The paper contains a survey of (mainly unpublished) adaptive logics of inductive generalization. These defeasible logics are precise formulations of certain methods. Some attention is also paid to ways of handling background knowledge, introducing mere conjectures, and the research guiding capabilities of the logics.

Inductive generalization, where people go beyond the data provided, is a basic cognitive capability, and it underpins theoretical accounts of learning, categorization, and decision making. To complete the inductive leap needed for generalization, people must make a key ‘‘sampling’’ assumption about how the available data were generated. Previous models have considered two extreme possibilities, known as strong and weak sampling. In strong sampling, data are assumed to have been deliberately generated as positive examples of a concept, (...) whereas in weak sampling, data are assumed to have been generated without any restrictions. We develop a more general account of sampling that allows for an intermediate mixture of these two extremes, and we test its usefulness. In two experiments, we show that most people complete simple one-dimensional generalization tasks in a way that is consistent with their believing in some mixture of strong and weak sampling, but that there are large individual differences in the relative emphasis different people give to each type of sampling. We also show experimentally that the relative emphasis of the mixture is influenced by the structure of the available information. We discuss the psychological meaning of mixing strong and weak sampling, and possible extensions of our modeling approach to richer problems of inductive generalization. (shrink)

The account of meta-induction (G. Schurz, Hume’s problem solved: the optimality of meta-induction, MIT Press, Cambridge, 2019) proposes a two-step solution to the problem of induction. Step 1 consists in a mathematical a priori justification of the predictive optimality of meta-induction, upon which step 2 builds a meta-inductive a posteriori justification of object-induction based on its superior track record (Sect. 1). Sterkenburg (Br J Philos Sci, forthcoming. 10.1086/717068/) challenged this account by arguing that meta-induction can only provide a (non-circular) (...) justification of inductive predictions for now and for the next future, but not a justification of inductive generalizations (Sect. 2). This paper develops a meta-inductive method that does provide an a posteriori justification of inductive generalizations, in the form of exchangeability conditions (Sect. 3). In Sect. 4, a limitation of the proposed method is worked out: while the method can justify weakly lawlike generalizations, the justification of strongly lawlike generalizations (claimed to hold for all eternity) requires epistemic principles going beyond meta-induction based on predictive success. (shrink)

Judging similarities among objects, events, and experiences is one of the most basic cognitive abilities, allowing us to make predictions and generalizations. The main assumption in similarity judgment is that people selectively attend to salient features of stimuli and judge their similarities on the basis of the common and distinct features of the stimuli. However, it is unclear how people select features from stimuli and how they weigh features. Here, we present a computational method that helps address these questions. Our (...) procedure combines image-processing techniques with a machine-learning algorithm and assesses feature weights that can account for both similarity and categorization judgment data. Our analysis suggests that a small number of local features are particularly important to explain our behavioral data. (shrink)

Our empirical beliefs beyond sense impressions and the memories of them cannot be justified. Logically we must be complete skeptics. This is the consequence of Hume’s skeptical argument against induction. Should we accept this conclusion or not? This is the so-called problem of induction. In this paper I propose a new solution that belongs to the 'dissolutionist’ tradition (Strawson 1952, Okasha 2001). Through criticizing the core argument in Hume’s skepticism I claim the following: Hume’s skepticism against induction does not succeed (...) in denying the soundness of induction. We are not required to justify induction generally. (shrink)

Inductive generalization asserts that what obtains in known instances can be generalized to all. Its original form is enumerative induction, the earliest form of inductive inference, and it has been elaborated in various ways, largely with the goal of extending its reach. Its principal problem is that it supplies no intrinsic notion of strength of support so that one cannot tell if the generalization has weak or strong support.

