding a pluralistic view of science, inwhich no method is the predominant one. (shrink)