Abstract
For a long time logicians have had various doubts concerning the justification of inductive inference. Is inductive inference justifiable, and when? Is inductive inference a correct one? Do the premises of inductive inference justify the acceptance of a conclusion, and when? Such are the questions which logicians have to answer, concerning the justification of induction.1
First published in Studia Logica V (1957). Translated by S. Wojnicki.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Preview
Unable to display preview. Download preview PDF.
References
This paper is a somewhat modified fragment of the doctoral dissertation prepared by the author.
Cf. M. Black âOn the Justification of Inductionâ, Language and Philosophy, pp. 63â64.
The terms âinductionâ resp. âinductive inferenceâ are used in so many different meanings that I think it an unfeasible task to find their analytical definition. In contemporary literature of logic the tendency may be observed to such understanding of these terms in which their range is very wide and embraces all the fallible inference methods which are usually approved in the empirical sciences. It is in this, very imprecise sense that I use the term âinductionâ resp. âinductive inferenceâ in the present paper.
M. Black, op. cit., pp. 66â68.
It cannot be generally established what probability should be called high probability. There is always some convention in choosing the lower limit of high probabilities, as for example in adopting the level of significance in statistical tests. Undoubtedly our inclination to consider a given probability high depends in practice to a large extent to the loss connected with the acceptance of a false conclusion.
J. M. Keynes, A Treatise on Probability, 1929.
J. Hosiasson, âO prawomocnosci indukcji hipotetycznejâ (âThe Problem of Justification of Hypothetical Inductionâ), Fragmenty Filozoficzne, Warszawa 1934.
G. H. von Wright, The Logical Problem of Induction, 1941.
R. Carnap, Logical Foundations of Probability, 1951.
Constructing such theories is nowadays fashionable among western logicians (cf. e.g. Carnapâs Logical Foundations of Probability and the lively discussion following it which has been going on in logical and philosophical periodicals). Instead of the term âlogical probabilityâ the term âdegree of confirmationâ is used, or shortly âconfirmationâ. I think construction of such theories is unnecessary for the solution of the problem of justification of induction for the following reasons: the existing axioms for the concept of confirmation usually repeat the axioms given for the mathematical concept of probability. (Cf. the axioms given by Keynes and Hosiasson in the works mentioned or the axioms given by A. Shi-mony in his paper âCoherence and the Axioms of Confirmationâ, Journal of Symbolic Logic 20.) On the other hand, attempts to give an explicit definition of the degree of confirmation usually reduce it to classical probability, defined not on events but on sentences (cf. Carnap, op cit.). I think this difference is not substantial. In my opinion it is obvious that classical probability may equally well be considered a function where arguments are events or sentences (cf. here i.a. K. R. Popper, Logik der Forschung, p. 188 etc., W. Glivenko, Kurs teorii veroyatnosti (Calculus of probability), Moscow â Leningrad 1939, pp. 231-231; R. Carnap, op. cit., pp. 29-30). Hence it seems that construction of special theories of confirmation as theories of classical probability defined on sentences is unnecessary. The confirmation theories which I know do not bring anything new, neither from the formal, nor from the substantial points of view, in comparison with mathematical theories of probability, and they only repeat what is already known from the mathematical theories. The construction of a special theory of logical probability becomes reasonable only when one holds the view that logical and mathematical probability differ not only in that the former is defined on sentences and the latter on events, but when there is a difference which makes impossible the adoption of the axioms of mathematical probability as ready made axioms for logical probability. This is, for instance, Popperâs opinion: according to him, logical probability (Hypothesenwahrscheinlichkeit), understood as the degree of justification of a hypothesis by facts, has some properties totally different from mathematical probability (Ereigniswahrscheinlichkeit), even if the latter were to be interpreted as function propositional arguments (cf. Logik der Forschung, pp. 188 ff, and POP-perâs articles in British Journal for the Philosophy of Science 5, 6). There is no place here to deal further with this interesting question.
Op. cit., pp. 235-236.
