As I recently mentioned, a Harvard University philosopher, Aviezer Tucker, just published a review of my book Proving History for the academic journal History and Theory (Vol. 55, February 2016, pp. 129-140), titled, The Reverend Bayes vs. Jesus Christ. Tucker is an expert in the methods and philosophy of history, so his review carries some weight. It’s significant, therefore, that he endorses the program of my book—that historians need to start using Bayes’ Theorem, as effectively as they can, to resolve questions in their field—and that in fact even when he criticizes my book, he does so by suggesting improvements that are either already in that book (and he merely overlooked them) or in my subsequent application of its program in its sequel, On the Historicity of Jesus. This is almost the best assessment one could expect. It lacks merely noticing that much of what he suggests, I already did. What I provide below is an analysis of his review that helps understand his points, and relates them to what I’ve already written.
The best line of the whole review is his declaration that (with my emphasis):
If philosophers of historiography accept that historiography is inferred and probable, they must come to terms with the extensive Bayesian literature, ask what it means for historiography, can it be applied for analyzing the relationship between historiography and evidence, and if so, does it need to be revised or adjusted for this task, and if so, how?
Amen. It is also important to note his qualifying remark: “I disclaim any expertise in the historiographical debate about Jesus.” He is thus not attempting to evaluate any claims in that field. He is only interested in looking at my Bayesian analysis of them and assessing how far it gets toward what he’d like to see done. But he does make a few observations about that field that are apt enough to mention.
For instance, Tucker argues that it is folly for anyone to assume the Gospels were written to be taken as literal history, rather than symbolic allegory. Ironically he says this as if to criticize me for making that assumption, even though of course I am renowned for arguing his very point! For instance, in Chapter 10 of On the Historicity of Jesus I lay out a detailed case for the Gospels being symbolic allegories—even the later ones, Luke and John, who posture themselves as reporting true history; I show in Chapter 4 how that was actually in accordance with ancient tactics of religious persuasion, in which texts would be written for two audiences, the yokels who only grasp stories meant literally, and the educated elite who understand (and in fact prefer) deeper allegorical meanings (Elements 13 and 14, pp. 108-24).
What threw Tucker off was that in Chapter 3 of Proving History I use as an illustrative example of how Bayesian reasoning works the darkness at the death of Christ reported in the Synoptic Gospels. Simply because it’s easy and obvious, and helps the reader see how they already reason with Bayes’ Theorem, they just don’t realize it. For that example I compare a literal meaning with a fictional one as the two competing hypotheses for how the story came to be in them (either because the event happened, causing it to be recorded, or because someone made it up, causing it to be recorded). Tucker mistook me as arguing that this was somehow related to the historicity of Jesus (even though I never connect it to that question), as if that one story being false made the historicity of Jesus less likely, which he rightly criticizes as invalid thinking. Which is a relief. Because that’s exactly what I argue in OHJ: that even the entire contents of the Gospels being fiction does not decrease the probability of Jesus existing. If, for example, they were solely and entirely intended to convey deeper truths, then the authors weren’t even concerned about any real history of Jesus. That wouldn’t mean there wasn’t one. I thus assign the Gospels no value to the debate (in OHJ, they end up with a likelihood ratio of 1/1, pp. 597-98, explained on pp. 506-09; the one exception being their mythic hero markers, which put Jesus in a distinctive reference class: p. 395).
Tucker proposes literary meaning rather than historical: he suggests the darkness merely conveys something the author wanted to say about the death of Jesus. Which falls under my hypothesis of fiction (that being just one of many causes of fiction). So really, he isn’t criticizing what’s actually in Proving History, but verifying that it’s correct. Tucker similarly approves of the approach in general, noting for example that when I reveal the Bayesian structure of literary mimesis as a cause of stories (i.e. the “emulation” or rewrite of earlier stories about other heroes to communicate ideas, as opposed to actual historical events being their cause):
Carrier’s sole but good example for likely mimetic relation between the Gospels and earlier texts is between the story of the resurrection of Jesus and Daniel’s salvation in the book of Daniel. He could have added the similarities between aspects of the Jesus narratives and the biblical stories about Abraham, Joseph, Moses, King David, Elijah, and Isaiah.
And indeed I do exactly that in OHJ. This is impressive endorsement. He gets it. Indeed, he concludes on this point:
Carrier’s explanation of some of the evidence in the Gospels is fascinating because, as far as I know, it is the first Bayesian reconstruction of structuralism and mimesis. It may be fruitfully applied to textual criticism, the history of literature, anthropology, and more.
In this and other respects Tucker views my project as achievable: we need to analyze all forms of historical reasoning in Bayesian terms, and use that analysis to make it better and more checkable.
On the other hand, Tucker does conclude with some pessimism about history as an academic field. He notices I “believe that historians are generally rational” and sharing Bayesian methods will “allow them to reach a consensus about most issues.” And he agrees “Bayesianism explains why historians do not have to agree precisely on their estimations of prior probabilities and likelihoods to generate similar enough posterior probabilities.” But he thinks I’m overly “optimistic that disagreements among historians, usually about background information and likelihoods, can be resolved by discussion and exchange of information.” He agrees that’s the only way it can be resolved. But he worries that there may be barriers. He calls attention to all the problems I found in the historiography of Jesus (after all, I conclude, “the methods so far used by Jesus historians are either invalid or invalidly employed,” Proving History, p. 204). And he rightly notes that I didn’t bother offering hypotheses as to why that is (which, BTW, was mainly because I am not sure how to test them).
