An Interview With Richard Carrier About His Book, "Proving History"

Richard Carrier has kindly agreed to answer some questions I posed after reading his soon to be released book Proving History: Bayes's Theorem and the Quest for the Historical Jesus (Amherst, NY: Prometheus Books. 2012). This book introduces Bayes’s Theorem as a basis for assessing whether or not Jesus existed, a subject he will deal with in a forthcoming book titled, “On the Historicity of Jesus Christ.”

Let me first go on record as saying I think Richard’s scholarship is deep, his attention to detail surprising, and his breath of research wide-ranging. I predict he will continue making a huge difference on a wide range of subjects. It’s good that he’s on our side, the side of reason and science. ;-)

John: What do you mean in the Preface when you say “I have no vested interest in proving Jesus didn’t exist. It makes no difference to me.”

Richard: As I go on to explain there, it really doesn’t matter to me, in the way it does to believers. I’m not invested in any theory proclaiming otherwise, and the historical Jesus, or perhaps I should say Jesuses (as there are several) proposed by mainstream scholars today pose no challenge to my worldview. If my method can be used to prove Jesus existed, I’d count that a win. Then we can finally move on to something else. I just don’t think that’s how it’s going to pan out in the end. But that’s calling the game too early. Let’s see what happens.

John: What is the central aim of this book?

Richard: It has two central aims, one particular and one general. On the particular end: to demonstrate that the methods presently used to prove Jesus existed are illogical or unsound, and to articulate what methods we should employ instead, which I show must be fundamentally Bayesian (meaning, in accord with a what’s called Bayes’ Theorem). But on the general end: to demonstrate that Bayes’ Theorem actually underlies all valid and sound historical arguments, in any field whatever; and realizing this has profound implications for how all historians reason, about any subject whatever, not just Jesus.

John: When it comes to historical studies about Jesus what is the specific problem you seek to address in this book?

Richard: Basically, every scholar who looks at the same evidence comes up with a completely different Jesus. As I show in the book, I’m not the only one noticing this. Many scholars in the field have been complaining about this for almost a decade now. If everyone applies the same method to the same facts and comes up with a completely different answer, there is clearly something fundamentally defective about that method. I show what those defects are, but in chapter one I summarize the general problem with them. But chapter five really tackles these problems detail-by-detail.

John: Why should non-experts leave good history to the experts?

Richard: My whole second chapter answers that question. But it can all be summarized by a single point: getting historical conclusions right requires such a vast specialized background knowledge, and such specialized training in methods and their pitfalls, that it is simply not possible for a layman to have this experience, unless they spend many dedicated years acquiring it (and that’s a lot harder to do on your own; although not impossible).

John: I think you are correct that we should leave good history to the experts. But how can non-experts judge between the experts to know which one of them is right when they disagree if the experts all adopt the math of Bayes’s Theorem? How can non-experts judge their competence when the math can be tough?

Richard: The math shouldn’t ever be that tough. In my talk for last year’s Skepticon, Bayes’ Theorem: Lust for Glory, I explain why we should all have at least high school math down pat (and if we don’t, we need to get on citizens of an advanced democracy, and as dedicated skeptics, it’s a must). But that’s all you need. Add that to the instruction I give in my book and you should be able to judge Bayesian arguments in history as well as you can judge any expert without being one yourself. And we all don’t have to be a plumber to be able to tell when your plumber sucks at his job or is really brilliant at it; you won’t know how to evaluate every detail, of course, which is often why we seek second opinions, but that’s what we do in history, too: we look at what the developing consensus is.

The only thing I’d add to what I already say about this in the book is that we should simply dismiss fundamentalists from here on out. If they can’t adduce any non-fundamentalist expert who agrees with them, then they are spewing propaganda, not scholarship. And if they can find one, often they pick someone obsolete (and thus don’t tell you no one agrees with him anymore), or a maverick (like in climate science, there are always a few nutjob “experts” who gainsay hundreds of experts—so when faced with that conundrum, the hundreds win; because if the maverick can’t convince them, why should he convince you?), or they pick a side in an unresolved debate and represent it as fact, when real scholars would tell you that because there is so much expert disagreement on that point there simply is no established fact of the matter—it should be treated as unresolved.

This is relevant to the historicity debate, because it means laymen should at this point side with the consensus of non-fundamentalist experts: that there was some sort of historical Jesus, but he was nothing like what the fundamentalists say he was. I will be challenging that consensus (and I show even in this book why that consensus is unusually ill founded), but the process still has to proceed: experts have to examine and debate my case; and then in a decade or two we’ll see where the challenge stands.

