John Earman is no doubt a brilliant philosopher and, for all I know, a good person. But when it comes to talking about David Hume on the internet, he is my sworn enemy.
Of course, I’m joking! But my frustration is 100% serious. Dr. Earman wrote a book over 25 years ago aiming to swiftly refute David Hume’s argument on miracles. If you’re a student of Hume studies, you realize both how influential Earman is and how bad that is for the public understanding of Hume, his argument, and his philosophy.
Earman concluded that:
“Of Miracles” is an abject failure. It is not simply that Hume’s essay does not achieve its goals, but that his goals are ambiguous and confused. Most of Hume’s considerations are unoriginal, warmed over versions of arguments that are found in the writings of his predecessors and contemporaries. And the parts of “Of Miracles” that set Hume apart do not stand up to scrutiny. Worse still, the essay reveals the weakness and the poverty of Hume’s own account of induction and probabilistic reasoning. And to cap it all off, the essay represents the kind of overreaching that gives philosophy a bad name.” (Earman 3)
This is really damning stuff! But unfortunately for Earman, he got Hume all wrong. Let’s correct the record.
Let’s Survey Popular Discourse On This Argument
We’ll get into the details of why Earman was wrong in a moment, but for now, let’s take a survey of popular philosophy discourse from Hume’s critics:
To start, the apologist that popularized Earman as a trump card to Hume was William Lane Craig, who has been name dropping Earman for decades in his various speeches. You can find one example here:
If I were a polemical person, I would say “Jesus Christ Bill, you’re literally lying to people in the first minute of this lecture! There is no general recognition that his argument is an abject failure. You made it up!”
…but I’m going to keep my cool here. I will invite anyone to actually look into this claim and see if there’s any data behind it outside of philosophers of religion, who are more likely to be theist and Christian.
I mention William Lane Craig as one of the main proliferators of the Earman meme because, quite frankly, most of the engagement I see with either Hume or Earman is repeating Craig’s talking points or just vaguely gesturing toward Earman. That is, it seems that apologists believe Earman rebutted Hume merely by insisting it so.
Recently, this culminated in Princeton PhD Noah McKay saying triumphalist things against Hume, in which he directly cites Earman. Another philosophy of religion popularizer, Amos Wollen, called McKay “incredibly based” for this:
One oddity I’ve noticed is that apologists believe that Earman so thoroughly rebutted Hume, that they see no value in explaining Hume or Earman. This “cite Earman and move on” strategy works only if Earman actually rebutted or debunked Hume. As I will show, he did not. And so, to be persuasive to a skeptical audience, miracle apologists need to return to Hume’s arguments and address them before they progress into the details of any specific miracle.
What Earman Said About Hume
Here are some of Earman’s main charges against Hume. It’s not a comprehensive list, but some of the main points.
Hume’s Straight Rule Of Induction
From Earman:
“So here in a nutshell is Hume’s first argument against miracles. A (Hume) miracle is a violation of a presumptive law of nature. By Hume’s straight rule of induction, experience confers a probability of 1 on a presumptive law. Hence, the probability of a miracle is flatly zero. Very simple. And very crude.” (Earman, 23)
In non-Bayesian language, Earman reads Hume as saying that laws are assumed to be certain, and therefore a miracle is impossible.
The Maxim is Trivial
Earman says Hume’s maxim, which we can find below, is trivial:
“That no testimony is sufficient to establish a miracle, unless the testimony be of such a kind, that its falsehood would be more miraculous, than the fact, which it endeavours to establish.”
Specifically, Earman says:
Most commentators have seen profound wisdom here.
I see only triviality. Suppose that we are in a situation where witnesses have offered testimony t(M) to the occurrence of the miraculous event M. Let E be the record of our past experience in favour of the lawlike generalization to which M is an exception. Then, as good Bayesians, our current degree of belief function should be the conditionalization on t(M)&E of the function P(·) we had before obtaining this evidence. Thus, the relevant probability of the event which the testimony endeavours to establish is P(M1t(M)&E), while the relevant probability of the falsehood of the testimony is P(~M1t(M)&E). To say that the falsehood of the testimony is more miraculous than the event which it endeavours to establish is just to say that the latter probability is smaller than the former, i.e.