This condition can be replaced with the following one: there is an Δ>0 such that P(vn, Kn-1 · ឥ) ⫠1-Δ. Keynes proved his theorems on the grounds of his own axiomatic theory of probability. They can however be as easily demonstrated on the grounds of the mathematical theory of probability.
Op. cit., pp. 251 f.
For instance, J. Nicod, Le problĂšme logique de lâinduction, pp. 76 ff.
If they are not finite, from the practical point of view it is important that the inductive generalization would hold for the subset of the examined set the elements of which we encounter in practical activities and this subset is always finite.
Cf. i.a. B. Russel, Human Knowledge, G. Allen, London 1948, p. 424.
As P(Vn, g) = l.
The assumption of finiteness of the set A does not exclude the possibility of an unlimited increase of the number of verifications, if only it is assumed that the examined elements of the set A are â statistically speaking â sampled with replacement. If we assumed that we are dealing with sampling without replacement, we would have to substitute the assumption that n tends to infinity by the assumption that n tends to N. In both cases the conclusions concerning the probabilistic justification of enumerative induction will be the same.
This measure of the degree of confirmation of hypothesis h by empirical data e corresponds to what Carnap calls ârelevance quotientâ (Logical Foundations of Probability, p. 567). We might just as well take as measure of the degree of confirmation of h by e any function strictly increasing with m(h, e). This has been done by I. J. Good, who takes as measure of the degree of confirmation of h by e (Good uses the term âweight of evidence for h, given eâ) the value of log P (h, e) + log P (e) + const. Cf. Probability and the Weighing of Evidence, p. 63.
Professor M. KokoszyĆska, in her contribution presented at the same conference and published in the same volume of Studia Logica, introduces among other things the non-relativized and the relativized concept of inductive fallacy. It is easy to see that my expression âe probabilistically confirms hâ has the same meaning as prof. KokoszyĆskaâs expression âh is correctly inferred from e (in the absolute sense)â, as well as the sentence âe probabilistically confirms h1 more powerfully than h2â means the same as prof. KokoszyĆskaâs: âthe inference of h1from e is less fallacious (more correct) than the inference of h2 from eâ.
Statisticians call such statistical tests parametric tests.
Cf. the definition of statistical test given by J. Neyman in First Course in Prob ability and Statistics, p. 258. The description of the procedure of verifying statistical parametric hypotheses has been taken from M. Fiszâs Calculus of Probability and Mathematical Statistics, in particular Chap. 13.
I assume here that elements of the set A examined with respect to the property B are sampled with replacement. The whole reasoning might be based on the assumption of sampling without replacements and lead to a similar result.
Cf. Baconâs well known opinions: âInductio enim quae procedit per enumeration nem simplicem res peurilis est; precario concluda et periculo exponitur ab instantia contradictoriaâ⊠And: âInductio mala est quae per enumerationem simplicem Principia concludit ScientiarumâŠâ(Novum Organum, Amsterdam 1964, I, 105, and I, 69). Among the contemporary logicians, Professor Kotarbinski says of enumerative induction: âA non-exhaustive induction, which proceeds by simple enumeration⊠is a very weak method of justifying generalizationâ, Kurs logiki (A Course in Logic), 2nd ed., p. 135.
I owe this information to Professor S. Romahnowa.
Editor information
Rights and permissions
Copyright information
© 1977 PWN - Polish Scientific Publishers - Warszawa
About this chapter
Cite this chapter
CzerwiĆski, Z. (1977). The Problem of Probabilistic Justification of Enumerattve Induction. In: PrzeĆÄcki, M., WĂłjcicki, R. (eds) Twenty-Five Years of Logical Methodology in Poland. Synthese Library, vol 87. Springer, Dordrecht. https://doi.org/10.1007/978-94-010-1126-6_5
Download citation
DOI: https://doi.org/10.1007/978-94-010-1126-6_5
Publisher Name: Springer, Dordrecht
Print ISBN: 978-94-010-1128-0
Online ISBN: 978-94-010-1126-6
eBook Packages: Springer Book Archive