BAYES ON TRIAL!
Imagine trying to convince someone of Einstein's equations, Isaac Newton's second law of motion (f=ma), or of biological evolution, by using Bayes Theorem. We never do that, nor do we use it on a great many other evidence based issues. The objective evidence is convincing all on it's own.
------
A.I. on Bayes Theorem. I typed in these words: "Criticisms of bayes theorem theory"
-------------
While Bayes’ Theorem is an undisputed mathematical formula, criticisms center on its application in Bayesian statistics and epistemology. The primary debates involve subjectivity in prior beliefs, methodological ambiguity, and computational/logical limitations.
1. The Subjectivity of Prior Probabilities (The Problem of the Priors)Bayes' theorem requires defining a "prior"—an initial assumption or belief about the probability of an event. Critics, particularly classical/frequentist statisticians (e.g., Fisher and Neyman), argue that because these initial beliefs are subjective, the resulting probability is also subjective, making it scientifically unreliable. If a prior is heavily biased or uninformed, it can take an excessive amount of new data to correct the final outcome.
2. Lack of Uniqueness (The Objective Bayes Dilemma)To counter subjectivity, "Objective Bayesians" use uninformative priors (representing ignorance). However, critics argue there is no single mathematically rigorous way to define a "true" state of ignorance. Depending on the mathematical method chosen, the resulting prior can inadvertently introduce hidden assumptions that distort the analysis.
3. Falsifiability and FlexibilityIn scientific modeling, the sheer flexibility of Bayesian frameworks (with choices of priors, likelihoods, and utility functions) allows researchers to continuously adjust models until they achieve desired results. Critics argue this makes Bayesian theories susceptible to becoming unfalsifiable, or used to spin "just-so stories" to validate pre-existing conclusions rather than testing them objectively.
4. The Bayesian Trap and PseudoscienceIn applied settings, misinterpreting the odds or failing to seek alternative explanations can lead to over-diagnosis and confirmation bias. Practitioners can fall into the "Bayesian trap", where strong prior assumptions or repeated observations lead to a false sense of certainty, blinding the researcher to anomalies or the need for new experiments.
5. Logical Omnipotence and UnificationWhen applied to human reasoning or epistemology, critics point out the problem of logical omnipotence. Pure Bayesian theory assumes that a perfectly rational agent instantly assigns probabilities to all possible alternative hypotheses and knows all logical consequences of their beliefs, which is an impossible cognitive burden for actual human beings or finite computers.
6. The "Bayesian Brain" CircularityIn cognitive science, the Bayesian brain hypothesis suggests that perception and learning are based on probabilistic inferences. Critics argue this approach can be ambiguous and methodologically circular—suggesting that because a system "looks" rational, it must be using Bayesian calculations, despite a lack of neurological evidence for such complex mathematics at the process level.
------
A.I. on Bayes Theorem. I typed in these words: "Criticisms of bayes theorem theory"
-------------
While Bayes’ Theorem is an undisputed mathematical formula, criticisms center on its application in Bayesian statistics and epistemology. The primary debates involve subjectivity in prior beliefs, methodological ambiguity, and computational/logical limitations.
1. The Subjectivity of Prior Probabilities (The Problem of the Priors)Bayes' theorem requires defining a "prior"—an initial assumption or belief about the probability of an event. Critics, particularly classical/frequentist statisticians (e.g., Fisher and Neyman), argue that because these initial beliefs are subjective, the resulting probability is also subjective, making it scientifically unreliable. If a prior is heavily biased or uninformed, it can take an excessive amount of new data to correct the final outcome.
2. Lack of Uniqueness (The Objective Bayes Dilemma)To counter subjectivity, "Objective Bayesians" use uninformative priors (representing ignorance). However, critics argue there is no single mathematically rigorous way to define a "true" state of ignorance. Depending on the mathematical method chosen, the resulting prior can inadvertently introduce hidden assumptions that distort the analysis.
3. Falsifiability and FlexibilityIn scientific modeling, the sheer flexibility of Bayesian frameworks (with choices of priors, likelihoods, and utility functions) allows researchers to continuously adjust models until they achieve desired results. Critics argue this makes Bayesian theories susceptible to becoming unfalsifiable, or used to spin "just-so stories" to validate pre-existing conclusions rather than testing them objectively.
4. The Bayesian Trap and PseudoscienceIn applied settings, misinterpreting the odds or failing to seek alternative explanations can lead to over-diagnosis and confirmation bias. Practitioners can fall into the "Bayesian trap", where strong prior assumptions or repeated observations lead to a false sense of certainty, blinding the researcher to anomalies or the need for new experiments.
5. Logical Omnipotence and UnificationWhen applied to human reasoning or epistemology, critics point out the problem of logical omnipotence. Pure Bayesian theory assumes that a perfectly rational agent instantly assigns probabilities to all possible alternative hypotheses and knows all logical consequences of their beliefs, which is an impossible cognitive burden for actual human beings or finite computers.
6. The "Bayesian Brain" CircularityIn cognitive science, the Bayesian brain hypothesis suggests that perception and learning are based on probabilistic inferences. Critics argue this approach can be ambiguous and methodologically circular—suggesting that because a system "looks" rational, it must be using Bayesian calculations, despite a lack of neurological evidence for such complex mathematics at the process level.
No comments:
Post a Comment