P(Mlt(M)&E) > P(- M1t(M)&E) (1)
which is equivalent to P(M1t(M)&E) > 0.5. (I) (2)
On this reading, Hume’s maxim is the correct but unhelpful principle that no testimony is sufficient to establish the credibility of a miracle unless the testimony makes the miracle more likely than not. (97)
Long story short, if we’re presupposing a Pascalian scaling of probability (that is, one where our confidence slides from 0 as impossible to 1 as certain), of course no testimony is sufficient to establish a miracle unless its falsehood is more miraculous (i.e. unlikely) than the event it seeks to establish. If the two propositions are mutually exclusive and encompass the totality of what is possibly the case, the probability (per Earman) has to be above .5 for the miracle. What’s more, we can say that Earman thinks Hume’s maxim is a token one about a single piece of testimonial evidence (as opposed to a type maxim about the entire type of testimony), and so it collapses into triviality.
The Diminution Principle Is Flawed
According to Earman, Hume’s Diminution Principle is:
“The evidence, resulting from the testimony, admits of diminution, greater or less, in proportion as the fact is more or less unusual.”
Earman criticizes Hume, citing Price, saying that “’improbabilities as such do not lessen the capacity of testimony to report the truth” and “that the diminution effect operates through the factors of the intent to deceive and the danger of being deceived, either by others or by oneself.
Finally, he says:
“Hume gets his intended moral only for one order of quantifiers; namely, for any given fallible witness (who may practise deceit or who is subject to misperception, deceit, or self-deception), there is a story so a priori implausible that we should not believe the story if told by that witness. It does not follow that there is a story so a priori implausible (unless, of course, the prior probability is flatly zero) that for any fallible witness, we should not believe the story if told by that witness. To get this latter implication one needs the extra postulate that there are in-principle bounds below which the probability of errors of reporting cannot be reduced. Hume’s cynicism suggests such a postulate, but cynicism is not an argument.” (97)
Put succinctly, Earman is saying that no attested event is so improbable a priori that we could dismiss any witness reporting such a thing. For Earman, Hume does not elaborate on an additional postulate from which we could get to that point.
Multiple Witnesses Could Warrant Belief
Hume argues that no witness testimony could establish a miracle. Earman responds:
“Suppose that there are N fallible witnesses, all of whom testify to the occurrence of an event M. For simplicity suppose that the witnesses are all equally fallible in that for all i =1, 2 , ..., N
P(ti(M)IM&E) = p, P(ti(M)/~M&E) = q (4)
And suppose that conditional on the occurrence (or non-occurrence) of M, the testimonies of the witnesses are independent in the sense that
P(t](M)&t2(M)&’” &tN(M)/-.!:.M&E) = P(t1(M)1-.!:.M&E)xP(t2(M)1 ±M&E)x... xP(tNCM)1 ±M&E) (5)
where ±Mstands for M or-M, and where the choice is made uniformly on both sides of the equality. Then the concurrent testimony of all N witnesses gives a posterior probability
P(M1t1(M)&t2(M)& ... &tNCM)&E) = [ 1 .o]() P(-MlE) q N 1+ P(MlE) P (6)
The witnesses may be very fallible in the sense that q can be as close to 1 as you like. But as long as they are minimally reliable in the sense that p > q, it follows from (6) that the posterior probability of M can be pushed as close to 1 as you like by a sufficiently large cloud of such fallible but minimally reliable witnesses-provided, of course, that P(M1E) >0. Using this result Babbage was able to give some nice examples where just twelve minimally reliable witnesses can push the posterior probability of an initially very improbable event to a respectably high level.
In sum, if we set aside Hume’s wildly optimistic account of induction on which P(M1E) = 0, then he must agree that fallible multiple witnesses can establish the credibility of a miracle, provided that their testimonies are independent in the sense of (5) and that they are minimally reliable in the sense that p >q. (Earman 101)
Note: I admit, I do not follow the math super well because I’m not habituated to reading and writing in Bayes theorem. Nevertheless, I will translate!
