sciencebriefs
all subjects →
10:37in productionCh. 1 · The inverse problem/ 10:37 · ceiling 15 min
Computing & AI · Applied science

Bayes' theorem

Bayes didn’t discover belief updating—he discovered how to bound a binomial probability parameter using conditional probability.

Bayes's 1763 essay solves a narrow inverse probability problem using conditional probability. It computes limits on a binomial parameter—not a general theory of belief updating. Its power lies in precision, not generality.

Chapters & takeaways4
  1. 1:09
    The inverse problem

    Bayes solved an inverse probability problem: infer hidden structure (e.g., urn composition) from observed outcomes.

  2. 3:00
    The algorithm

    He built a specific algorithm—Proposition 9—to compute limits on the binomial probability parameter using conditional probability.

  3. 4:50
    The editor’s hand

    Richard Price edited, framed, and presented Bayes’s work—shaping how it entered scientific discourse.

  4. 6:39
    The theorem, not the theory

    The 1763 essay contains a theorem about conditional distributions—but only for one case, not a general rule.

Worth your time?

Yes. Study the whole thing.

4.5/ 5
What works
  • inverse probability
  • binomial parameter estimation
  • conditional distribution derivation
What does not
  • general theory of belief
  • subjective probability
  • modern Bayesian inference
Study it if
  • statisticians
  • historians of science
  • AI practitioners
Skip it if
  • casual readers seeking life advice
  • those expecting a 'discovery of rational thinking'
The written brief1 min read

What the work claims

The essay claims a solution to an inverse probability problem: given observed successes and failures (e.g., black/white ball draws), compute limits on the unknown true probability of success. It frames this as a problem of inferring urn composition from samples—not as a universal rule for belief revision.

How it was done

Bayes used conditional probability to construct an algorithm—Proposition 9—that calculates limits on an unknown parameter using observed evidence. He solved an inverse probability problem posed by Abraham de Moivre, specifically inferring the probability parameter of a binomial distribution from outcomes. Richard Price edited Bayes’s unpublished manuscript for two years before presenting it at the Royal Society on 23 December 1763.

What holds up

The core mathematical result holds: given observed outcomes (e.g., draws from an urn), Bayes’s algorithm computes bounds on the underlying probability parameter. This was verified through conditional probability logic applied to the binomial setting. The result appears in the essay as a theorem about the conditional distribution of R given X₁,…,Xₙ.

What does not

The 1763 essay does not state Bayes’ theorem in its modern general form. It does not define prior or posterior distributions as abstract concepts. It does not apply beyond the binomial case. It makes no claim about subjective belief, learning, or scientific method.

Why it matters beyond the lab

It established the first formal method for inverting probability statements—a conceptual pivot that later enabled statistical inference, machine learning, and probabilistic AI. But its cultural expansion into ‘Bayesian thinking’ far exceeds what the 1763 text supports.

Is it worth your time

Yes—if you work with evidence, uncertainty, or inference. Bayes’s method remains foundational for updating beliefs in statistics, computing, and decision-making. But it is not a general theory of reasoning; it is a narrow, mathematically precise solution to one type of inverse problem.

Same field · Computing & AI4 of 32
Up next in Science

Benoit Mandelbrot

· 9:57

Fractals do not explain nature — they rename its roughness with precision.

9:57