All I must say at this point is that the theory that Jesus did not exist is more plausible than historians currently give credit, but only in a non-crazy version of the theory. And this is because experts haven’t seen all the evidence in one place, and structured in a logically sound way from premise to conclusion. They have only seen badly argued, fact-challenged versions of the theory, and it’s no wonder they dismiss it. Really, if that were the best anyone could do, they should dismiss it. And so did I. Until I realized there was a better case to be made.

Part of that case is realizing that historicists have been resting their conclusions on a house of cards: a completely bankrupt methodology. And Proving History aims to communicate that fact. I know many experts in the field who don’t know this, even though it’s been said in numerous academic press publications by prominent experts in the field already—information just doesn’t disseminate fast enough, due I suspect to information overload (there are too many books and articles about Jesus published in a year to keep up), and too many experts are so dogmatically set on their own “theory” that they never even look for scholarship that might be challenging their fundamental assumptions. And thus they never discover it, even when there is plenty of it, easy to find. So if you run into an expert who doesn’t know this, tell them to read my book. That’s how word will get around, and transform the field.

John: You provide twelve good axioms that represent the epistemological foundation of rational-empirical history. Tell us about Axiom 4: “Every claim has a nonzero probability of being true or false (unless its being true or false is logically impossible).”

Richard: Basically, it’s the obvious epistemological point: we can always be wrong about anything. And that means there is some probability of our being wrong. And that means there is some probability that what we think is false, is actually true. We have to acknowledge this. Because it forces us to ask what that probability is. Is it too low to worry about? How do we know? How would we know? Proving History is about answering questions like that, in the subject of history. One result is that by thus considering alternative explanations of the evidence, our “favored” explanations sometimes don’t look so certain anymore. That’s in fact the biggest mistake made by historians: failing to consider alternative explanations. They end up slaves of verification bias. Bayes’ Theorem corrects that.

John: Tell us about Axiom 5: “Any argument relying on the inference ‘possibly, therefore probably’ is fallacious.”

Richard: Quite simply, too many historians (and not just in Jesus studies; in every field) think they have made their case if they can come up with any plausible explanation. “Well, it could have been...” is assumed to be a sufficient rebuttal to anything. But that’s fallacious.

The most obvious example of this mistake appears in fundamentalist “harmonizations” of Gospel contradictions: they think they have “rebutted” the conclusion that the Gospels are contradicting each other if they can think of “any” possible way to harmonize the accounts, developing a fanciful “just so” story that makes everything fit, by assuming a hundred things not in evidence. But that ignores the fact that that account is actually extremely improbable. That Matthew is deliberately contradicting Mark because he is arguing against Mark is vastly more probable than that Matthew and Mark are correctly describing exactly the same events. Thus, the fact that the latter is “possible” is irrelevant to what we should conclude. We should conclude with what is by far the most probable explanation of the evidence, not what is merely possible. This is an easy example because no sane scholar accepts such harmonization nonsense. Only crazy fundamentalists think it’s convincing. But varieties of the fallacy appear even in sane, mainstream arguments.

This does not mean, however, that all mere possibilities should be ignored. If there are several possible explanations of the same evidence and we can’t determine which is more probable (or we can at best establish only a slight advantage to one over the others), then we have to accept that we don’t know what happened—it could be any of the most probable possibilities. And historians of antiquity have long been comfortable with this, although human resistance to uncertainty does sometimes plague even them.

John: You also enumerate twelve rules you would like to see all historians consistently follow. Rule 1 is to “Obey the Twelve Axioms and Bayes’s Theorem.” Then you say. “This does not mean you must use Bayes’s Theorem in any mathematical sense, only that any historical argument you employ must not violate Bayes’s Theorem.” For my readers would you explain Bayesian reasoning without the math?

Richard: Basically, whatever argument you make, we have to be able to model it in Bayes’ Theorem and get the same answer. Because if we don’t, then you’ve violated basic logic somewhere. This means, however, that all sound and valid arguments in history are already Bayesian. Their authors just don’t know it. Since all historians make plenty of valid and sound arguments, all historians have been using “Bayesian reasoning without the math” all along. My book just opens their eyes to that fact. And once your eyes are thus opened, you can see more fallacies (and thereby avoid them), and more ways to soundly argue, and thus you will improve as a historian. We can also see more ways to analyze and criticize historical arguments generally, not just our own.

