Abstract
This note employs the recently established consistency theorem for infinite regresses of probabilistic justification (Herzberg in Stud Log 94(3):331–345, 2010) to address some of the better-known objections to epistemological infinitism. In addition, another proof for that consistency theorem is given; the new derivation no longer employs nonstandard analysis, but utilises the Daniell–Kolmogorov theorem.
Similar content being viewed by others
Notes
An event U is called contingent given a probability measure P if and only if 0 < P(U) < 1.
The symmetric difference of two events A, B is defined as \(A\Updelta B:=(A\setminus B)\cup (B\setminus A). \)
We have nevertheless decided to stick with the term ‘justification’ for much of this paper simply because it seems to be the more common terminological choice in the literature.
In fact, the property of being a finite ordinal, i.e. a natural number, can even be defined in set-theoretically “robust” terms: It is definable through a \(\Updelta_0\) and thus through an absolute formula. In other words, whether a given set is a natural number does not depend on the interpretation in a particular transitive model of set theory (cf. Jech 2000, Lemma 12.10).
It should be mentioned that the Daniell–Kolmogorov theorem admits a particularly simple proof by means of nonstandard analysis, cf. Herzberg (2011).
To see this, apply the Transfer Principle to the sentence \({\forall n\in{\bf N}}\left( S(n+1)\Rightarrow \forall k\leq n\left(S(k)\right)\right), \) which in turn follows from the hypothesis \({\forall n\in{\bf N}}\left(S(n+1)\Rightarrow S(n)\right)\) by mathematical induction.
In the special case where \(\varepsilon=\frac{1}{2}, \) we have \(Q^{j,\infty}_{i}=\frac{1}{2}\) for all \(j\in{\bf N}\) and each \(i\in\{0,1\}. \)
References
Atkinson, D., & Peijnenburg, J. (2010). The solvability of probabilistic regresses A reply to Frederik Herzberg. Studia Logica, 94(3), 347–353.
Bauer, H. (2002). Wahrscheinlichkeitstheorie. 5., durchgesehene und verbesserte Auflage. de Gruyter: Berlin.
Fumerton, R. (1998). Replies to my three critics. Philosophy and Phenomenological Research, 58, 927–937.
Fumerton, R. (2001). Epistemic justification and normativity. In Steup, M., (eds), Knowledge, truth, and duty: Essays on epistemic justification, responsibility and virtue, pp. 49–61, Oxford University Press: New York, NY.
Gillett, C. (2003). Infinitism redux? A response to Klein. Philosophy and Phenomenological Research, 66, 709–717.
Herzberg, F. (2008a). A definable nonstandard enlargement. Mathematical Logic Quarterly, 54(2), 167–175.
Herzberg, F. (2008b). Addendum to “A definable nonstandard enlargement”. Mathematical Logic Quarterly, 54(6), 666–667.
Herzberg, F. (2010). The consistency of probabilistic regresses. A reply to Jeanne Peijnenburg and David Atkinson. Studia Logica, 94(3), 331–345.
Herzberg, F. (2011). Simple nonstandard proofs of Daniell–Kolmogorov-type theorems. Positivity. doi:10.1007/s11117-011-0135-7.
Jech, T. (2000). Set theory. The third millennium edition. Springer Monographs in Mathematics. Berlin: Springer.
Kanovei, V. and Reeken, M. (2004). Nonstandard analysis, axiomatically. Springer Monographs in Mathematics. Berlin: Springer.
Kanovei, V., & Shelah, S. (2004). A definable nonstandard model of the reals. Journal of Symbolic Logic, 69(1), 159–164.
Klein, P. (1998). Foundationalism and the infinite regress of reasons. Philosophy and Phenomenological Research, 58, 919–925.
Klein, P. (1999). Human knowledge and the infinite regress of reasons. Philosophical Perspectives, pp. 297–325.
Klein, P. (2005). Infinitism is the solution to the regress problem. In Steup, M. and Ernest, S., (eds), Contemporary debates in epistemology, pp 131–140, Malden, M.A. Blackwell.
Klein, P. (2007a). How to be an infinitist about doxastic justification. Philosophical Studies, 134, 25–29.