Tucker proposes to fill that gap with the following assessment of his own:
From a Bayesian perspective, the following nonexclusionary and nonexclusive hypotheses may explain the remaining differences [among modern Jesus historians]: Historians may be working with different Jesus hypotheses (both positive and negative) because “Jesus” may have different meanings [to them]. The absence of consensus may then be semantic, not methodological. Historians may start their investigations with different prior evaluations of the probabilities of various Jesus hypotheses. They may be more dogmatic than historians in general and hold on to these priors despite evidence that they should revise their beliefs. Historians disagree on the reliability and relevance of evidence. All Jesus historians agree on the relevance of the Synoptic Gospels and the Epistles as evidence, but they may disagree about the reliability of parts of those texts and may admit or deny the relevance of other evidence, for example, Gnostic Gospels. Finally, some of the scholars who wrote about Jesus may not have been historians but theologians who used entirely different nonhistorical [methods] …
Some of this I agree with and some of it I don’t.
Tucker is certainly right that theology often gets substituted for history in this subject all too often. This has even explicitly been said by the famed Jesus historian John Dominic Crossan, that the study of Jesus could meet objective standards, “at least when it is not a disguise to do theology and call it history, do autobiography and call it biography, do Christian apologetics and call it academic scholarship.” This is Tucker’s point as well. Whatever they are doing, it’s disingenuous to call it history. I concur. Likewise Tucker’s point about stubborn dogmatism in the field, which plagues even secular and liberal historians of Jesus; much more so Christians and conservatives. That being psychological and not logical, obviously you will never persuade someone who is immune to evidence and reason.
But the other three points I think Bayesian reasoning can solve, provided we can get past those two barriers I concur exist.
Indeed there are disagreements on what counts as evidence in the question of Jesus, for example what position one takes on that famed paragraph in Josephus (wholly forged? partly forged? wholly authentic?) or whether the Gospel of Thomas was composed before or after the Synoptics. And so much more. But I say these are themselves all resolvable by Bayesian debate involving the exchange of information by parties actually interested in getting at the truth (and not just defending their a priori assumptions with whatever handy rationalizations). For example, many historians who trust any part of the Josephan testimony are basing that judgment on the Arabic fragment being more original than the Eusebian manuscripts, not being aware that subsequent research confirmed that fragment to be derivative from Eusebius and not closer to the original after all (OHJ, pp. 336-37). With that new information, if they are reasonable, they will update their probabilities. Likewise when one reads Mark Goodacre’s demonstration that Thomas is actually based on the Synoptics and not independent of them as some have thought.
And though indeed “historians … start their investigations with different prior evaluations of the probabilities of various Jesus hypotheses,” that is precisely the kind of thing an exchange of information among committed Bayesians would resolve, by bringing their evaluations closer together (though their estimates don’t ever have to be the same, though, as I explain in Proving History, a point Tucker acknowledges; but they will converge on a general region of the probability space). This is partly the objective of Chapters 3 through 6 of OHJ: to provide the information that should change the estimates historians place on various competing hypotheses, by dispelling inaccurate assumptions and gaps in knowledge they had previously been relying on. And finally, all semantic disagreements are automatically resolved with the exchange of information. Once everyone clarifies what they mean by “Jesus existed,” there will no longer be any confusion. This is partly the role of Chapter 2 in OHJ.
But I even suspect Tucker would agree with me on these points.
Did Nietzsche Have Syphilis?
The first part of Tucker’s review is just philosophical background and a description of my approaches and aims (pp. 129-33). He gets to the meat of it on page 133. But the preliminary section is a valuable read. Although, readers of his review might be confused, as I was, by one of his statements regarding the role of likelihoods in Bayesian reasoning about history. Tucker says that “if the evidence is the dementia of Nietzsche, there is no world where the hypothesis that he suffered from syphilis is not true; its posterior probability is 1.”
I had to ask Tucker about this because it made no sense to me at first. He confirmed in personal correspondence that he meant what I would have called syphilitic dementia (i.e. a specific type of dementia with symptoms unique to syphilis), and not merely a general medical diagnosis of dementia. Indeed that would be tautologically only indicative of syphilis. I’m not so sure there is such a thing (I think diagnosing syphilis requires more tests than that). But certainly if there is, then his statement makes sense. Indeed it’s simply then true by definition.
But if we want to help people grasp Bayesian reasoning, I would have used the general diagnosis of dementia. Because there are other causes of dementia (in the generic sense) than syphilis, so there would be some worlds where Nietzsche both had dementia and not syphilis. And Bayesian reasoning involves estimating or finding out what the frequency is of such possible worlds, relative to worlds where he has both. I would also have noted that the “evidence” properly speaking is not really “the dementia” of Nietzsche, since, as Tucker points out with a separate analogy to the reign Louis Philippe I, the dementia of Nietzsche can no longer be observed (as Tucker puts it, it “cannot be deduced from universal truths nor be perceived,” therefore it has to be inferred). Even when it could, dementia is still only a diagnosis, which is an inference from observations, not something we directly observe. Rather, the evidence consists only of records of things Nietzsche said and did that are indicative of dementia.
And that adds another question to our analysis. Because there are also possible worlds where all those indications exist on the record but Nietzsche did not in fact suffer from dementia! Those worlds might be few, but they would correspond to the frequency of the misdiagnosis of dementia on the same observations or reports. Especially since not all reports are necessarily honest, nor always accurate even when honest, and even when accurate and honest, could be misunderstandings or have other causes, however unlikely. For instance, Nietzsche could have been pretending to have dementia.