What Earman is saying is that as long as there are a large number of witnesses who are each minimally reliable, you can update the posterior probability for an initially very improbable event to a level where you can be confident that it happened. The key point is that these witnesses must be independent of each other.1
Earman continues to point out an inconsistency in Hume, namely that Hume is open to this idea for “secular” miracles, but not for religious ones. To Earman, this is cynicism again and not an argument.
Putting It All Together:
Earman is unimpressed with Hume’s argument against miracles. He essentially repeats the (mistaken) analysis of C.S. Lewis that Hume defines miracles as impossible, albeit using Bayesian language and math.2
He points out that Hume’s maxim is trivial because of course the falsehood of a testimony would have to be more improbable than an attested miracle to establish the miracle. That’s basic Bayesian probability!
The Diminution Principle does not hold because Hume does not make an argument about why we should dismiss any witness testifying to a sufficiently improbable event, and there should not be any sort of event that is a priori so unlikely that it’s dismissed.
And finally, under a Bayesian lens, we have good reason to think a sufficient number of minimally reliable witnesses could establish the credibility of a miracle.
So that’s what Earman has to say. What’s wrong with it?
How Earman Got Hume Wrong
The Straight Rule Isn’t Real
The first thing to understand is that Hume’s “straight rule” is just not real. Earman made it up! No Hume scholar agrees with this interpretation of Hume.
Earman misunderstands Hume’s vocabulary of “proof” and “laws of nature” by reading them as logical categories, when they are instead epistemological. To the extent that we can determine laws of nature, they are matters of fact, not logical necessities. We formulate our understandings of the laws of nature from a posteriori observation, not by extrapolating them in the armchair like the laws of logic.
So when Earman accuses Hume of putting the probability of a miracle occurring as “flatly zero” (i.e. impossible) he is making a category mistake, as doing so would make Hume’s laws of nature a priori logical necessities, not a posteriori matters of fact.
As Vanderburgh points out, there are many textual reasons to doubt the authenticity of Hume’s straight rule. Namely, Earman “quotes four passages from the Enquiry, none of which imply the straight rule when Hume’s epistemology is properly understood.”
Further:
“Hume does not think the probability is 1 that the future will resemble the past. He admits, for example, that it is possible that the sun will not rise tomorrow (EHU 4.2; SBN 25–6), and that bread will not be nourishing the next time we eat it (EHU 4.16; SBN 34). For Hume no amount of past experience could make it that “the probability that all As are Bs is 1.” (Vanderburgh 48)
And so, Earman:
…gets Hume wrong in a fundamental sense: matters of fact can never be certain; moreover, since induction is fallible, Hume would never have intended to suggest that even a very large number of uniform observations produces a probability of 1. This is true even for the exceptionless regularities upon which laws are founded. When Hume says “no room for doubt”...he means that we have a strong psychological tendency to expect that the contrary will not occur. There are no grounds for doubt (an epistemic claim), but this is not the same as saying that the contrary is impossible (a logical claim). (48)
In all, these misunderstandings render many of Earman’s criticisms moot.
Hume’s Probability Is Factual, Not Numerical
According to Hume scholar Dorothy Coleman, Hume’s approach to evidential probability has an entirely different structure and basis than does the mathematical theory of probability Earman employs in his attack.3 From Coleman:
“[W]hat all conceptions of probability have in common is that they provide different criteria for grading degrees of provability, and that degrees of provability allow for two kinds of scales. Pascalian scales take the lower extreme of probability to be disprovability or logical impossibility; the Baconian scale takes the lower extreme to be only non-provability or lack of proof.”
In this way, it’s helpful to look at the Pascalian probability as numerical probability, and the non-Pascalian probability4 as factual probability.
Critics may say that factual probability is an inferior or outdated probability model, but this is far too rash. Per Franklin, factual probability has a long tradition that extends back to ancient Greek and Roman thought, and that these two different kinds of probability have been treated separately throughout most of history. Indeed, factual probability is often more compatible with how we speak about various scientific theories. As Franklin explains:
“The big bang theory of the universe is much more probable, on present evidence, than the steady-state theory. But it is a rare scientist who can be found to say exactly how much more probable—or even approximately how much.”