Klein, P. (2007b). Human knowledge and the infinite progress of reasoning. Philosophical Studies, 134, 1–17.
Loeb, P. (1975). Conversion from nonstandard to standard measure spaces and applications in probability theory. Transactions of the American Mathematical Society, 211, 113–122.
Nelson, E. (1977). Internal set theory: A new approach to nonstandard analysis. Bulletin of the American Mathematical Society, 83(6), 1165–1198.
Peijnenburg, J. (2007). Infinitism regained. Mind, 116(463), 597–602.
Peijnenburg, J. (2010). Ineffectual foundations: Reply to Gwiazda. Mind, 119(476), 1125–1133.
Peijnenburg, J., & Atkinson, D. (2008). Probabilistic justification and the regress problem. Studia Logica, 89(3), 333–341.
Peijnenburg, J., & Atkinson, D. (2011). Grounds and limits: Reichenbach and foundationalist epistemology. Synthese, 181(1), 113–124.
Podlaskowski, A., & Smith, J. (2011). Infinitism and epistemic normativity. Synthese, 178, 515–527.
Schuster, P., Berger, U., & Osswald, H., (eds) (2001). Reuniting the antipodes—constructive and nonstandard views of the continuum, volume 306 of Synthese Library. Dordrecht: Kluwer Academic Publishers.
Turri, J. (2009). On the regress argument for infinitism. Synthese, 166, 157–163.
Author information
Authors and Affiliations
Corresponding author
Appendix: A “Standard” Proof of the Consistency Theorem for Probabilistic Regresses
Appendix: A “Standard” Proof of the Consistency Theorem for Probabilistic Regresses
For all of this section, we fix two sequences \(\langle\alpha_k\rangle_{k\in{\bf N}},\langle\beta_k\rangle_{k\in{\bf N}}\in[0,1]^{\bf N};\) as in all of this paper, N denotes the set of all nonnegative integers. In this section, we shall prove the following result, which is a slight generalisation of the consistency theorem in Herzberg (2010):
Theorem A.1
There exists some probability measure P and some sequence of P-measurable events \(\langle S_k\rangle_{k\in{\bf N}}\) such that \(P\left(\left.S_k\right|S_{k+1}\right)=\alpha_k\) as well as \(P\left(\left.S_k\right|\complement S_{k+1}\right)=\beta_k\) for all \(k\in{\bf N}\).
Essentially, the proof is based on the construction of a reversed non-homogeneous Markov chain with state space {0,1} that was started an infinitely long time ago. We use the notation \(\prod\nolimits_{\leftarrow}\) to refer to products of matrices whose order of multiplication has been reversed, so that \({\prod}_{\mathop{\leftarrow}\limits_{k=m}}^{n}T^k=T^n\cdots T^m\) for all \(n\geq m\in{\bf N}. \) Likewise, \({\prod}_{\mathop{\leftarrow}\limits_{k=m}}^{\infty} T^k=\lim\nolimits_{n\rightarrow\infty} T^n\cdots T^m. \)
For all \(k\in{\bf N}, \) put
For all \(j,j^{\prime}\in {\bf N}\) with j′ ≥ j and for all \(i,i'\in\{0,1\}, \) let
wherein \(e_0=\left(\begin{array}{c}0\\ 1\end{array}\right)\) and \(e_1=\left(\begin{array}{c}1\\ 0\end{array}\right), \) and let
for some \(c=\left(\begin{array}{c}1-\varepsilon\\ \varepsilon\end{array}\right)\) with \(\varepsilon\in (0,1). \) Footnote 7 Define for all finite \(J\subseteq {\bf N}\) with \(\{j_1,\ldots,j_{\# J}\}, \) a measure P J on {0,1}J by means of the equation
for all \(\vec i\in \{0,1\}^J. \)
Lemma A.2
P J is a probability measure on {0,1}J for all finite \(J\subseteq {\bf N}. \)
Proof
Note that all the T k are transition probability matrices, that the set of transition probability matrices is closed under multiplication and that the set of transition probability matrices is a closed subset of the set of matrices of that dimension. Therefore, \({\prod_{\mathop{\leftarrow} \limits_{k}}}T^k\) is a transition probability matrix. By definition, this means that each entry is nonnegative and the sum of each row vector equals 1. Formally,
thus \({^\top \!\! e_{i}}\left({\prod_{\mathop{\leftarrow} \limits_{k}}T^k}\right)\left(e_0+e_1\right)=1\) for each \(i\in\{0,1\}. \) This immediately entails that
for all \(j,j'\in{\bf N}\) and each \(i'\in\{0,1\}. \) However, it also means that
which implies that
for all \(j\in{\bf N}. \) So,
for all \(j,j'\in{\bf N}\) and each \(i'\in{\bf N}. \) This can be used to prove inductively in #J that \(\sum_{\vec i\in\{0,1\}^J}P_J\{\vec i\}=1. \) Since clearly \(P_J\{\vec i\}\geq 0\) for all \( \vec i\in\{0,1\}^J, \) we conclude that P J must be a probability measure on {0,1}J. \(\square\)
Let \(\mathcal{H}\) be the set of all finite subsets of N.