In reinforcement learning, a decision maker searching for the most rewarding option is often faced with the question: What is the value of an option that has never been tried before? One way to frame this question is as an inductive problem: How can I generalize my previous experience with one set of options to a novel option? We show how hierarchical Bayesian inference can be used to solve this problem, and we describe an equivalence between the Bayesian model (...) and temporal difference learning algorithms that have been proposed as models of RL in humans and animals. According to our view, the search for the best option is guided by abstract knowledge about the relationships between different options in an environment, resulting in greater search efficiency compared to traditional RL algorithms previously applied to human cognition. In two behavioral experiments, we test several predictions of our model, providing evidence that humans learn and exploit structured inductive knowledge to make predictions about novel options. In light of this model, we suggest a new interpretation of dopaminergic responses to novelty. (shrink)

Aris Spanos and Deborah Mayo’s error-statistical approach to statistical modeling and inference adopts the reliability of inductive inference as a primary criterion for statistical model and estimator selection (e.g., curve fitting). In this paper, we expand the error-statistical approach’s adoption of reliable inductive inference by scrutinizing the epistemic legitimacy of contemporary techniques leveraged in data science. We argue that data validation and testing potentially provides a direct, measurable method of evaluating evidence for reliable inductive inferences in cases (...) where the error-statistical approach is not easily applied, and conclude with an exploration of core methodological foils to the reliability of inductive inference revealed by this argument. (shrink)

Probability theory, epistemically interpreted, provides an excellent, if not the best available account of inductive reasoning. This is so because there are general and definite rules for the change of subjective probabilities through information or experience; induction and belief change are one and same topic, after all. The most basic of these rules is simply to conditionalize with respect to the information received; and there are similar and more general rules. 1 Hence, a fundamental reason for the epistemological success (...) of probability theory is that there at all exists a well-behaved concept of conditional probability. Still, people have, and have reasons for, various concerns over probability theory. One of these is my starting point: Intuitively, we have the notion of plain belief; we believe propositions2 to be true (or to be false or neither). Probability theory, however, offers no formal counterpart to this notion. Believing A is not the same as having probability 1 for A, because probability 1 is incorrigible3; but plain belief is clearly corrigible. And believing A is not the same as giving A a probability larger than some 1 - c, because believing A and believing B is usually taken to be equivalent to believing A & B.4 Thus, it seems that the formal representation of plain belief has to take a non-probabilistic route. Indeed, representing plain belief seems easy enough: simply represent an epistemic state by the set of all propositions believed true in it or, since I make the common assumption that plain belief is deductively closed, by the conjunction of all propositions believed true in it. But this does not yet provide a theory of induction, i.e. an answer to the question how epistemic states so represented are changed tbrough information or experience. There is a convincing partial answer: if the new information is compatible with the old epistemic state, then the new epistemic state is simply represented by the conjunction of the new information and the old beliefs. This answer is partial because it does not cover the quite common case where the new information is incompatible with the old beliefs. It is, however, important to complete the answer and to cover this case, too; otherwise, we would not represent plain belief as conigible. The crucial problem is that there is no good completion. When epistemic states are represented simply by the conjunction of all propositions believed true in it, the answer cannot be completed; and though there is a lot of fruitful work, no other representation of epistemic states has been proposed, as far as I know, which provides a complete solution to this problem. In this paper, I want to suggest such a solution. In [4], I have more fully argued that this is the only solution, if certain plausible desiderata are to be satisfied. Here, in section 2, I will be content with formally defining and intuitively explaining my proposal. I will compare my proposal with probability theory in section 3. It will turn out that the theory I am proposing is structurally homomorphic to probability theory in important respects and that it is thus equally easily implementable, but moreover computationally simpler. Section 4 contains a very brief comparison with various kinds of logics, in particular conditional logic, with Shackle's functions of potential surprise and related theories, and with the Dempster - Shafer theory of belief functions. (shrink)

