Nenad Filipovic

Certain characteristics make machine learning a powerful tool for processing large amounts of data, and also particularly unsuitable for explanatory purposes. There are worries that its increasing use in science may sideline the explanatory goals of research. We analyze the key characteristics of ML that might have implications for the future directions in scientific research: epistemic opacity and the ‘theory-agnostic’ modeling. These characteristics are further analyzed in a comparison of ML with the traditional statistical methods, in order to demonstrate what it is specifically that makes ML methodology substantially unsuitable for reaching explanations. The analysis is given broader philosophical context by connecting it with the views on the role of prediction and explanation in science, their relationship, and the value of explanation. We proceed to show, first, that ML disrupts the relationship between prediction and explanation commonly understood as a functional relationship. Then we show that the value of explanation is not exhausted in purely predictive functions, but rather has a ubiquitously recognized value for both science and everyday life. We then invoke two hypothetical scenarios with different degrees of automatization of science, which help test our intuitions on the role of explanation in science. The main question we address is whether ML will reorient or otherwise impact our standard explanatory practice. We conclude with a prognosis that ML would diversify science into purely predictively oriented research based on ML-like techniques and, on the other hand, remaining faithful to anthropocentric research focused on the search for explanation.

The ecological approach to rationality, exemplified in theories such as Gerd Gigerenzer’s Fast and Frugal Heuristics, posits that different decision-making procedures are rational in various decision environments. In this regard, the ecological approach sharply contrasts with standard normative theories like Expected Utility Theory, which is often viewed merely as an analysis of bounded rationality. In this article, I aim to provide philosophical foundations for the ecological approach as a framework for normative rationality. I draw on John Norton’s Material Theory of Induction to argue that the appropriate method for addressing uncertain evidence is domain-dependent and sensitive to the environment, as the ecological approach suggests. Furthermore, I argue that, motivated by insights from material theory, we should re-examine the claim that consistency requirements are universal standards of rationality in three ways: first, the plausibility of consistency axioms depends on the decision environment; second, the application of standard theories is contingent upon the structural conditions of that environment; and third, arguments for axioms, such as money-pump arguments, are plausible only in specific domains. Finally, I argue that the ecological approach can provide a substantive account of rationality norms and that its performance-related criteria for these norms should facilitate the evaluation and modification of agents’ decision-making. These arguments underscore the advantages of the ecological approach over standard theories while situating it within the broader philosophical framework of induction.

In this article, I examine the consequences of Mamou’s recent solution to the so-called problem of reindividuation. The problem is first proposed by John Broome as an argument against the Humean view of rationality and subjective decision theory. Essentially, Broome argues that the reindividuation of outcomes must be constrained in some way if the axioms of rationality are not to be vacuous, but the constraints on individuation cannot be consistent with the Humean view of rationality. Mamou offers an elegant Humean solution to this problem. We do not need such rules for reindividuation since the subjective decision theory operates under the assumption of maximal relevance of description, and only under that assumption do the axioms tell anything. Thus the problem of reindividuation dissolves in a quite natural and simple way. However, I argue that if we connect this solution to the topic of justification for the requirements of rationality, problems arise. I examine the influence of the solution on the common argument for the axioms of decision theory, the so-called money-pump arguments. These arguments are a staple in the justification of the axioms of subjective decision theory. I argue that they are in collision with Mamou’s solution. This collision brings us to the dilemma: either accept Broome’s conclusions or lose the money-pump arguments as an effective tool for justification of the axioms. I conclude that neither of the options is really good for the defenders of the Humean view of rationality.

Sake i ostali inkongruentni protivdelovi su dovoljni da odbacimo relacionizam, ili je bar tako mislio Kant, jos od nekih svojih prekritickih spisa. Argumenti sa inkongruentnim protivdelovima su elegantni i efektni, i privukli su prilicno veliku paznju kod odredjenog broja autora, koji su argumente kritikovali na razne nacine, ili ih branili. Rasprave su se pritom odaljile od originalnog Kantovog argumenta, i od duha tog vremena, i primila obelezja moderne filozofije i geometrije. U ovom tekstu bi trebalo da pokazem da Kant, kao i oni koji su ga kasnije branili, nisu postigli svoj cilj - nije dat konkluzivan argument protiv relacionizma.