In this way, people have historically used factual probability when numerical probability has been inappropriate (52). As Vanderburgh explains:
One problem can be teased out this way. Bayes and Price (and, by extension, other Bayesians such as Earman) are forced to make assumptions about the distribution of chances across contrary possible outcomes, assumptions which are rarely if ever justified outside of highly constrained and artificial experimental situations. Bayes, for example, develops his argument in terms of the equal chances of a perfectly round ball coming to rest at any given place on a perfectly flat table. His conclusions do indeed follow for such idealized cases. But, to speak metaphorically, we usually do not have round balls and flat tables, or at least we cannot be sure that we do. The assumption of the equipossibility of contrary outcomes is therefore usually not justified in actual cases. (53-54)
What’s more:
“Having a non-numerical theory of evidential probability does not mean that we must give up all hope of precision and rigor in our assessments of the probability of empirical hypotheses. There are other ways to achieve this: In discussions of temperature, for example, classificatory concepts (hot, warm, very cold, etc.) and comparative concepts (hotter than, etc.) are possible without any numerical scale. Similarly the non-numerical tradition of evidential probability has given us perfectly serviceable classificatory concepts (improbable, probable, highly probable) and comparative concepts (more probable than, etc.). (See Franklin 328.) If this seems insufficient, we should ask whether quantitative theories of probability can really do better: “Has the quantification of probability helped in the evaluation of uncertain evidence in science?” Franklin (369) answers, “In the restricted cases in which statistical tests apply, it has, but for more general theory evaluation, it seems not.” The desire for increased precision is laudable but, as the quotation from Aristotle at the head of this article points out, it can be taken too far” (54)
Further, Vanderburgh makes a compelling case in his book David Hume on Miracles, Evidence, and Probability that Hume was familiar with factual probability, due to his legal and research background, while at the same time not seeing numerical probability as relevant to his argument about miracles. What’s more, Hume never speaks of probabilities in numerical terms. This evidence, along with further discussions forthcoming about Hume’s “weighing” of probability, makes it very likely that Hume’s probability is factual, not numerical.
And so, the mistake Earman and many contemporary critics of Hume make is assuming that Hume is just like them, treating factual and numerical probability in the same way, using Bayes theorem. As a result of this misreading, Earman attributes positions to Hume that Hume would not accept, and so many of his criticisms of Hume collapse.
Hume’s Maxim Is About The Base Rate Fallacy
Now that we have established that Hume’s probability is that of factual probability instead of numerical probability, let’s get into more details about what that entails. This will be pivotal for this section and the next two about the base rate fallacy, the diminution principle, and whether eye-witness accounts could together warrant belief in a miracle.
To review, Hume’s account of probability is about weighing different probabilities, not numerically measuring them. Think of an old fashioned scale, with one side counterweighing the other.
When Hume assesses the probability that the sun will rise tomorrow, he recalls the count of times the sun has risen each morning compared to the count of the time it hasn’t. The probability that has the highest count (the sun rises) has more weight, and thus is the probability we can be most confident in.
This is obviously a much different way than Earman or the Bayesian account of probability and confidence. Hume isn’t dividing the probability of one by two and seeing which hypothesis is measured above .5 (and therefore what we should be confident in), but weighing two probabilities based on the instances they have been credibly observed, instilling confidence in the probability that is most frequently observed.
This kind of thinking is especially useful when we’re comparing the probabilities of two mutually exclusive events with extreme disparity. Specifically, Hume’s maxim illustrates the base rate fallacy. Millican explains:
“Fred wants to know whether he suffers from some genetic condition G which afflicts one person in a million. He has no other evidence either way, but a test is available which seems very reliable, in that whoever is tested, and whether they actually have the condition or not, the chance that the test will give a correct diagnosis is 99.9%, and an incorrect diagnosis only 0.1%...When Fred later leaves the clinic in distress at having tested positive for G, how convinced should he be that he does indeed have that condition?
Most people would, in my experience, judge Fred’s likelihood of having G in this situation to be very high, but in fact the reverse is the case, as Hume would recognize. As Fred stumbles out despondently through the clinic door, Hume might greet him with a consoling comment something like this: Consider whether it be more probable, that this kind of test should be mistaken, or that you should really have condition G(cf. E 10.13).