In ‘Generalization and Induction: Misconceptions, Clarifications, and a Classification of Induction’, we comment on Lee and Baskerville’s paper ‘Generalizing Generalizability in Information Systems Research’, which attempts to clarify the concept of generalization and classify it into four types. Our commentary discusses the misconceptions in their paper and proposes an alternative classification of induction. Their response ‘Conceptualizing Generalizability: New Contributions and a Reply’ perpetuates their misconceptions and create new ones. The purpose of this rejoinder is to highlight the major (...) problems both in their original paper and in their reply and to provide further clarifications. Lee and Baskerville’s so-called ‘new language’ of describing research activities based on their concept of generalization is confusing. Their classification abuses the term ‘generalize’ and is self-contradictory. Hence, contrary to their claim, their classification and ours are not compatible. Also contrary to their claim that we advocate paradigmatic domination, our commentary is just about the correct use of terms such as ‘generalize’, ‘induction’ and ‘deduction’. (shrink)

The cognitive unbinding paradigm suggests that the synthesis of neural information is attenuated by general anesthesia. Here, we analyzed the functional organization of brain activities in the conscious and anesthetized states, based on functional segregation and integration. Electroencephalography recordings were obtained from 14 subjects undergoing induction of general anesthesia with propofol. We quantified changes in mean information integration capacity in each band of the EEG. After induction with propofol, mean information integration capacity was reduced most prominently in the γ band (...) of the EEG . Furthermore, we demonstrate that loss of consciousness is reflected by the breakdown of the spatiotemporal organization of γ waves. We conclude that induction of general anesthesia with propofol reduces the capacity for information integration in the brain. These data directly support the information integration theory of consciousness and the cognitive unbinding paradigm of general anesthesia. (shrink)

In Generalizing Generalizability in Information Systems Research Lee and Baskerville (2003) attempt to clarify generalization and distinguish four types of generalization. Although this is a useful objective, what they call generalization is often not generalization at all in the proper sense of the word. We elucidate generalization by locating their major errors. A main source of these is their failure to understand the depth of Hume’s problem of induction. We give a thorough explication of the (...) problem and then give a solution. Lastly, we propose an alternative taxonomy of generalization: theoretical, within-population, cross-population, contextual, and temporal. (shrink)

The first example of a simultaneous inductive-recursive definition in intuitionistic type theory is Martin-Löf's universe á la Tarski. A set U 0 of codes for small sets is generated inductively at the same time as a function T 0 , which maps a code to the corresponding small set, is defined by recursion on the way the elements of U 0 are generated. In this paper we argue that there is an underlying general notion of simultaneous inductive-recursive definition (...) which is implicit in Martin-Löf's intuitionistic type theory. We extend previously given schematic formulations of inductive definitions in type theory to encompass a general notion of simultaneous induction-recursion. This enables us to give a unified treatment of several interesting constructions including various universe constructions by Palmgren, Griffor, Rathjen, and Setzer and a constructive version of Aczel's Frege structures. Consistency of a restricted version of the extension is shown by constructing a realisability model in the style of Allen. (shrink)

We present a framework that provides a logic for science by generalizing the notion of logical (Tarskian) consequence. This framework will introduce hierarchies of logical consequences, the first level of each of which is identified with deduction. We argue for identification of the second level of the hierarchies with inductive inference. The notion of induction presented here has some resonance with Popper's notion of scientific discovery by refutation. Our framework rests on the assumption of a restricted class of structures (...) in contrast to the permissibility of classical first-order logic. We make a distinction between deductive and inductive inference via the notions of compactness and weak compactness. Connections with the arithmetical hierarchy and formal learning theory are explored. For the latter, we argue against the identification of inductive inference with the notion of learnable in the limit. Several results highlighting desirable properties of these hierarchies of generalized logical consequence are also presented. (shrink)

Kevin T. Kelly. General Characteristics of Inductive Inference Over Arbitrary Sets of Data Representations.

