Key Arguments
Central Problem: How can credence obey probability laws when there’s no underlying objective probability (especially for logical/mathematical statements)?
Solution: Logical Induction
Introduced by Garrabrant, Benson-Tilsen, Critch, Soares, Taylor (2016).
What It Does:
- Rigorously formalizes epistemic uncertainty about logical/mathematical statements
- Assigns probabilistic credences to logical propositions
- Ensures rational coherence and consistency
- Generates sequence of credences that become more accurate as evidence accumulates
- Crucially: Satisfies probability axioms despite no empirical probability
How It Works (Market Analogy):
- Logical uncertainty = “market”
- “Traders” = algorithmic strategies/computational heuristics
- Traders “bet” on logical statements (true/false)
- As new information emerges (proofs, computations, heuristics), traders update
- Market credences gradually converge toward accurate beliefs
- Analogy ensures internal coherence, rational updating
Resolution of Philosophical Tension:
- Credences must obey probability laws (rational consistency)
- But logical credences are epistemic, not empirically objective
- Logical Induction shows how probabilistic consistency maintained without objective probabilities
Result: Rigorous embrace of epistemic probabilities—can assign rational, probabilistic credences even without empirical probabilities
Connection to Framework
Defending Bayes Sequence:
- Part 5 (Batch 2): Empirical vs logical credence distinction
- Part 6 (this post): Formal tool for logical credences
- Part 7 (this batch): Application to theories
Epistemological Foundations:
- Distinguishes Measure (objective QBU probability) from Credence (subjective epistemic)
- Logical Induction provides formal basis for credences in non-empirical domains
- Resolves how credences can be rational without corresponding to physical probabilities
AI Safety Connection:
- Paper from MIRI (Machine Intelligence Research Institute)
- Relevant to AI reasoning about logical uncertainty
- Connects framework to AI alignment community
Evolution Tracking
New Technical Tool:
- First appearance of Logical Induction in archive
- Provides formal underpinning for credence theory
- Borrowed from AI safety research, applied to philosophy
Methodological: Uses technical result from computer science to resolve philosophical problem
Integration: Logical Induction becomes load-bearing concept for defending Bayesianism against Deutsch/Hall critique
Cross-References
- Backward: Defending Bayes, Part 5
- Forward: Defending Bayes, Part 7
Notes
Significance: Imports technical result from AI research to solve philosophical problem. Cross-disciplinary synthesis.
Accessibility: Uses market analogy to explain complex formal concept. Generally readable despite technical content.
AI Safety Bridge: Connects Axio framework to AI alignment community’s work. Suggests shared intellectual ecosystem.
Formal Rigor: Cites actual research paper, engages with technical literature (not just philosophy).