Bayes' rule
A principle of probability theory describing how to update one's degree of belief in a hypothesis given new evidence.
As defined in 343 | Tom Griffiths on The Laws of Thought
Where it comes up · 6
- Roland Fryer on Race, Diversity, and Affirmative Action2 claims
- 143 | Julia Galef on Openness, Bias, and Rationality2 claims
- 343 | Tom Griffiths on The Laws of Thought2 claims
- The Search for the Truth with the Anonymous Master1 claim
- AMA | February 20211 claim
- AMA | September 20211 claim
Most important claims · 9 of 9Search all claims for “Bayes' rule” →
- The radical innovation of Bayes was applying probability theory—previously a mathematics of gambling and dice (Cardano, Pascal)—to thought itself, recognizing that the same axiomatic system that characterizes dice also characterizes what goes on in our heads when we change our beliefs.343 | Tom Griffiths on The Laws of Thought
- The laws of thought are not codified like the laws of thermodynamics; rather, we have mathematical systems (with candidate law-like principles such as Bayes' rule for inductive inference and modus ponens for valid deduction) that characterize what thinking should look like in different circumstances.343 | Tom Griffiths on The Laws of Thought
- In experiments on climate change, affirmative action, and the death penalty, people screened to hold opposing beliefs who were given identical information left more polarized—each side interpreting the information as supporting their prior—contradicting the classic game-theory expectation that shared information makes beliefs converge.Roland Fryer on Race, Diversity, and Affirmative Action
- Being Bayesian decomposes into separable parts: Bayes' rule itself is a theorem inherent in conditional probability and is non-optional, the practice of updating can be a bounded-rationality approximation that one should still try to approximate as well as possible, and the subjective view of probability (credences updated by new information) is a philosophical commitment Carroll endorses against those who reject it.AMA | February 2021
- Rather than Bayesianism being dressed-up induction, Popperian falsificationism is dressed-down Bayesianism: Bayes' rule is a mathematical theorem, and Bayesian reasoning assigns prior credences to propositions and updates them with data so credences never reach zero—solving Hume's problem of induction without claiming metaphysical certainty—whereas falsificationism is just a crude approximation that pretends to assign no credences when in reality we always implicitly do, favoring simpler theories with higher priors.AMA | September 2021
- Two people can observe the same publicly observable signal and interpret it completely differently because they update through their prior beliefs first; Fryer's double-updating model holds that unclear signals are first resolved into a clear signal using priors and then used for Bayesian updating, which can drive belief divergence rather than convergence.Roland Fryer on Race, Diversity, and Affirmative Action
- Statistical extrapolation works by comparing two possibilities, that the pattern came from chance or that it came from a real bias or causal signal, and the extremely unlikely pattern under the chance hypothesis is what justifies inferring a signal, as with ten thousand heads in a row on a supposedly fair coin.The Search for the Truth with the Anonymous Master
- If evidence X would have increased your belief in a hypothesis, then observing the opposite of X must decrease it; people commonly neglect this by 'accounting for' contrary evidence without recognizing that the opposite evidence would have strengthened their belief.143 | Julia Galef on Openness, Bias, and Rationality
- Bayes' rule states your confidence in a hypothesis should rise in proportion to the ratio of how likely the observed evidence would be if the hypothesis were true versus if it were false; even without precise numbers, being mindful of this ratio improves reasoning over default human intuition.143 | Julia Galef on Openness, Bias, and Rationality