In this paper I present a game-theoretical approach to the problem of induction. I investigate the comparative success of prediction methods by mathematical analysis and computer programming. Hume's problem lies in the fact that although the success of object-inductive prediction strategies is quite robust, they cannot be universally optimal. My proposal towards a solution of the problem of induction is meta-induction. I show that there exist meta-inductive prediction strategies whose success is universally optimal, modulo short-run losses which are (...) upper-bounded. I then turn to the implications of my approach for the evolution of cognition. In the final section I suggest a revision of the paradigm of bounded rationality by introducing the distinction between local, general and universal prediction strategies. (shrink)

In this paper I show that Einstein made essential use of aim-oriented empiricism in scientific practice in developing special and general relativity. I conclude by considering to what extent Einstein came explicitly to advocate aim-oriented empiricism in his later years.

In this paper Venn's account of probability inference and induction is examined, tracing their differences as well as how they ‘co-operate’ in inferences from particulars to particulars. We discuss the role of mathematical idealizations in making probability inferences, the celebrated rule of succession and we delve into the nature of the reference class problem arguing that for Venn it is a common problem for both induction and inference in probability. Our approach is both historical and philosophical attempting to sketch Venn's (...) position both in the philosophy of probability and induction of his time and in relation to the twentieth-century frequentism. (shrink)

In section I the notions of logical and inductive probability will be discussed as well as two explicanda, viz. degree of confirmation, the base for inductive probability, and degree of evidential support, Popper's favourite explicandum. In section II it will be argued that Popper's paradox of ideal evidence is no paradox at all; however, it will also be shown that Popper's way out has its own merits.

When an empirical prediction E of hypothesis H is observed to be true, such observation is said to confirm, i.e., support (although not prove) the truth of the hypothesis. But why? What justifies the claim that such evidence supports the hypothesis? The widely accepted answer is that it is justified by induction. More specifically, it is commonly held that the following argument, (1) If H then E; (2) E; (3) Therefore, (probably) H (here referred to as ‘hypothetico-deductive confirmation argument’), is (...) inductively strong. Yet this argument looks nothing like an inductive generalization, i.e., it does not seem inductive in the term’s traditional, enumerative sense. If anything, it has the form of the fallacy of affirming the consequent. This paper aims to solve this puzzle. True, in recent decades, ‘induction’ has been sometimes used more broadly to encompass any non-deductive, i.e., ampliative, argument. Applying Bayesian confirmation theory has famously demonstrated that hypothetico-deductive confirmation is indeed inductive in this broader, ampliative sense. Nonetheless, it will be argued here that, despite appearance, hypothetico-deductive confirmation can also be recast as enumerative induction. Hence, by being enumeratively inductive, the scientific method of hypothetico-deductive confirmation is justified through this traditional, more restrictive type of induction rather than merely by ampliative induction. (shrink)

In this paper we consider a natural generalization of the Principle of Instantial Relevance and give a complete characterization of the probabilistic belief functions satisfying this principle as a family of discrete probability functions parameterized by a single real δ ∊ [0, 1).

1. Introduction. Attempts to utilize the probability calculus to prove or disprove various inductive or inductive skeptical theses are, I believe, highly problematic. Inductivism and inductive skepticism are substantive philosophical positions that do not allow of merely formal proofs or disproofs. Often the problems with particular alleged formal proofs of inductive or inductive sceptical theses turn on subtle technical considerations. In the following I highlight such considerations in pointing out the flaws of two proofs, one (...) an alleged proof of an inductive sceptical conclusion due to Karl Popper, the other an alleged proof of an inductivist thesis originally due to Harold Jeffreys and later advocated by John Earman. Surprisingly, in examining Popper's argument it is shown that certain apparently weak premises, often embraced by both inductivists and deductivists, lend themselves to inductive conclusions. However it is argued that those premises are still philosophically substantive and not amenable to a purely formal demonstration. The lesson to be learnt here is twofold. First, we need to be very careful in determining which formal theses entail, and which are entailed by, inductive skepticism and inductivism. Second, we need to take great care in laying out and examining the assumptions presumed in formal arguments directed for and against such formal theses. In a follow up article I will consider various attempts by David Stove and Karl Popper and David Miller to identify the exact content of inductive skepticism and propose a new identification based on the theory of content developed in Gemes 1994. Finally I will compare this new version with that proposed recently in Alberto Mura 1990. (shrink)