I bring this up because it illustrates yet another value to thinking this way about history. Analyzing the problem in Bayesian terms—and in Tucker’s way, of considering and counting up possible worlds, since that is what we are trying to find out when doing history: which of those possible worlds it is most likely the case we are actually in—forces us to realize there are alternative explanations of everything that we must give an account for, either to provide a valid reason why we can ignore them, or to determine their share of the probability space and what effects that has on everything else we want to know.
And here we have two probability spaces to worry about: that of what caused the evidence of dementia (Nietzsche pretending? Witnesses of it lying? His actually having dementia?); and that of what his having dementia entails once we conclude that he had it (Nietzsche having syphilis, or any of at least thirteen other known ailments known to cause dementia, or something even unknown to us, since medical science has surely not discovered everything already). This corresponds to my demarcation in Proving History (p. 48) of two different types of questions we ask in history: What happened? (Did Nietzsche really suffer from dementia?) And: Why did it happen? (Was it syphilis?)
Make note of that. Because it becomes important below.
Major Criticism Alpha: Defining the Hypotheses
But that had nothing to do with Tucker’s critique of Proving History. The first major criticism Tucker has of the actual book is his concern for the need to carefully define the hypotheses to be compared in respect to whether “Jesus existed” is true or false. Tucker says “Carrier does not define explicitly the question he is investigating,” meaning the question of the existence of Jesus, but that’s both besides the point and partly incorrect. A reader familiar with my book might wonder what he means, the question I am investigating in Proving History, or the question I investigate in its sequel, OHJ? He confirmed to me in personal correspondence he meant the latter.
But that’s not the question investigated in Proving History. This is clearly stated on its first page (p. 7, first paragraph of the Preface):
This book is the first of two volumes examining a daring question: whether Jesus … never really existed as a historical person. … The present book does not test that claim (as the next volume, On the Historicity of Jesus Christ, will), but rather begins the inquiry by resolving the central problem of method: How does one test a claim like that?
Accordingly, I define the hypotheses to be tested about Jesus in the second book, On the Historicity of Jesus. There, one whole chapter each is devoted to defining the two hypothesis to be compared: minimal historicity and minimal non-historicity. This is explained again in the closing pages of Chapter 5 of PH (pp. 204-05), where I actually talk about a general approach to defining those hypotheses (so his saying I didn’t is partly incorrect…I say partly because it’s true that in PH I only make a general point about how to do, and do not lay out what results when you do). So here Tucker is criticizing Proving History for not covering what it explicitly states it does not cover and was not intended to cover.
And yet, what he then discusses as his suggested path toward completing this step is precisely what I do in On the Historicity of Jesus! So in fact, Tucker has ended up reinforcing and supporting what I actually have argued. So this isn’t really a criticism after all.
This makes more sense of his complaint that I don’t mention what he calls a “Kripkean” definition of a historical Jesus. Obviously, because defining the historical Jesus is not the function of Proving History. When I do get to that step, in OHJ (Chapter 2), I fully incorporate exactly what Tucker recommends: a minimal criteria-based description of a man who lies in direct causal relation to subsequent Christian claims. I confirmed this is what he meant in personal correspondence, where Tucker says what we need is a hypothesized man “who initiated an uninterrupted causal chain that led to” the Gospels, and therefore “any other properties, including crucifixion and resurrection, are inessential,” meaning, it wouldn’t matter if he wasn’t really crucified or resurrected.
I concur. Criterion 2 of my definition in OHJ (p. 34) reads: “…the same Jesus who was claimed by some of his followers to have been executed by the Jewish or Roman authorities.” If they merely claimed this, even if it was false, I conclude we still have a historical Jesus in a meaningful sense. “By contrast,” Tucker explains, “Santa Klaus was invented in Germany in the 19th century by the owner of a department store” and “there is no causal chain that maintains reference to connect it with Saint Nicholas” of the Middle Ages on whom Santa Claus was based. Analogously, if the story of Jesus at its inception was wholly based on Moses or Socrates or Julius Caesar, there would not be any meaningful sense in which Jesus “existed,” masked man fallacies aside. The key step is uninterrupted causal connection to (in effect) the masked man, so that no matter who he turns out to be, he at least turns out to be a person who actually did something relevant that sparked his worship. Everything else is moot. (To the mere question of historicity, that is.)
Indeed, in Chapter 2 of OHJ I define a minimal historical Jesus as not even someone requiring the name Jesus, as that could have been assigned later to the actual person who started the religion by making certain claims and getting killed and then being believed to have been resurrected. It’s those events starting the religion that require a historical person for the name to refer to. The name is actually incidental, if for example the apostles simply chose afterward to call their deceased leader “our Savior and Messiah,” since Jesus Christ literally means “Savior [and] Messiah.”
The alternative I explore (and explain in Chapter 3 of OHJ as the most defensible alternative, among many others I conclude aren’t defensible at all) is that the person claimed to have been crucified and resurrected (e.g. in the letters of Paul) was not a historical person at all, but a supernatural being who was believed to have been crucified far up in the sky. Which of course they believed to be a historical event, but that we now of course would not believe was, but rather something someone only imagined or claimed happened. I then show how these two hypotheses exhaust enough of the probability space that we can ignore everything else.