Given that the test is wrong one time in a thousand, while G afflicts only one person in a million, there is clearly a far greater likelihood of a mistaken test than of Fred’s actually suffering from G. And so a positive test report does relatively little to indicate that he actually has the disease: in fact, it changes the probability from a negligible one in a million to the only slightly more worrying 1 in 1,002. Hume’s maxim, therefore, is entirely correct in this case, and it also gives the correct answer for other relevantly similar cases…
…Hume deserves credit for enunciating a principle which clearly anticipates—by two and a half centuries—the identification of the base rate fallacy by psychologists Amos Tversky and Daniel Kahneman. This is a very common error in human thinking, whereby we naturally find it all too easy to ignore the background “base rate” of some phenomenon when assessing the significance of evidence for it. So on receiving a disappointing test report for condition G, most people would be far more struck by the specific immediacy of that result—and the test’s apparent reliability of 99.9%—than by the memory of the general probability for G of one in a million. They would thus be seriously mistaken, and Hume’s maxim is potentially of considerable value as a vivid reminder of the need to take base rates into account.” (Millican 278)
Now, there are shortcomings to this account of probability that I won’t get too much in detail here.5 I will just acknowledge that maxims are most helpful (and perhaps maybe only helpful) for weighing probabilities with the most egregiously disparate probabilities, such as miracle claims.
I’ve heard philosophers such as Kevin Scharp refer to Hume’s maxim as the confidence argument, and that gets at the heart of the issue. Just as you can be confident you don’t have a highly unlikely disease based on one data point or a few low quality data points, you can also be confident that a miracle did not occur due to just testimony or a few low quality data points.
Sure, a miracle may have happened, just as you may have a disease you tested positive for with an unreliable test. But to be confident of such things, we need high quality evidence which, as a matter of fact (a posteriori), we don’t have with miracle testimony, let alone miracles in general.
Types of Testimony And the Diminution Principle
In the last section, we’ve gone through the “weighing” aspects of Hume’s probability, but to understand how Earman misunderstands the diminution principle, we have to understand Hume’s vocabulary.
Put simply, when we’re weighing probabilities, proofs are strong probabilities with no countervailing evidence, and laws of nature are the strongest proofs. Impossibilities are the opposite of laws of nature, in the sense that they don’t even amount to probabilities, and the weight of the laws of nature gives us moral certainty that they don’t happen.
When we’re weighing the probabilities, laws of nature (often) outweigh proofs (because they are better attested), proofs outweigh probabilities, and probabilities outweigh impossibilities. With increased observations, impossibilities can be upgraded to probabilities, which can be upgraded to proofs, which can be upgraded to laws of nature. Conversely, laws of nature can be downgraded to probabilities, which could conceivably be downgraded to impossibilities as well. What upgrades or downgrades each classification is quality evidence, frequently observed.
But how do we weigh proofs against proofs? In 1761, Hume wrote on this subject in a letter to David Blair saying:
“The proof against a miracle, as it is founded on invariable experience, is of that species or kind of proof, which is full and certain when taken alone, because it implies no doubt, as is the case with all probabilities…but there are degrees of this species, and when a weaker proof is opposed to a stronger, it is overcome.” (Hume 1932, 1.350)
In his criticism of Hume, Earman omitted the final (bolded) clause of the above sentence when discussing the diminution principle. The last clause is essential to understand Hume, as it’s the foundation for understanding what tips the scales when he weighs proof against proof. That is, if we’re weighing two different kinds of proof against each other, all else being equal, the scales would balance, and we would thus suspend judgment on which is more likely (miracle or human error). However, there are different types of proofs of different quality that may lead us to not suspend judgment.
Per Millican:
“Hume’s idea seems to be that different “kinds” of testimony (specified in terms of the character and number of the witnesses, the manner of delivery etc.) carry a different typical probability of truth and falsehood independently of the event reported.” (276)
If you don’t believe Millican, just refresh your memory on Hume’s maxim:
“No testimony is sufficient to establish a miracle, unless the testimony be of such a kind, that its falsehood would be more miraculous than the fact which it endeavours to establish.”