Inductive thinking is a universal human habit; we generalise from our experiences the best we can. The induction problem is to identify which observed regularities provide reasonable justification for inductive conclusions. In the natural sciences, we can often use strict laws in making successful inferences about unobserved states of affairs. In the social sciences, by contrast, we have no strict laws, only regularities which most often are conditioned on ceteris paribus clauses. This makes it much more difficult to (...) make reliable inferences in the social sciences. In particular, we want knowledge about general causal relations in order to be able to determine what to do in order to achieve a certain state of affairs. Knowledge about causal relations that are also valid in the future requires experiments or so called ‘natural experiments’. Only knowledge derived from such experiences enable us to draw reasonably reliable inferences about how to act in order to achieve our goals. (shrink)

The Problem of Old Evidence is a perennial issue for Bayesian confirmation theory. Garber (Test Sci Theor 10:99–131, 1983) famously argues that the problem can be solved by conditionalizing on the proposition that a hypothesis deductively implies the existence of the old evidence. In recent work, Hartmann and Fitelson (Philos Sci 82(4):712–717, 2015) and Sprenger (Philos Sci 82(3):383–401, 2015) aim for similar, but more general, solutions to the Problem of Old Evidence. These solutions are more general because they allow the (...) explanatory relationship between a new hypothesis and old evidence to be inductive, rather than deductive. In this paper, I argue that these solutions are either unsound or under-motivated, depending on the case of inductive explanation that we have in mind. This lends support to the broader claim that Garber–Style Bayesian confirmation cannot capture the sense in which new hypotheses that do not deductively imply old evidence nevertheless seem to be confirmed via old evidence. (shrink)

Philosophers who regard some mathematical proofs as explaining why theorems hold, and others as merely proving that they do hold, disagree sharply about the explanatory value of proofs by mathematical induction. I offer an argument that aims to resolve this conflict of intuitions without making any controversial presuppositions about what mathematical explanations would be.

I examine Du Châtelet’s methodology for physics and metaphysics through the lens of her engagement with Newton’s Rules for Reasoning in Natural Philosophy. I first show that her early manuscript writings discuss and endorse these Rules. Then, I argue that her famous published account of hypotheses continues to invoke close analogues of Rules 3 and 4, despite various developments in her position. Once relevant experimental evidence and some basic constraints are met, it is legitimate to inductively generalize from observations; general (...) hypotheses can thereafter be assumed as true until contrary experiments show otherwise. I conclude by arguing that this account of induction plays an essential role in her metaphysics, both in an argument for simple substances—which has an inductive premise—and in her attempt to distinguish acceptable and unacceptable metaphysical commitments. (shrink)

We introduce a new axiom called inductive dichotomy, a weak variant of the axiom of inductive definition, and analyze the relationships with other variants of inductive definition and with related axioms, in the general second order framework, including second order arithmetic, second order set theory and higher order arithmetic. By applying these results to the investigations on the determinacy axioms, we show the following. (i) Clopen determinacy is consistency-wise strictly weaker than open determinacy in these frameworks, except (...) second order arithmetic; this is an enhancement of Schweber–Hachtman separation of open and clopen determinacy into the consistency-wise separation. (ii) Hausdorff–Kuratowski hierarchy of differences of opens is faithfully reflected by the hierarchy of consistency strengths of corresponding parameter-free determinacies in the aforementioned frameworks; this result is valid also in second order arithmetic only except clopen determinacy. (shrink)

