Jaynes’s book (first part): https://bayes.wustl.edu/etj/prob/book.pdf

Permanent class page: https://www.wmbriggs.com/class/

Uncertainty & Probability Theory: The Logic of Science

*Link to all Classes.* Jaynes’s book (first part):

**Video**

Links:

Bitchute (often a day or so behind, for whatever reason)

**HOMEWORK:** In the homework, assume in the same box you observed D = 10 good and 3 bad widgets. What is Pr(A|DX)?

**Lecture**

**Probability Of Hypotheses**

Everything I did on the board in the video is done in full detail in Jaynes, so there is no point repeating that here. The links to his book are above. It is very simple math, and beautifully explained. You must read it.

I highlight here only one thing. Recall all (as in all) probability fits the schema Pr(A|X), where A is our proposition of interest and X the evidence we assume is true. As in assume is true. Whether it is true is completely beside the point. We want to judge the uncertainty of A with regard to X. That is logic, which is to say, probability.

“A”, as we have learned over and again, can be any kind of proposition. It can even be a—*drumroll*—hypothesis!

* <<Audience gasps. Realization dawns. Slow applause crescendos into an ovation.>>*.

When it is given this glorified name, *hypothesis*, probability becomes *Science*. A hypothesis is no more, and no less, than a proposition.

Since Pr(A|X) is the uncertainty of A with respect to evidence X, we have just done—men, steel yourselves; ladies, set down those dishes—*a hypothesis test.*

Sort of. Because that’s not what is ordinarily called a “hypothesis test”. We’ll learn the official version later, when I recommend against it with all the strength I can muster. We have, though, “tested” A with respect to X.

Notice very very most very carefully that Pr(A|X) is a **not** decision that A is true or false. Unless, of course, the evidence X insists this is so. Official hypothesis testing makes decisions for you, it makes you say “Treat A as true (or false)”, which is one of its main weaknesses. In real life, whether A should be treated as true or false depends on the probability of it with respect to X, whether X is important to the decision maker, and whether Pr(A|X) is large or small enough to be important to that decision maker. Decisions and bets *are not probability*.

Now suppose you want to entertain not only X, but also evidence D. No problem for probability. We now want Pr(A|DX). Done! It really is that simple.

Of course, if X assumes some mathematics, sometimes Pr(A|DX) can be written in a nice way to facilitate computation (as in Jaynes), but that’s all that is happening. We don’t need Bayes. It is only a helpful tool, which is only helpful sometimes. What we really want is Pr(A|DX), and any way we can get it is good enough.

This should remove all mysticism statistical probability brings. “A” does not “have” a probability. There is no “true” probability of A. *Nothing has a probability*. Propositions only have probabilities with respect to assumed evidence. Change the evidence, change the probability. It is as simple as that.

I’ve said it dozens of times so far, and will repeat it dozens more, but this is *all of probability*. Once you have learned, and truly assimilated this lesson, all the rest is minor detail.

Like, for instance, what if want to judge X itself. We want X is to true about the world, as as close to true as we can make it. That is, we want Pr(X|W) ≈1 for evidence W about the world. All right. Then we can compute this Pr(A|DXW)! We’ll work out the niceties of this at a later class.

**Subscribe or donate to support this site and its wholly independent host using credit card click here****. Or use the paid subscription at Substack. Cash App: $WilliamMBriggs. For Zelle, use my email: matt@wmbriggs.com, and please include yours so I know who to thank.**

As I recall Bayes' theorem, it doesn't quite align with how you're presenting it. Here's what I remember: P(A) is the prior probability, P(A∣DX) is the posterior probability, and P(DX) is the llkelihood. But the terminology in Bayes' theorem has always puzzled me—there’s nothing inherently 'prior' about the prior nor is it necessary for posterior to come after. As you mentioned, prior(and posterior) is not tied to time or sequence in the way the term might suggest. Anyway, that's how I recall it. And what’s this about deep kimchi? Kimchi is a staple in Korea.