So, contrary to Earman, Hume’s maxim is about types of testimony, not individual testimony (aka token testimony)! In the context of Hume’s weighing of probability, that means that (Per Millican):
“the confidence we place in the testimony…will depend on the extent to which the testimonial proof... over-balances its antagonist. We have “proof against proof,” with the overall credibility given not by either “proof” individually, but by the result of weighing them against each other. Neither side of the contest alone yields the appropriate credibility measure: that comes only from the comparison between them.” (276)
The Complexity of The Diminution Principle
To be completely honest with you, the diminution principle is a little confusing to me when it comes to comparing proofs. I think I understand it, but I think I actually agree with critics such as Earman and Price that it appears somewhat wrong, or at the very least it’s not communicated effectively.
For Hume’s purposes, it doesn’t undermine his argument at all, for reasons I’ll get into in a moment. At the very least, it’s salvageable, as you can find or extrapolate principles from Hume’s probability to communicate why the proof of reliable testimony is outweighed by the experience of a natural laws.
How Earman Got The Principle Wrong
Earman and Price and others misunderstood the diminution principle because they looked at the principle through a Bayesian lens. That is, Hume is not saying what we would call the baseline improbability of an event inherently makes it untrue or impossible. In Hume’s probability, he’s already controlled for the relative likelihood of either hypothesis by classifying them as laws of nature and proofs. And so, Hume believes when we weigh the probabilities and both sides amount to a proof, the kind or type of testimony can be considered as something of a “tie breaker.”
And so, someone’s testimony for a miracle may amount to a proof because they are considered reliable, but because of aspects of their type of testimony (that they are biased toward a religion, that the event is extremely improbable, etc.) that lowers the probability of it being true, leading to the natural law to “destroy” that proof.
The Problem With The Principle
The problem with this principle as Hume communicates it is that it appears like double counting against the miracle, which is probably fallacious.6 After all, in our initial “weighing” of the probabilities, we’ve already taken into consideration possible errors by witnesses and the high probability that the laws of nature aren’t violated. Why would we take those into consideration again? It just does not seem correct.
Salvaging The Principle
The good news is that we don’t need to work that hard to salvage the principle, we just need to communicate it differently. Just as how we weigh probabilities by the frequency they are observed, we should also weigh proofs by the frequency they are observed.
Let’s say you met a stranger and they told you 10 improbable things that nevertheless turned out to be true. After their 10th testament, they also claim that they have the ability to supernaturally levitate. Should you believe them? No!
The reason why is because your observation that people don’t levitate is more frequently and reliably observed than the mere 10 testimonies of the stranger. His reliability may be considered a proof at first, but it’s unwise to believe that stranger (or suspend judgment on him) just because he was truthful about mere 10 improbable (but not impossible) things. That may sound somewhat incredible, but if you consider the law of gravity something you observe every second of your life (so 3600 observations an hour), you’re going to need more data than the proof of one witness.
It’s possible that I’m misreading or misunderstanding Hume here, and my solution to this problem is the actual diminution principle. I’m not sure! I welcome anyone who may know better to correct me, and I will put that edit here.
Any Possible Error Here Doesn’t Undermine Hume’s Argument
If you’ll notice, I’ve done something my critics believe I’m incapable of doing, and that’s to admit that Hume may have gotten something wrong!
Before they dance on the grave of Hume’s argument, it’s worth mentioning that the shortcomings or lack of clarity when it comes to the diminution principle does absolutely nothing to undermine Hume’s maxim or argument on miracles.
That’s because Hume would object that any testimony for a miracle would ever amount to a proof! The reason why is because miraculous testimony is uniformly bad and corrupted by religious motivation, among other things.
As a law of nature always diminishes or destroys mere probabilities, and miracle testimony is a probability at best,7 this lack of clarity on how proofs can destroy each other doesn’t affect the outcome of Hume’s argument.
Hume is seemingly granting for the sake of argument that testimony could be a proof, and (demonstrating through the diminution principle) that it still wouldn’t be enough to establish a miracle because of flaws inherent to miraculous testimony. The entire purpose of the second part of Of Miracles is to elaborate on why miraculous testimony as a type or kind of testimony is unreliable, and thus not a proof. He did not need to concede that much to miracle proponents, as testimony itself is likely never a proof, let alone miracle testimony.
