Bayesian Epistemology
Belief modeled as probability and updated by Bayes' rule on new evidence.
Bayesian Epistemology is the philosophical position that rational degrees of belief (credences) should be modeled as probabilities satisfying the axioms of probability theory, and that rational belief revision in light of new evidence should follow Bayes's theorem (updating prior probabilities to posterior probabilities by conditioning on new evidence). The framework draws on Frank Ramsey's 1926 'Truth and Probability' and Bruno de Finetti's 1937 'La prevision' for foundational Dutch Book arguments showing that an agent whose credences violate probability axioms can be exploited through bets. Subsequent contributors include Rudolf Carnap (logical probability), Richard Jeffrey (probability kinematics for uncertain evidence), Patrick Maher, and contemporary figures including Branden Fitelson, James Joyce, and Kevin Kelly. Bayesian epistemology offers a unified account of confirmation, evidential support, learning, and inference, with applications to philosophy of science (Bayesian confirmation theory), formal epistemology, decision theory under uncertainty, and the philosophy of induction. The framework is distinct from Bayesian Inference as practiced in statistics and machine learning (where it is formal-scientific) — the philosophical position is that belief itself should be probabilistically modeled, not just that probabilistic methods are useful tools. The framework has substantial subjective and objective variants, with the subjective Bayesian permitting wide latitude in priors and the objective Bayesian seeking principled constraints on prior selection.
Core components
- Credences (degrees of belief) modeled as probabilities
- Probability axioms (non-negativity, normalization, additivity)
- Bayes' theorem for belief updating
- Dutch Book argument (probabilistic incoherence enables exploitation)
- Conditionalization as rational update rule
- Jeffrey conditionalization for uncertain evidence
- Subjective vs objective Bayesianism
- Connection to decision theory under uncertainty
- Distinction from frequentist statistics
Primary use case
Formal epistemology and philosophy of science; philosophical foundations for statistical methodology; theory of confirmation and evidence; decision theory under uncertainty; foundation for some contemporary discussions of rationality; growing influence in cognitive science and computational rationality research; influence on rationalist communities (LessWrong, effective altruism) for everyday reasoning.
Common criticisms
- Problem of priors: in subjective Bayesianism, priors can be arbitrary, and yet evidence interpretation depends on them
- objective Bayesianism's attempts to constrain priors (maximum entropy, indifference principles) have not produced consensus
- old evidence problem: Bayesian conditionalization works poorly for evidence already known to the agent
- logical omniscience problem: standard Bayesianism assumes agents have perfect logical and mathematical knowledge, which no real agent does
- 'lottery paradox' and similar puzzles for the relationship between credences and binary belief
- integration with empirical research on actual human reasoning is mixed — humans systematically deviate from Bayesian updating in characterizable ways
- the framework's mathematical sophistication can obscure substantive philosophical commitments
- competing frameworks including Dempster-Shafer theory and imprecise probabilities argue Bayesianism is too restrictive.
Lineage
- Siblings
- Falsificationism, Reliabilism