As there are many different ways a Bayesian argument can proceed (as it rests on four premises regarding key probabilities, and probabilities can exist in all kinds of ratios to each other, making for hundreds of different combinations), there is no “one way” to describe Bayesian reasoning without the math. I provide two flow charts in the book’s appendix to help with that, and I start chapter three with an example of covertly Bayesian reasoning (you won’t even know you were doing math until later in the chapter). But one of many ways to describe Bayesian reasoning in ordinary terms is this: the correct explanation of anything is always more likely to be what typically explains such things (because typical = more common = more frequent = more probable), so the more atypical an explanation is, the more improbable the evidence must be on any other explanation before we can believe an atypical explanation is more probably correct than a typical one. Accordingly, Bayes’ Theorem proves that an argument like “That body of evidence is very unlikely unless x is true” is a good argument, unless it’s even less likely that x is true on prior knowledge. So we have to take into account prior knowledge. And we have to take into account how well the evidence fits each competing theory—not merely that it does fit (since any body of evidence can be made to fit almost any theory). And these all require measures of degree. Which is what gets you back to the math. But the fact is, our brains are already doing this math. All the time. Every historical argument ever made in history has been fundamentally mathematical. We just don’t realize it, because we don’t examine the logic of what we are saying and thinking.

John: Would you briefly explain the math of Bayes’s Theorem?

Richard: In one form (which I provide in the appendix), called the “odds form,” the ratio of prior probabilities is multiplied by the ratio of consequent probabilities, to give you the ratio of posterior probabilities. So what does all that mean? The “ratio of prior probabilities” is the probability of your explanation on prior knowledge, in ratio to the probability of all other alternatives on prior knowledge; and the “ratio of consequent probabilities” is the probability of the evidence (the probability that we would have that evidence, exactly as we have it, instead of something else) if your explanation is true, in ratio to the probability of that same evidence if any other explanation is true instead; and the “ratio of posterior probabilities” is simply the probability your theory is true (full stop), in ratio to the probability that some other theory is true, which is the conclusion you are aiming for: how likely is your explanation, all things considered. The standard form of Bayes’ Theorem gives you a more straightforward conclusion, a straight-up probability that your explanation is true. I teach both in Proving History.

John: What does math have to do with history?

Richard: As I noted just above, all historical argument is already fundamentally mathematical. Any time you say something is “more likely” than something else, that an explanation is “improbable,” or “almost certainly true,” or “implausible,” and so on, you are making mathematical statements. Any time something is “more” than something else, that’s math. Just try to make a historical argument without ever referencing how likely anything is, or how plausible it is, or making any other statements of probability or degree, and you’ll realize: history is math. So we should pop the hood and look inside what’s going on in our brains when we speak and think this way, and figure out how to check the math and do it right. Otherwise we are relying on our untrained intuition, which is not only unreliable, it also can’t be checked or critiqued, since we don’t even know what it’s doing. My book pops the hood, and shows you what it’s doing.

John: How would a Bayesian approach to history help historians come to an agreement and make progress possible?

Richard: By making their intuitions and assumptions explicit and thus subject to examination and criticism—by themselves, too, not just by others (and any honest historian should want that, to make sure they are reasoning correctly).

For example, if a historian rebuts an opponent by saying his theory is “implausible,” what exactly does that mean? How does he arrive at that conclusion? How can we critique or check whether that conclusion is valid, when we don’t even know what premises it was based on, or by what logic the conclusion is being reached from those premises? And indeed, are all implausible explanations false? If not, then calling an explanation “implausible” is not a sufficient rebuttal. Whereas if all implausible explanations are false, then why? And how do we know when something is that implausible?

Bayes’ Theorem forces historians to confess what probabilities they are estimating for what things, and why they are coming to those estimates instead of others. Their assumptions are thus exposed. And once exposed, often they won’t stand up to criticism. Or if they do, then we will have ended up with a much more robust proof of their conclusion. Either way, we end up with better history.

John: When it comes to extraordinary claims what questions does Bayesian math ask in order to assess their probability?

Richard: By definition, an “extra-ordinary” claim is an improbable claim: because the very word means not ordinary, and ordinary = common = frequent = probable, so not ordinary = not common = not frequent = not probable. Usually, of course, we reserve the word “extraordinary” not merely for something uncommon, but something extremely uncommon, or even so uncommon that we really don’t have any reliable evidence it ever happens at all. Thus, when we use the word in that sense, we are saying that that claim is extremely improbable on all our prior knowledge of history and humanity and the universe.

