likelihood function
In Bayesian inference, the probability of obtaining the observed data if a given hypothesis were true.
As defined in 200 | Solo: The Philosophy of the Multiverse
Where it comes up · 6
- AMA | August 20215 claims
- AMA | June 20213 claims
- 200 | Solo: The Philosophy of the Multiverse1 claim
- AMA | April 20221 claim
- AMA | August 20221 claim
- AMA | December 20221 claim
Most important claims · 10 of 12Search all claims for “likelihood function” →
- Finding life on Europa or Enceladus would substantially raise credence that life is frequent because the likelihood of such a finding is far higher in a life-rich universe than a life-poor one, but you cannot draw the conclusion from Bayesian reasoning alone—you must state your priors, since with a sufficiently tiny prior the update would be negligible; conclusions depend on both likelihood functions and priors.AMA | April 2022
- Applying Bayes' rule to cosmological scenarios is stuck on the likelihood function because the only data is 'I exist'; it is unclear whether to ask the probability that any observers exist, observers like me, or exactly me, and whether one's own existence is genuinely new evidence or old evidence already baked into the priors (the problem of old evidence), making the update ambiguous.200 | Solo: The Philosophy of the Multiverse
- The objection that Bayesian updating requires 100% confidence in evidence is wrong; you do not need to leave the Bayesian paradigm but must be honest about the likelihood function, which is the probability of obtaining that data given the theory, and that likelihood must account for the possibility that your data-collecting or interpreting procedure is mistaken.AMA | August 2022
- Bayesian reasoning is an idealization that doesn't map onto the real world because we are finite and computationally bounded; likelihoods aren't truly objective when propositions are ill-defined, and even people agreeing on priors and likelihoods will reach different conclusions because they pay attention to different subsets of the overwhelming available information — attention, not information, is the scarce quantity.AMA | August 2021
- Being a good Bayesian about UFOs means almost all the disagreement comes from priors (which Bayesianism cannot adjudicate), while the likelihood of fuzzy unexplained sightings is near 1 under both alien and non-alien hypotheses, so grainy footage provides essentially no evidential leverage.AMA | June 2021
- You cannot use experiment to rule out a fluctuating universe and prove the past hypothesis, because Bayesian theory choice requires multiplying the likelihood by the prior; in a randomly fluctuating universe the prior probability that current ordered observations just fluctuated into existence is so overwhelmingly large that no amount of new likelihood data can beat it down, so the past hypothesis must be assumed rather than derived.AMA | December 2022
- The fine-tuning argument for theism is flawed not because Bayesian reasoning is illegitimate, but because its proponents cherry-pick a single fact (life exists) while ignoring that most observed features of the universe — its vast size, age, and messiness — are far more likely under naturalism than theism; a good Bayesian must weigh ALL the evidence.AMA | August 2021
- Empirical induction (seeing white swans and concluding all swans are white) is invalid because tomorrow's black swan refutes it, lacking the certainty of mathematical induction; the proper approach is Bayesian — assigning prior credences to competing theories and updating on observations, where the same data (twelve white swans) strongly disconfirms a 50-50 theory but gives zero evidence under a hemisphere-clustering theory.AMA | June 2021
- Bayesian reasoning is an idealization that does not fully map onto reality because likelihood functions are not truly objective when propositions (like 'theism') are not well-defined, because people assign different credibility to the same information, and crucially because finite, computationally bounded humans pay attention to different subsets of the available data and thus update differently.AMA | August 2021
- Popperian falsificationism, treating all conjectures as candidates and slicing away the refuted ones, is a crude approximation to Bayesian reasoning that fails because refutations are never cut-and-dried — an anomaly may indicate a new planet (Neptune saving Newtonian gravity) rather than a falsification (Mercury requiring general relativity); Bayes corrects this by never setting likelihoods exactly to 0 or 1, always allowing for experimental error.AMA | June 2021