So, on the one hand, this is all not discussed in PH because PH is not about these hypotheses. It’s about the method by which one would build and test any hypotheses, whatever they end up being (as explained on pp. 7 and 204-05). While, on the other hand, where I do discuss the hypotheses to be tested in this case (in OHJ, the sequel to PH), I actually employ an approach in full agreement with what Tucker proposes. So in what looks like a criticism, in fact Tucker anticipates and validates exactly how I go on to deploy Bayes’ Theorem to the question of the historical existence of Jesus.
I mentioned already Tucker’s mistaken notion that I over-attend to “literalism” as a competing hypothesis (a hypothesis he regards as too ridiculous for any historian’s attention), and how that was simply a misreading of my choice of example in Chapter 3. Tucker defines such hypotheses helpfully as “Bedelian” (as opposed to Kripkean and Humean, and his discussion of these options is interesting). This is thus again validating, since I explicitly rule out in OHJ any concern for Bedelian hypotheses. I there assume all objective experts already agree no such Jesus existed—the Jesus of the Gospels read literally. The only interesting question, I declare in OHJ, is whether there was some other more mundane man behind that myth. I thus wholeheartedly agree with Tucker’s beautiful statement, “I do not think it is necessary to resort to heavy Bayesian artillery to demolish literalist soft targets.” Amen.
Another validating point Tucker makes is when he points out, again thinking this is a criticism, that “even if Carrier can prove that the historicity of Jesus has a low posterior probability, it may still be rational to uphold it for lack of a more probable alternative.” In fact that is why I address in OHJ the definite need for a better hypothesis, and wrote that book precisely to explore if in fact we have one; otherwise historicity prevails. In Chapter 3 of OHJ I even mention some of the alternative hypotheses that are so implausible that they could not prevail over even a problematic historicity hypothesis (unless we observed the world to operate very differently than in fact we do, but there is no need to entertain that counterfactual).
Strictly speaking, of course, Tucker’s statement is incorrect. Since Bayesian analysis always calculates only a relative probability, it can never be the case that historicity has “a low posterior probability” and is still the more probable option. Unless by “a low posterior probability” Tucker means to include conclusions like “there is a 60% chance Jesus existed.” Then it’s tautologically the case that we should conclude that more probably than not he existed. But Tucker might have meant that, let’s say, though “there is a 30% chance Jesus existed,” that could still be higher than the probability of any alternative, and so we should conclude he probably existed. But that’s not quite correct either (as I explain in OHJ, p. 247, w. n. 18). If there was a 30% chance Alexander the Great died from murder, while every alternative natural cause of his death had only a 5% chance of being the case, we still should conclude that Alexander more probably died of natural causes, we just don’t know which one. Because the probability that he died “of some natural cause” would be 70%.
Even then, Tucker’s point can be reframed to mean that, if we have no alternative to historicity that has a significant prior probability or that can explain the evidence at all well, then no matter how unlikely historicity may be in absolute terms, we will find that its “posterior probability” will actually be high (just as for my example of theft vs. resurrection in explaining the empty tomb in my last post about this). And that’s precisely what I allow in OHJ—the only place where I actually explore the hypothesis space regarding the historicity of Jesus.
Either way, Tucker is saying we need a good explanation of the evidence other than historicity, if we are to ever make a respectable case against historicity. And that’s precisely what I argue in OHJ.
Major Criticism Beta: Literalist vs. Literary Hypotheses
As I noted already, Tucker mistook my use of a literalist example in Chapter 3 as somehow having to do with the historicity of Jesus, and as representative of that debate—when in fact my discussion of mimesis criteria is closer to what I think is representative of the debate: finding reasons for the authors of the texts we have to have written what they did, and comparing those hypotheses (did Nietzsche even have dementia?), before we can take the next step of asking where the evidence leads (did Nietzsche actually have syphilis?).
Here is where Tucker says, “Carrier seems to accept the literalist perception that their hypotheses are identical with the propositional content of evidence, rather than an anachronistic, decontextualized, and historically insensitive misinterpretation of it.” Tucker states that as a criticism. Yet in fact it’s exactly what I myself have said. I’ve always agreed with him: literal readings of the Gospels (as straightforward recollections of events) are obviously anachronistic, decontextualized, and historically insensitive misinterpretations of what the authors of the Gospels are even doing.
Hence I only used the example of a literal reading of the sun going out at Christ’s death as a foil in Chapter 3 to explain how Bayesian reasoning works. I don’t connect that debate to the historicity of Jesus—I do not say that if the event was made up, then Jesus did not exist. Yet I think Tucker’s discussion of it could confuse readers into thinking that’s how I used it. More importantly, though, I actually compare the literalism of the eclipse claim with the alternative Tucker himself mentions. Because his alternative, that this was a figurative use of storytelling instead of a literal one, would certainly be one type of “Christians just made that up” hypothesis that I define as the alternative to literalism in PH (p. 44).
In Proving History I am most explicit about this when I point out that “we must first ascertain if our author is just reporting the facts as witnessed by or told to him (or what he purports to be those facts) or constructing a story for some purpose other than making a record of what happened (such as producing myth, fiction, or any other form of storytelling whose aims are more subtle than superficial veracity)” (p. 160). This certainly acknowledges and includes what Tucker means by the possible nonliteral intent of the sun’s darkness in Mark’s Gospel. So I did not exclude such hypotheses even in PH. Though I do explore them in much more detail in OHJ (especially in Chapters 9 and 10), Tucker’s readers might still be misled into thinking this possibility wasn’t mentioned even in PH. But it was.