What Bayes’ Theorem tells us is that it is illogical to believe such a claim, unless the evidence for that claim is extremely improbable on any other explanation but the extraordinary one; in fact, the evidence must be even less probable on other explanations than the extraordinary explanation is improbable on all prior knowledge. Thus, extraordinary claims require extraordinary evidence. I give a logical demonstration of that in Proving History, so the principle can never be gainsaid again.

John: I must say that your defense of using Bayes’s Theorem for historical research is convincing. One criticism is the problem of subjectively assigning probabilities to prior probabilities. How do we assign numbers to prior probabilities that a person was healed in Lourdes, France, or that a witch flew the night to have sex with the Devil, or that an axe head floated, or that the hand of Moses turned leprous?

Richard: It depends on what your goals are. If you want to know the actual probability of those things, then you are faced with a different problem than if you simply want to know whether you should believe them. You don’t need to know the “actual” probability of a meteorite killing you in the next five minutes in order to know that, whatever that probability is, it is certainly so small you needn’t believe it’s going to happen. Thus, when faced with absurd claims like that, subjective probabilities are irrelevant. Because we can generate objective probabilities that are good enough to reach a conclusion by.

In Proving History I explain a whole system for doing this in detail. Basically, it operates on two principles: (1) if you start with a premise of the form “the probability is [x] or less” then your estimate can be wildly off and still get you a conclusive result, since the conclusion will then be “the probability is P or less” and if P is then extremely small, we don’t need to know what the actual probability is, because we know it’s vastly smaller than P; and (2) if your “wildly off” estimate is honestly as “wildly off” as you can reasonably believe it ever to be, then your conclusion, P, will necessarily be as high as you can reasonably believe it ever to be, so if P is then extremely low, you are fully justified in rejecting the claim. In fact, that conclusion follows necessarily, as a matter of inviolable deductive logic. No matter how subjective the facts are. That is, as long as you aren’t ignoring anything you do in fact know. Otherwise, what you know is the best you can ever judge by. So the only way to legitimately change your conclusion is to get better evidence. Period. There is no other logically valid way out of that conclusion.

Of course, sometimes P will not be so low. But then we can analyze the case more carefully to see how much we can honestly narrow our estimates, or else accept whatever degree of uncertainty P warrants.

John: In chapter 5 you analyze the various proposed criteria to judge the probability of an authentic saying of Jesus in the Gospels (like dissimilarity, coherence, multiple attestation, etc), and you find them all wanting for various reasons. You spend a lot of time on the criterion of embarrassment. Why is that considered by you inadequate, especially when it comes to the claim that Jesus was crucified?

Richard: Not just sayings, by the way, but all facts, such as actions, events, and facts (like whether he was really ever a resident of Nazareth). I spent the most time on the argument from embarrassment because it is the most used, the most crucial for establishing historicity, and the most important for understanding why it is invalid. It then becomes an excellent model for seeing the deficiencies in all the other criteria, which are often much easier to see the faults of. The basic reason it doesn’t work is that it rests on assumptions that aren’t true most of the time, especially when applied to the documents we have for Jesus. I explain at great length in the book why that is.

Notably, as I was sending in the final proofs of Proving History and had already completed my fully peer reviewed case, I discovered that Mark Goodacre and several other prominent scholars were preparing extensive critiques very similar to mine, to appear in another book that may be out later this year (Jesus, History and the Demise of Authenticity, ed. Chris Keith and Anthony LeDonne), dismantling the entire method of criteria. As I show in chapter one of my book, this adds to a rising trend in the field. Basically, every expert who has specifically examined the validity of the criteria, and published books or articles on them, has concluded they are defective. This is becoming the new consensus. Indeed it already is the consensus, in the sense that all experts who have become specialists in the criteria are in agreement on this point.

When it comes to the crucifixion argument, the basic version you hear is that that was so embarrassing no Christian would claim it unless it were true. But this can be refuted with a single example: the castration of Attis was also embarrassing, yet no one would argue that therefore there must really have been an Attis who really did castrate himself. Arguably this was even more embarrassing than being crucified, as heroically suffering and dying for one’s beliefs was at least admirable on all the value systems then extant, whereas emasculating yourself was regarded as the most shameful of all fates for any man. Yet “no one would make that up” clearly isn’t a logically valid claim here. Attis did not exist, and a non-existent being can’t ever have castrated himself. So clearly someone did make that up. It’s being embarrassing did not deter them in the slightest. And in fact that is true throughout the history of religions: embarrassing myths were (and in all honesty, still are) the norm, not the exception.