In this chapter we will deal with “mechanizing” induction, i.e. with ways in which theoretical computer science approaches inductive generalization. In the field of Machine Learning, algorithms for induction are developed. Depending on the form of the available data, the nature of these algorithms may be very different. Some of them combine geometric and statistical ideas, while others use classical reasoning based on logical formalism. However, we are not so much interested in the algorithms themselves, but more on (...) the philosophical and theoretical foundations they share. Thus in the first of two parts, we will examine different approaches and inductive assumptions in two particular learning settings. While many machine learning algorithms work well on a lot of tasks, the interpretation of the learned hypothesis is often difficult. Thus, while e.g. an algorithm surprisingly is able to determine the gender of the author of a given text with about 80 percent accuracy [Argamon and Shimoni, 2003], for a human it takes some extra effort to understand on the basis of which criteria the algorithm is able to do so. With that respect the advantage of approaches using logic are obvious: If the output hypothesis is a formula of predicate logic, it is easy to interpret. However, if decision trees or algorithms from the area of inductive logic programming are based purely on classical logic, they suffer from the fact that most universal statements do not hold for exceptional cases, and classical logic does not offer any convenient way of representing statements which are meant to hold in the “normal case”. Thus, in the second part we will focus on approaches for Nonmonotonic Reasoning that try to handle this problem. Both Machine Learning and Nonmonotonic Reasoning have been anticipated partially by work in philosophy of science and philosophical logic. At the same time, recent developments in theoretical computer science are expected to trig- ger further progress in philosophical theories of inference, confirmation, theory revision, learning, and the semantic and pragmatics of conditionals. We hope this survey will contribute to this kind of progress by building bridges between computational, logical, and philosophical accounts of induction. (shrink)

Carnap's Inductive Logic, like most philosophical discussions of induction, is designed for the case of independent trials. To take account of periodicities, and more generally of order, the account must be extended. From both a physical and a probabilistic point of view, the first and fundamental step is to extend Carnap's inductive logic to the case of finite Markov chains. Kuipers (1988) and Martin (1967) suggest a natural way in which this can be done. The probabilistic character of (...) Carnapian inductive logic(s) for Markov chains and their relationship to Carnap's inductive logic(s) is discussed at various levels of Bayesian analysis. (shrink)

Roy Harrod; III—The General Structure of Inductive Argument, Proceedings of the Aristotelian Society, Volume 61, Issue 1, 1 June 1961, Pages 41–56, https://doi.

Induction, conceived as the class of rational non-truth preserving inferences, has been a perennial problem in philosophy. Aside from the problem of justification of induction, a less debated issue is the problem of properly describing inductive inferences. The purpose of this book is to conceptually investigate this descriptive problem of induction from the standpoint of the nonmonotonic logical tradition raised inside the field of Artificial Intelligence in the last thirty years. As we try to show, an essential part of (...) this enterprise involves the analysis of a pragmatically oriented notion of probability, called by us inductive plausibility. This in turn is made from the viewpoint of paraconsistency ? which according to some is the unavoidable companion of nonmonotonicity ? and modal logic. We therefore put forward a nonmonotonic, paraconsistent and modal approach to induction and plausibility. The book shall be useful to philosophers, logicians and computer scientists interested in the problem of induction in general and in the logical analysis of induction and plausibility in particular. (shrink)

Inductive reasoning, initially identified with enumerative induction is nowadays commonly understood more widely as any reasoning based on only partial support that the premises give to the conclusion. This is a tad too sweeping, for this includes any inconclusive reasoning. A more moderate and perhaps more adequate characterization requires that inductive reasoning not only includes generalizations, but also any predictions or explanations obtained in absence of suitable deductive premises. Inductive logic is meant to provide guidance in choosing (...) the most supported from a given assembly of conjectures. (shrink)

Alvin plantinga and michael slote, Following ayer, Have attempted to formulate the argument from analogy for the existence of other minds as an enumerative induction. Their way of avoiding the 'generalizing from a single case' objection is shown to be fallacious.

Philosophers of mathematics commonly distinguish between explanatory and non-explanatory proofs. An important subclass of mathematical proofs are proofs by induction. Are they explanatory? This paper addresses the question, based on general principles about explanation. First, a recent argument for a negative answer is discussed and rebutted. Second, a case is made for a qualified positive take on the issue.