Nevertheless, in actual fact, the earliest Christians we have who commented on this story in the Synoptics actually did read the text literally. They even attempted to prove the event happened (see note 2 on p. 299 of PH, which contains all the scholarship on that point). That’s why I mention Thallus (on p. 44). As early as c. 200 A.D. he was being cited by Christians as confirming the event happened. He did not; but what matters for the present point, is that they thought he had (see my journal article on Thallus reproduced in Hitler Homer Bible Christ). Though I do believe there were Christians who read it figuratively (and I believe Mark intended his whole Gospel to be read figuratively, at least by insiders), their commentaries were destroyed by the Christians who gave us the version of Christianity we know now—yet those commentaries were evidently used by polemical heresiologists like Irenaeus, so we know they existed. Which was earlier? The literal or the figurative reading? It’s hard to be certain, because the historical record has been heavily doctored by the literalists.
I discuss all of this in Proving History‘s sequel, OHJ (e.g. Chapters 8.4, 8.12, 7.7, and Element 21 in Chapter 4). Including a full discussion of the allegorical role of myth writing (Chapter 10.2 and Elements 13 and 14 in Chapter 4), and whether and to what extent the Gospel authors did or did not intend what they said to be taken literally (in Chapter 10; the answer is not a simple one). This is foreshadowed by my treatment of mythmaking throughout PH, which does indeed illustrate allegorical reasoning as a motive (e.g. pp. 31, 132, 142, 153-54, 155, 160, 162-63, 191). Tucker seems not to have noticed this. So when he criticizes Proving History for not considering this hypothesis, a hypotheses he rightly thinks too credible to ignore, he is again not actually criticizing Proving History. For it already does consider that hypothesis (the section on mimesis that impressed him is an example of that). And that hypothesis becomes a very important one in OHJ. Just as Tucker would have wanted. So he is, again, actually validating my project.
Major Criticism Gamma: On Assigning Probabilities
There is one point where I think Tucker outright misread the text of Proving History. It’s another example of what he says the book should have said being what the book actually said. Tucker refers to that same page in PH I just referenced above (page 160) and says that there, “Carrier assigns to evidence (the Gospels) a prior probability, rather than likelihood or expectancy, as the Bayesian theorem requires.” But that’s not true. On that page I am clearly talking about the prior probability of competing hypotheses regarding the causes of what’s in the Gospels, not of the Gospels themselves.
What I discuss there is how we might assign prior probabilities to the contents of the Gospels being truth claims or fictions. And I explicitly frame this as comparing hypotheses of author intent:
[Y]ou must ascertain how frequently that author, in that document, particularly in analogous literary and narrative contexts, fabricates data rather than reports what he learned from reliable historical sources. If there are many instances of doubtless fabrication or the use of unreliable sources or methods, then the prior probability that this author reports data because it’s true (rather than only because it suits his story, or his source’s story) is low even if he purports to be recording what happened. (p. 161)
Clearly I am not here assigning priors to “the Gospels,” but assigning priors to the competing causes of that evidence (the contents of the Gospels). In other words, how likely is it that the next story a Gospel tells us is historically true, given that all the stories it has told up til now have been fictions? Once we resolve that question, then we ask whether that next-to-be-examined story in that Gospel is expected on each hypothesis (that the author is recording what happened, or that the author is doing something else, like writing figuratively or in parable). That’s how we determine whether the content of Gospels are historical evidence: by determining how likely it is that what they contain was caused by historical events (being, as Tucker said before, at the end of a continuous chain of causes beginning with an actual historical event involving Jesus) or by the author’s other motives (like writing figuratively or in parable). Or if it’s indeterminate (which in Chapter 10 of OHJ is what I find to be the case most of the time). That’s where the content of the Gospels becomes e; the cause of that content remains h from beginning to end—and it’s the cause of that content that is being discussed on pages 160-61.
And when it comes to generating a prior probability that any given story in the Gospels is a record of historical facts or was written fictively, I explicitly describe how this could be a prior probability generated as the posterior probability of previous runs of the equation, as we go through a Gospel pericope by pericope (Proving History, p. 168). The end result of that process could instead be estimated—we could just approximate what such a series of runs would get us, and thus skip to our estimated prior—but as I explain, you needn’t estimate like that if you are concerned about over- or under-estimating. You can instead construct the prior by starting with a neutral (50/50) prior and going through the text claim by claim, updating your prior with every run of the equation.
Although in broader context, genre also must affect these priors. If the Gospels correspond to a type of literature that is usually fake—like, say, in the set of all sacred stories written by religious fanatics about worshiped deities—then that generates a prior probability that carries over to the Gospels before we even start running numbers only on them. Just like the “urban legend” genre today, whose being caused by true events we immediately assign a low prior to as soon as we encounter an instantiation of it. But again, in PH, I say one needn’t even do that, if one wants to treat the Gospels in isolation with a starting unbiased prior. Although, IMO, that is actually introducing bias, not removing it. Because disregarding the genre and historical-literary context of the Gospels is removing data from b, our background evidence, which will not give us correct results. But I don’t even bother making that point in PH.
But note again, that what Tucker thinks is a criticism of PH, is actually an endorsement of PH. He says the problem would be fixed by treating the Gospels as evidence for a hypothesis to explain. And that’s precisely what I do. He is thus in fact endorsing my actual procedure.