There are many other reasons why the argument fails here, but they all reduce to the same Bayesian point: there are other explanations of the evidence (other reasons why a god or hero would be depicted as humiliated and murdered, like the goddess Inanna was, or the god Prometheus was) that are not sufficiently improbable for us to assume “it’s true” is automatically the best explanation. Thus “embarrassment” just isn’t a valid argument. You need to look at all the available explanations and compare their relative probabilities.

John: Several criticisms of the Bayesian method have been proposed which you effectively deal with in your last chapter. One such criticism was made by philosopher John Earman that we can never know all logically possible theories that can explain the evidence and thus we can never know what their relative priors would be, even though they must all sum to 1, which would seem to leave us in a bind. How do you answer that?

Richard: The same way Earman did. “Lost theories” are no different than “lost evidence.” We can only conclude what’s most probable from what we know at the time. More information will change that. And in fact Bayes’ Theorem specifically allows for this: the probability that a claim is false is precisely the probability that new information will correct us. Discovering a new theory is just another form of new information. Thus Bayes’ Theorem is not a magic oracle but simply a theory of warrant: it tells us what we are warranted in believing at any given time. Nothing more. It does not give us certainty. No method can. It just gives us probabilities. Which is all we ever have, whether we acknowledge Bayes’ Theorem or not.

However, this does not justify carelessness. We can only conclude “some new theory will come along and change everything” is improbable if we’ve actually done our due diligence in (a) examining the theories so far actually proposed and (b) thinking through all the obvious logical possibilities. How likely we are to still be wrong after all that, is precisely how likely an alternative explanation is. We then plug that into the equation.

John: Since historical evidence is weak and because improbabilities happen all of the time, as you say, what are we to do when several different historians come to conclusions that do not achieve any higher probability above slightly improbable (40%) or even improbable (20%)?

Richard: We do what honest historians have always done when this happens (and in my field, ancient history, it happens a lot): we conclude that we don’t know what happened. We might be able to say that some things are somewhat more likely than others, but that also several of those alternatives are not too improbable to rule out.

For example, did Alexander the Great ambush the Persians at Granicus, or charge pell mell across the river and wage an amazing hand-to-hand melee with them in the water? The latter we have from an eyewitness. The former we have from a later informed expert. But the former is vastly more probable on prior evidence (of how battles and wars are typically waged, and how generals as successful as Alexander typically make engagement decisions, and what unit tactics we know Alexander relied on to defeat the Persian empire generally). So at best they seem equally balanced; and in fact, I am inclined to doubt the eyewitness. You could go either way. Certainly one of them is false. The evidence simply isn’t sufficient to know for sure. I think all honest historians who examined all the arguments pro and con would side against the eyewitness, but none of us would bet our lives on that conclusion. And so it goes.

John: In your forthcoming book you’ll test between two hypotheses: h = “Jesus was a historical person mythicized” and ~h = “Jesus was a mythical person historicized.” Care to give us an advanced introduction to that book and/or where your research has led you so far based on Bayesian methodology?

Richard: It’s no secret that I’ve come to the conclusion that ~h is more likely. And the more I’ve researched it, the more certain I am of that. I keep finding evidence supporting ~h; whereas evidence for h keeps disappearing the more I examine it. However, my conclusion does come close to the Granicus example above. I am not supremely certain. I just think it’s more likely than not. But this won’t be any comfort to Christians, since the next most probable hypothesis is that Jesus existed but we know essentially nothing about him. Which, incidentally, a lot of experts in the field are starting to agree with. It’s slowly becoming the consensus position. There are still hold outs, like Bart Ehrman, but I don’t think their position is going to survive in the long run. There are just too many cats out of the bag at this point. But what will be the fate of the next-step position, that there wasn’t even a Jesus at all? Time will tell. But someone needs to present the case properly before it can be conclusively accepted or refuted. No one has done that yet. My future book On the Historicity of Jesus Christ will. In the meantime Proving History does a good job already of showing why that currently growing consensus is correct; and it’s just one step from there to full mythicism.

John: Thanks so very much for your time. I hope this present book is read widely and helps advance the discussion.