Why Don't You Like Bayes Theorem?

Support Paulogia at

Paulogia Audio-Only-Version Podcast

Follow Paulogia at

[TestifyApologetics]

This is the one where I had the shower thought, dang it I should've said ____ instead.

I'd probably say something more like "Bayesian modeling is just that...modeling. It isn't a substitute for good judgment, which we agree on. Certainly you can see garbage in, garbage out when it comes to numbers plugged into Bayesian equations like with Carrier. Which again, we agree on the GIGO thing.

Bayes is certainly no panacea. But it also guards against many false moves, if one understands it. The structure of it is a guard. It is also important for understanding why hard-line "classical" models of apologetics that claim that you *must* first show that God exists before the historical argument are wrong. This is a big reason why I'm an evidentialist.

And perhaps most importantly to me, the modeling of cumulative cases is also assisted by an understanding of Bayesian probability."

[SpaceLordof]

Bayes theorem, used in the sciences, is incredibly useful, when you have data, and not an opinion.
Before you can honestly set a prior for a historical event, you need a baseline. What’s the baseline for a miracle?
Erik’s hot tea analogy doesn’t work when you are looking at unique, miraculous events.

[ziploc]

They appear to want to give their claims credibility by using Bayes Theorum instead of "and then a miracle occurred", by plugging in some miracle numbers as the input variables.

[bengreen]

funny how Christians think Bayes Theorem is only dodgy when mythicists use it.

[rillyx]

Erik (Tea Analogy Summarized):
If we see something that we have experience with (steeping tea) in an unlikely situation, then it can override our assumptions of that situation, therefore if we have non eye witness stories decades later of something we don't have experience with (resurrection of humans) it should also override our assumptions.

That analogy doesn't track for me.

[angelmendez-rivera]

What Erik and Kyle fail to understand about cumulative cases is that Bayes' theorem, and Bayesian analysis in general, *does not* support there being a cumulative case in cases where the individual pieces of evidence are so bad that they do not cross the threshold. If you have two independent pieces of evidence, A, and B, and a hypothesis H you want to investigate, then P(H | A) quantifies how well H explains A, and P(H | B) quantifies how well H explains B. If you take A & B combined as a single, unified piece of evidence, then you expect P(H | A & B) to be greater than both P(H | A) and P(H | B). However, the analysis reveals this is only true when both of the above are greater than P(A & H)·P(B & H)/P(A & B & H). If the probabilities are lower than this threshold, then this indicative of A or B being actually bad pieces of evidence for H. This accurately reflects something the Paulogia has said in previous occasions: combining multiple pieces of really bad evidence does not produce good evidence.

[gornser]

"So, you're saying there is a chance" as a cliche/meme sums up religions for me pretty much

[brianstevens]

Bayes works great as long as the priori are axiomatic, when you plug in crap you get crap out.

[lreadlResurrected]

I think a bunch of stories can make some people have "prior's" that are on a par with "once upon a time".

[fepeerreview]

0:47 I hugely respect you and your work, Paulogia. I want to challenge you a bit on Bayes. I think you actually stated it's real value. It forces people to assign probabilities to their priors. This puts them "on record". They have to commit. It is because Kyle had to commit in this way that you are able to challenge his priors more effectively. Kyle's mistake is to think that he can *arbitrarily* assign high probabilities to his priors. Now he can be challenged on those assignments. The mistake both you and Kyle are making is to believe the assigned quantification is "arbitrary" when it absolutely should NOT be arbitrary. It should be based on a data set of similar pieces of evidence and how they resolved themselves into correct and incorrect conclusions.

Example - Suppose we know that a certain author has 10 factual statement we've been able to verify, with 8 being false and 2 being correct. Now we encounter another factual statement by said author. With the information we have available we can assign a LOW probability this statement will be correct. If we can actually prove it is correct or incorrect through other means then we can add it to our data set. I'm simplifying a bit. But I'm trying to show that the process is not supposed to be "arbitrary".

3:20 Kyle has to misstate Dr. Carrier's position in order to make his point. That's a big red flag.

People weren't talking much about Bayes theorem until Dr. Carrier came along. He talked about it because he recognized the very real problem in biblical history, a complete lack of discipline in being *objective* about the value of the evidence being examined. By trying to find a way to quantify it he is trying to find a way to hold historians to more rigorous standards on the strength of the evidence they use. The academic study of history very much needs this kind of discipline and rigor if it is going to remain a relevant area of study. Obviously it can never be treated as a hard science like chemistry. But it *must* start working within a more objective framework.

2:05 Erik and the hot cup of tea - This is an interesting example because by any standard it MUST be assigned an extremely high probability of indicating human presence, since the science behind how quickly a cup of tea would lose its heat is completely understood and CALCULABLE. So he has actually shown us why Bayes is helpful here. His problem is that he has no "cup of tea" equivalent for the resurrection.

[fluffysheap]

Bayes is math, you can't be for or against it.

If it's misused, you still can't be for or against it, but it can be invalid.

[Obi-Wan_Pierogi]

“You’re saying there’s a chance” 😂

Swing and a miss

[zencaser]

Paul's very complaint exposes why Bayesian modeling is useful. If Erik's priors are high and Paul's priors are low - both have put a stake in the ground and committed. There is now a point for evidence and argumentation and examination. They need to justify their priors. The audience/peers/posterity/etc. can examine the priors and the weighting of the evidence.

If history is just someone writing a book or a blog post saying, "hey I like my pet theory and I think it's probable", without being forced to justify this conclusion with priors and evidence - well, then we're just being entertained with an interesting narrative.

History is science, only with comparatively terrible data.

[mugglescakesniffer]

Ooooh so, your saying....there's a

[scottduke]

Great critique of Bayes Theorem (the way it is used by many people)! Well put.

[theintegrator]

When I was a college professor (teaching tech writing) the engineering school and students preferred we give the papers a number score instead of a letter grade, suggesting that a 93 in a paper was more objective than an A-. The ultimate number I chose, of course, was just as subjective but numbers mean a lot to certain people.

[XDRONIN]

What prior event (equals) the resurrection for Bayes to work? The answer is none

[JohnnieHougaardNielsen]

Where Bayes theorem can work really great is when there's reasonable agreement about the values for the priors (like risk factors). Without that part, it indeed gets to be somewhat arbitrary and up to widely varying personal opinions about probabilities. With that agreement, the math to combine probabilities leads to good results which can be surprising to people not doing "Bayesian thinking", which can very well be done without going through the steps of explicitly applying Bayes theorem.

[drewharrison]

Oh, you haven't read Carrier but, you know what his motivation is? Good one, Erik.

[Sam_on_YouTube]

So you're not against Bayes Theorem, your against misusing it in these types of situations. It works where it was actually intended to be used.