There is one point in this discussion where I disagree with what Tucker appears to say, however. And that’s where he says “the question whether this evidence [meaning the contents of the Gospels] is historiography or myth imposes on the evidence categories foreign to it,” meaning that the authors may have grasped no distinction between them. That’s not true (ancient authors, even fanatical Christians, certainly understood the distinction; I quote many of them in OHJ explicitly discussing their awareness of it). But it wouldn’t matter even if it was, since what we want to know today is whether a given tale in a Gospel is evidence of an event in the life of the man we mean by Jesus, or evidence of some other authorial intent, and that distinction is ontologically real—it does not require the author himself to have understood it. It remains a question we today must answer, before we can even call the Gospels “evidence,” of anything—even if our answer is “we don’t know” (and thus 50/50 truth or fiction, fact or figure).
So when Tucker says that, instead, “The relevant question for a modern historian is what information is nested in the evidence and how to extract it,” he is again saying exactly what I actually said in PH: the information we need to extract from the evidence is whether the evidence was caused (ultimately) by an event that actually occurred resembling the tale now being told (like the tales of Julius Caesar crossing the Rubicon), or whether it was caused (ultimately) by someone making it up (like the tales of Santa Claus). And that’s precisely what I am talking about on pages 160 through 168: how to extract that information; and how to know when we can’t.
Major Criticism Delta: Multiple Attestation vs. Independent Testimonies
Tucker’s review I think is very confusing on another particular point. I was able to clear up some of what he meant by personal correspondence, but in the form still published, I don’t think the average reader will get it. And that relates to what he says about the power of independent testimonies. Which adds a valuable piece to what Proving History should have covered, and only didn’t due to its irrelevance in Jesus studies. He seems to be saying something in defense of applying the Criterion of Multiple Attestation to Jesus, but not really.
Tucker is an expert on the analytical logic of independent testimonies (e.g. “The Generation of Knowledge from Multiple Testimonies,” Social Epistemology 2015). He has established the mathematical logic of how and why independent testimonies can greatly increase the reliability of information testified to. But what he means by “independent testimonies” are actually independent sources of information, e.g. three separate witnesses to an event who do not speak to each other to coordinate their testimony. This is not what Jesus historians mean by “multiple attestation” in the Criterion of Multiple Attestation, which I address in detail in Proving History (pp. 172-75). They mean merely having multiple sources that might be independent of each other. What I (and other scholars I cite) point out is that this cannot generate a higher probability for a multiply attested claim unless it can be established that the testimonies are independent. And that means independent in the sense Tucker means (e.g. they can’t all stem from a single earlier fabricated source either, but must go back to actual independent witnesses of events). But that is not the case for any surviving testimony about Jesus.
The Gospels were all written with knowledge of both Mark and the letters of Paul. There is no independence here. Even the first of the Gospels, Mark, of which all the others are redactions, appears well aware of Paul. And even if there was a Q document (from which both Matthew and Luke theoretically copied), it cannot be established that it was independent of either Mark or Paul either—indeed, any attempt to argue Q was not known to or used by Mark or did not know or use Mark is fallaciously circular; and this becomes patently obvious the more you realize that Matthew is in fact Q, a conclusion even the fiercest remaining defender of Q, John Kloppenborg, had to concede, finally admitting that Q’s existence is only “at least as plausible” as its absence and that defenders of Q are far too confident in their assertions (OHJ, p. 270 n. 34). And frankly, Kloppenborg is still being wildly generous to Q. The evidence for Q is actually essentially non-existent (OHJ, pp. 269-70, 470-73; Loren Rosson concurs). You begin to notice this when you start discovering how many lies you’ve been told about Q (that it only contains sayings and not narratives—lie; that it only contains Jesus material—lie; that it did not contain nativity or crucifixion scene material—lie; that Luke does not use Markan material as transformed by Matthew—lie; that it was composed before the Jewish War or in a language other than Greek—lie; and so on). But even without having to conclude that, we still cannot establish Q is independent of Mark and Paul (indeed Q material appears well aware of Paul’s letters)—or independent of any fabricator who may have preceded and been used by Mark.
This is actually well enough known to experts in the field. It is generally recognized as lying if someone claims that the Gospels are independent of each other (I’m looking at you, Bart Ehrman). Which is why the Criterion of Multiple Attestation fails in Jesus studies. Even if it would be of use in other subjects or on other questions. So when Tucker says, “There is probably more rational historiographical consensus on inferences from the Synoptic Gospels than Carrier admits, for example, that Matthew and Luke were affected by Mark and by a lost source, Q,” I think it would be fair to remind you that he disclaimed “any expertise” in that subject. Certainly we all agree Matthew and Luke are redacting Mark (so no independence there). But there is no rational consensus on their having also redacted “Q.” Really, Luke just redacted Matthew. And again, even if he used a Q instead of Matthew, we can’t establish Q as an independent source either. (Notably Tucker excludes John as evidence altogether, as most experts now would agree he should; but it’s also now the consensus of specialists in John that John used the Synoptics as a source as well. So no independence even there.)
Tucker seems a little uncomfortable with this realization. He knows the value of multiple independent testimonies. So he wants to apply that knowledge here. But what we are telling him is that the evidence doesn’t give him any foothold for it. In a note of mild despair, he aptly observes that, so far as he can tell, “The problem with the Synoptic Gospels as evidence for a historical Jesus from a Bayesian perspective is that the evidence that coheres does not seem to be independent, whereas the evidence that is independent does not seem to cohere.” Right. Thus, no independent testimonies to be had in any of this.