Lemma A.3
\(\left\langle P_J\right\rangle_{J\in\mathcal{H}}\) is a projective family of probability measures.
Proof
Let H = J ∪ {h} with \(h\not\in J. \) Without loss of generality, let h > max J = j #J . Let p H J be the projection of H onto J. Then
Now, it is not difficult to verify that
Indeed, \(e_0 {^\top \!\! e_0} + e_1 {^\top \!\! e_1}= \left(\begin{array}{cc}1 & 0\\ 0 & 1 \end{array}\right), \) hence
Inserting Equations (5) into (4) yields
whence by definition \(P_H\left({p_J^H}^{-1}\{\vec i\}\right)= P_J\{\vec i\}\) for arbitrary \(\vec i\in \{0,1\}^J\) and all finite \(H,J\subseteq {\bf N}\) with \(J\subseteq H\) and \(\#\left(H\setminus J\right)=1. \) This implies (cf. Bauer (2002, p. 307, Bemerkung 2)) that \(\left\langle P_J\right\rangle_{J\in\mathcal{H}}\) is a projective family of probability measures.\(\square\)
In the following theorem, \(\mathcal{P}(\{0,1\})^{\otimes {\bf N}}\) denotes the infinite product of countably many identical copies of the power-set of {0,1}; it is the smallest σ-algebra with respect to which all projections are measurable. Given any \(J\in\mathcal{H}, \) the map π J is the projection onto {0,1}J defined as the map \(\pi_J:\{0,1\}^{\bf N}\rightarrow\{0,1\}^J,\quad \omega\mapsto \langle\omega(k)\rangle_{k\in J}\)
Theorem A.4
There exists a probability measure \(P:\mathcal{P}(\{0,1\})^{\otimes {\bf N}}\rightarrow[0,1]\) such that \(P_J=P\left(\pi_J^{-1}(\cdot)\right)\) for all \(J\in \mathcal{H}\).
Proof
By the Daniell-Kolmogorov theorem, there must be a probability measure P on \(\Upomega=\{0,1\}^{\bf N}\) such that the finite-dimensional distributions of the process X defined by \(X_k:\Upomega\rightarrow\{0,1\},\quad\omega\mapsto \omega(k)\) (for all \(k\in{\bf N}\)) coincide with the probability measures P J , \(J\in\mathcal{H}. \) \(\square\)
Now we can, at last, prove the consistency theorem stated at the beginning as a corollary to Theorem A.4:
Proof of Theorem A.1
Let P and X be as in the proof of Theorem A.4. For all \(k\in{\bf N}, \) set \(S_k=\left\{X_k=1\right\}, \) so that \(\complement S_k=\left\{X_k=0\right\}, \)
Then, for all \(n>k\in{\bf N}, \)
hence
as well as
whence
Comparing this with the definition of T k in Equation (3) leads us to conclude
for all \(k\in{\bf N}. \square \)
Rights and permissions
About this article
Cite this article
Herzberg, F. The Consistency of Probabilistic Regresses: Some Implications for Epistemological Infinitism. Erkenn 78, 371–382 (2013). https://doi.org/10.1007/s10670-011-9358-z
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10670-011-9358-z