Evidence and Objective Bayesian Epistemology

• Applicant: Dr. Jürgen Landes
• Subject Area: Theoretical Philosophy
• Term: from 2018 to 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 405961989

• Final Report Year: 2022

Final Report Abstract

Subjective Bayesian epistemology is currently widely perceived to be the paradigm of evidential reasoning in the philosophy of science and of rational degrees of belief underpinning the dominant account of decision making, as well as many of our statistical methods. Objective Bayesian epistemology is emerging as an alternative epistemology challenging the subjective account. Objective Bayesians hold that rational degrees of belief are determined by the agent’s evidence while subjective Bayesians claim that there is a range of rational degrees of belief given an agent’s evidence. This project aimed to strengthen the case for the objective approach. Evidential reasoning in subjective Bayesian epistemology is based on two long-standing interconnected parts: it is grounded in expected utility maximisation which is implemented by updating beliefs via conditionalisation. Evidential reasoning in objective Bayesian epistemology is based on less solid foundations and is also lacking a comprehensive framework for an implementation of evidential reasoning. Without these foundations and such an implementation objective Bayesian epistemology cannot aspire to stand up to the subjective approach. This project hence developed foundations and methodology for evidential reasoning in objective Bayesian epistemology. I have argued that the provided foundations and the developed implementation are superior to those of the subjective approach in key aspects and thus strengthened the case for the objective approach. Foundationally, I have argued that the objective approach does a better job of doing justice to two widely-accepted epistemological rationales: the Principal Principle and the Principle of Total Evidence. I have developed methodology for evidential reasoning for i) data integration problems, in which studies collected data on overlapping variable sets and a probability distribution is sought on the entire domain of interest and ii) infinite domains of interest represented by a first order predicate language. This research informed a core debate within formal epistemology, namely the question of whether the subjective or the objective approach better characterises the norms governing belief. The outcome of this debate is highly relevant to rational decision making in policy-making, science and daily life since the two approaches single out different actions as rational, in general. The outcome is furthermore relevant to statistics and artificial intelligence, where Bayesian principles are widely used and where there is little agreement as to whether subjective or objective Bayesian principles are best. I have furthermore investigated and developed tools for evidence aggregation in Bayesian epistemology. These developments can be applied within the subjective and the objective Bayesian approach. Bayesian networks, a graphical tool for modelling probability distributions, have played a key role in these works.

Publications

• A softmax algorithm for evidence appraisal aggregation. PLoS ONE, 16(6):1–23, 2021
  F. De Pretis and J. Landes
  
• A Triple Uniqueness of the Maximum Entropy Approach. In J. Vejnarová and N. Wilson, editors, Proceedings of ECSQARU, volume 12897 of LNAI, pages 644-656, Cham, 2021. Springer
  J. Landes
  
• Artificial Intelligence Methods For a Bayesian Epistemology-Powered Evidence Evaluation. Journal of Evaluation in Clinical Practice, 27(3):504–512, 2021
  F. De Pretis, J. Landes, and W. J. Peden
  
• Bayesian Epistemology. In Oxford Bibliographies in Philosophy. Oxford University Press, New York, 2021
  J. Landes
  
• The Entropy-limit (Conjecture) for Σ2 -Premisses. Studia Logica, 109: 423–442, 2021
  J. Landes
  
• The Principal Principle, admissibility, and normal informal standards of what is reasonable. European Journal for Philosophy of Science, 11, 2021
  J. Landes, C. Wallmann, and J. Williamson
  
• The variety of evidence thesis and its independence of degrees of independence. Synthese, 198:10611–10641, 2021
  J. Landes
  
• Towards the Entropy-Limit Conjecture. Annals of Pure and Applied Logic, 172, 2021
  J. Landes, S. Rafiee Rad, and J. Williamson
  
• Special Issue: Bayesian Epistemology: Current Perspectives and Challenges. Kriterion, 2022
  J. Landes