Nevertheless, Tucker is right to say that “Carrier does not analyze in Bayesian terms what is in my opinion the main method for Bayesian” evidentialism in history (or intelligence, criminal forensics, and other testimony-reliant fields—or even, we should note, science, since the requirement of multiple replications of an experiment produces essentially the same mathematical effect). And that’s looking for a cumulative effect of independent testimonies. I do actually make this point in a more general sense (in PH, p. 297, n. 5; plus a reference to the idea atop p. 43), but not in a more comprehensive way as I should have. Just because there are no relevantly independent sources for any claim about Jesus, does not mean this isn’t the case for other questions in history, and I did intend PH to be generally applicable. So I think Tucker is right to find fault in this. A small section on it would have been handy.
Not only could I have treated the mathematical point (and fruitfully cited Tucker’s paper in Social Epistemology), but I could have addressed the methodological challenges as well. For example, what exactly does it take to have an independent testimony? (What does independence mean.) And what does it take to know that’s what you have? (Since that actually affects your probabilities.) These two questions are not always easy to answer.
Consider all we’ve learned about the unreliability of eyewitness testimony (such that it is now a part of police standards to treat eyewitness testimony as among the least reliable evidences of a crime). As just one example: witnesses who are allowed to share their accounts before being interviewed by police become highly unreliable, as the memories of one witness contaminate the others, and their continued conversation even creates new confabulated memories. Police procedure now is to separate witnesses to an incident as soon as possible to minimize memory contamination. But it’s almost impossible to do this—rarely do you have witnesses who remain informationally isolated from each other until first being interviewed or recording their testimony in some more permanent form (which is highly recommended you do if you want to ensure the stable reliability of your own testimony to an event—this is what experienced journalists now do, even when they check their reports among each other). When it comes to the transmission of written texts over decades this problem is impossibly compounded. There is no possible way Mark, writing forty years after the events he supposedly relates, could be composing his stories independently of all oral lore and all previous writings by and among Christians. It’s just as impossible to know that any source that he may have had was independent of any of his other sources. Even Mark could not possibly know that.
So Tucker’s method of accumulating independent testimonies simply cannot be applied to any of the materials we have for Jesus. And it’s a challenge to find any cases in history where it can be applied. Until one gets closer and closer to modernity, where documentation increases exponentially, and you actually can have, for example, state officials recording an event in separate official archives completely independently of each other. The basic idea of Tucker’s approach is that “when priors are sufficiently low, the more testimonies there are, the lower is their expectedness” (“Generation of Knowledge,” p. 5). That is, you want testimony that is very a priori unexpected (like “someone just shot the President of the United States in a theater,” something that does not happen every day, and is especially specific even in the location). Then, when two independent witnesses, who have not collaborated or contaminated each other, both report those same two details, the odds of this happening by chance are extremely low. Their agreement is therefore very strong evidence for the claim. And as witnesses accumulate, even more so. I could have made this point, for example, in my discussion of what evidence it would take to convince us the sun went out for three hours in 1983 (PH, pp. 56-58). Tucker has a lot else of value to say on the subject in his Social Epistemology paper.
Finally, there were three points where Tucker inadvertently mischaracterized my discussions of method.
First, I do not believe it was accurate of him to say I “just dismiss[…] the similarities between the Synoptic Gospels” and thus their value as evidence, since, on the one hand, I routinely grant what all experts agree, that they are copying each other (thus explaining their similarities); and, on the other hand, when I dismiss their evidence as reliable records, whether because or or in spite of this, every single time I give arguments for that dismissal. In fact that’s what the whole of Chapter 5 consists of doing. And that negates the applicability of his adverb “just” in this sentence. So I fear he is misleading his readers into thinking I didn’t actually make a case every single time this comes up, and that I didn’t cite several experts doing the same nearly every single time. His expression makes it sound like I idiosyncratically do this (when in fact it’s several members of an expert community that are doing it) and that I do it without any basis. But it’s one thing to argue the basis is unsound; it’s another to imply there was no basis given that one must assess. Tucker gives no argument against the basis for any of my dismissals. He thus hasn’t actually voiced any identifiable criticism against any instance of my concluding we can’t extract history from the Gospels.
