Objective Bayesianism
My account of objective Bayesianism is epistemological: a theory of rational degree of belief, rather than a theory of statistical inference.
The account is distinctive in that:
- It rejects the usual Bayesian identification of conditional degree of belief with conditional probability.
- It doesn’t presuppose that degrees of belief should be updated by Bayesian conditionalisation.
- It rejects a common objective Bayesian assumption that evidence uniquely determines a rational belief function.
- It takes objective chances to play a central role in determining rational degrees of belief.
The account is built on three kinds of norm:
- Structural. An agent’s belief function should be a probability function.
- Evidential. If the agent establishes from evidence that the chance function is in some set of probability functions, then her belief function should be in the convex hull of that set.
- Equivocation. The agent’s degrees of belief should otherwise be equivocal, adopting committal degrees of belief (near 0 or 1) only where they are forced by structural or evidential norms.
*
Motivation
For some recent arguments for this sort of approach, see:
Jon Williamson: Bayesianism from a philosophical perspective and its application to medicine, International Journal of Biostatistics 19(2): 295-307, 2023. doi: 10.1515/ijb-2022-0043
Jon Williamson: A Bayesian account of establishing, British Journal for the Philosophy of Science 73(4):903-925, 2022. . doi: 10.1086/714798
Jon Williamson: Direct inference and probabilistic accounts of induction, Journal for General Philosophy of Science 54:451–472, 2023. . doi: 10.1007/s10838-021-09584-0
For an introduction to the approach, see:
Jon Williamson: In defence of objective Bayesianism, Oxford University Press, 2010.
*
Objective Bayesian inductive logic
I’m also interested in the use of objective Bayesianism to provide a new approach to inductive logic. I’m currently collaborating with Juergen Landes and Soroush Rafiee Rad on this.
Recent work includes:
Jon Williamson: Where do we stand on maximal entropy? In Logic for data, eds Hykel Hosni & Juergen Landes, Springer, 2024.
Juergen Landes, Soroush Rafiee Rad and Jon Williamson: Determining maximal entropy functions for objective Bayesian inductive logic, Journal of Philosophical Logic 52:555-608, 2023. doi: 10.1007/s10992-022-09680-6
Juergen Landes, Soroush Rafiee Rad and Jon Williamson: Towards the Entropy-Limit Conjecture, Annals of Pure and Applied Logic 172(2):102870, 2021. . doi: 10.1016/j.apal.2020.102870
For an introduction to the approach, see:
Jon Williamson: Lectures on inductive logic, Oxford University Press, 2017. Errata.
*
Objective Bayesian nets
Graphical models can be used to represent and reason with objective Bayesian probabilities.
Recent work includes:
Juergen Landes and Jon Williamson: Objective Bayesian nets for integrating consistent datasets, Journal of Artificial Intelligence Research 74: 393-458, 2022. . doi 10.1613/jair.1.13363
For an introduction, see:
Rolf Haenni, Jan-Willem Romeijn, Gregory Wheeler & Jon Williamson: Probabilistic logics and probabilistic networks, Synthese Library, Springer, 2011.
Jon Williamson: Bayesian nets and causality: philosophical and computational foundations, Oxford University Press, 2005.