Second, when Tucker says “Carrier repeatedly asks rhetorically why would the redactors allow information” into their text that they didn’t like, and proposes his own “answer,” the way he phrases it is ambiguous as to whether he means that’s the answer I give (as in fact I do: pp. 158-69, esp. 159, 163, and 167) or that I failed to give (which, per the material I just cited, would be untrue). In personal correspondence he confirmed it’s the latter. Which means, he is simply not correctly describing the book. The answer he gives, that “some stories or sayings could have been too well-known among the intended audience of the texts” to alter or doctor out, is actually exactly what I myself say:
The answer to that question is … usually that the most likely reason [an embarrassing] statement was included was that it was true but this author couldn’t omit it or change it despite having ample reason to on account of its embarrassing nature. But that requires explaining why that author could not omit it or even change it (and why no one else could, in all the decades before). Because some explanations of that odd fact will be far more far-fetched than others … and because you are obligated to prove that that explanation was in fact guiding that author’s construction of the text … and because some explanations will provide just as much reason to invent a fact as to have reported it if true … (p. 159)
[For example] … when a neutral or hostile tradition exists that is highly reliable and well known, such that a friendly source cannot easily ignore it or change it. Thus, if we can show an embarrassing claim had already been reliably and widely established by a neutral or hostile source by the time a friendly author wrote, then we can reverse this probability, because the frequency of those claims that will be true will be much greater. (p. 163)
[And yet on] the hypothesis that an author reported something he didn’t want to report because he was compelled to for some reason, odds are the author would express this, or in some fashion explicitly defend his interests against the implications of the embarrassment, which means if this didn’t happen, then the evidence is not as expected on our hypothesis. (p. 167)
In other words, if there is no evidence an author is embarrassed by what he is saying, we are not warranted in concluding he was (I make an expansive case for that conclusion, based on the concurring agreement of many experts: pp. 126-38). But certainly, when we have that evidence, then we can start to work the logic. The problem is, for Jesus, we can’t work it back to anything usable. For example, Matthew may show evidence of being embarrassed by Mark’s empty tomb and baptism stories, but Mark does not. So all we can prove is that Matthew was embarrassed by Mark. Not that Mark was embarrassed by the empty tomb and baptism stories. And since we can’t show that Mark didn’t invent them (in fact, his lateness in the timeline entails he almost certainly did, since embarrassment was only triggered after he wrote them down: pp. 126-28), that doesn’t get us to any relevant historical event. As I explain in PH, this isn’t the case in every historical investigation (p. 158). It’s just the case for Jesus. So the Criterion of Embarrassment may have its uses (pp. 158-69). But it has no applicable uses for the study of Jesus (pp. 138-57).
So once again, Tucker offers as a criticism, what is actually something my book already says. He thus vindicates my own arguments and procedure. As to the case I made for the inapplicability of this criterion to Jesus, Tucker doesn’t say anything specific. He simply reiterates a possibility, one that I already explicitly acknowledged and refuted. He does not address any component of my refutation.
Third, Tucker writes something very perplexing about how we can use average rates-of-change to estimate the dates of speciation events for stories, a principle that “underlies historical philology and evolutionary biology.” He says that my arguments (which happen to be the arguments of several experts I cited, a fact he neglects to mention) against applying something like this principle to extract historical facts from the Gospels (PH, 183-84) somehow contradicts the principle itself. But I couldn’t fathom what he meant. Nowhere in PH do I question the use of a statistical average rate of textual distortion to date divergences. All I conclude is that even knowing what that average rate was does not tell us which stories are true. All it can tell us is when (roughly) they diverged from earlier versions. But that’s not pertinent to the veracity of the story itself. Knowing the date of the earliest version of a story is not helpful to determining if that story is true when that date is already decades after the event it purports to describe. I even make this distinction explicit on page 184. So his criticism is bafflingly off the mark here.
I also point out there, citing experts on the fact, that deciding which textual variant (which “speciation event”) is the earlier is often not possible. A fact Tucker also does not address. So not only can’t we tell the chronology of speciation, but even if we could “stochastically” date one of these speciation events, that gets us no useful information about whether the story originates in a historical event rather than an author’s imagination decades later. Tucker seems to be confusing what I need to do. I need to ascertain the relative probability of two hypotheses (“caused by a historical event involving Jesus” and “caused by a subsequent author’s imagination”), not ascertain the reliability of subsequently transmitting the resulting story. Even a miraculously 100% reliable transmission mechanism would have no bearing on the question of which hypothesis is true regarding what caused the story in the first place.
An average rate of speciation in storytelling could perhaps be useful if we could establish chronological order (we usually can’t) and also establish a version of the story as being written within a few years of Jesus’s life. But I explain why neither is possible (PH, pp. 183-84), and my explanation is based on bona fide experts making this same point whom I actually cite (Porter, Avalos, Bird, and Goodacre: PH, p. 322, nn. 110 and 111). The fact is, no instance of this occurs in the case of Jesus, so the principle can’t be applied. Even when you might think it can (e.g. the correspondences between Paul’s and Mark’s account of the eucharist event: OHJ, Chapter 11.7), it collapses on examination. Because Paul actually tells us the source for his story: a hallucination, 1 Cor. 11:23—so much for getting to a real historical Jesus. Mark simply historically embellished an account of a vision, originally related by Paul.
One trivial point to close with is that when Tucker says of my discussion of “general” and “particular” claims, he “would rephrase that as tokens rather than types, terminology that Carrier is not familiar with,” that’s an unwarranted inference. I actually do know those terms. I chose not to use such obscure terminology in PH because I was aiming to write for as wide an audience as possible. This is explicitly stated on page 9, where I specifically mention I shall avoid unfamiliar terminology as much as possible; and thus whenever I needed to say something, I chose the most widely familiar words available. So that’s another criticism off the mark.
Apart from those misreadings of the text, Tucker’s review is uniformly informative. In fact it’s a valuable adjunct to Proving History, and advances the cause of getting historians to see the importance of “com[ing] to terms with the extensive Bayesian literature” and “what it means for historiography.”
Even when he gets the book wrong, what he says in an attempt to correct it is correct! It just happens to always be what the book itself already says. So even when he criticizes the book, he is actually unknowingly praising and validating it. This is the case even with his point about using speciation rates—though completely irrelevant to anything argued in PH, it is nevertheless a correct description of a usable method. Just not one of any use for Jesus.
So I have to count this review as a uniform win for my project. Tucker says nothing that contravenes anything argued in Proving History or On the Historicity of Jesus. Everything he says actually in fact verifies them. In all he voiced only one genuine criticism, which is that I could have added a section assisting historians with cases where multiple independent testimonies actually do exist (and even how to identify them). And with that I agree.