Multiple Witnesses?
The “multiple witnesses” objection is probably the climactic point of disagreement between Humeans and Bayesians (not that they’re an insoluble dichotomy). The two sides have radically different opinions on this point because of how they formulate probability.
The reason why Earman and the Bayesians believe that multiple witnesses could affirm the plausibility in a miracle is because of the math of Bayes theorem on top of some assumptions about multiple minimally-credible witnesses. I’m not going to go so far to say that Earman is completely wrong on the mathematical side of things, so much as I will cite a modern Humean who would object to the possibility of truly independent witnesses; objections that Hume would likely share.
The Humeans further disagree because they believe that testimony is never enough to establish a miracle, as both the baseline probability of a miracle is so low, and the fact that the quality of miracle testimony is also low. As a scholar such as Fogelin would say, the standards of establishing a miracle are very high, and testimony comes nowhere near that standard.
As Hume wrote in Of Miracles and as I have elaborated upon in the past, we consider eye witness testimony reliable because of our experience with it being reliable. And unlike other forms of evidence, testimony is always inferior because we have to infer things about the testimony based on factors particular to the witness on top of the actual facts attested.
A Court Metaphor
On a practical level, I side with the Humeans because we can’t reliably generate accurate numerical probabilities for most points of factual dispute. I also think having standards for evidence quality is a more helpful and realistic method for people to implement in their daily lives than assigning numerical credence to specific data points.
When we conceptualize evidence like it’s a criminal trial (as the Humean does), it helps communicate why we don’t take witness testimony as seriously as other forms of evidence. Let’s say we have a group of very reliable witnesses to a crime. How reliable and how many are irrelevant; they could be the most or least reliable witnesses and the point would still stand. Our witnesses see two men, Jeffrey and Ted, beating up another man, Cam. Jeffrey takes out a knife and stabs Cam. All of this is caught on camera, and it clearly shows Jeffrey stabbing Cam. The knife is recovered and Jeffrey’s fingerprints are found on it, but not Ted’s. Yet, for some reason, all of the witnesses say they saw Ted stab Cam!
If we were a jury or otherwise evaluating this situation, how would we evaluate these competing pieces of evidence? On one side of the scale, you have the testimony of a number of reliable witnesses. On the other, you have camera footage and fingerprint evidence that says something completely different, and there’s no evidence that this evidence was tampered with. Who do we convict for stabbing Cam?
The answer, I would like to think, is pretty obvious. We convict Jeffrey! Even though the witnesses are numerous and reliable, they are not as reliable as camera footage that’s functioning properly or fingerprint evidence that followed proper chain of custody. We have a stronger proof for the reliability of camera footage and fingerprint evidence than we do eyewitness testimony, and so we are more confident in the conclusions of the former.
A good defense attorney may argue that there’s sufficient reasonable doubt to convict Jeffrey, but that’s the best they can do. Just as the proof of a miracle based on testimony cannot overcome the proof of the laws of nature, the proof of the witnesses cannot overcome the proof of the evidence of fingerprints and camera technology. Given our experience of witnesses, we can be assured that they’re almost certainly in error somehow, and so their testimony is not a proof.
The Verdict?
Multiple witnesses cannot attest to a miracle sufficiently that one can be confident that a miracle happened. To be clear, that’s not the same thing as saying multiple witnesses can’t attest to something spectacular sufficiently that one can be confident that it happened. It’s also not the same as saying that no evidence could establish the reality of a miracle.
Eye witness testimony is reliable for probable events, and that’s why we use them in criminal investigations (as crimes are probable). But when it comes to expanding our understanding of the universe to the point that we are confident that our current understanding is wrong, we need evidence of a higher quality than witness testimony.
Final Thoughts
In all, I appreciate John Earman’s criticisms of Hume, because they’re an excellent entry point in communicating misunderstandings about Hume, but also bridging the gap between proponents for numerical probability and factual probability.
Having said that, I do not hold the same appreciation of the people who lazily echo Earman’s criticisms and don’t check their accuracy. Many of the scholars I’m citing wrote these responses over 20 years ago and yet we’re still pretending like Earman dropped the mic and ended Hume during the waning months of the Clinton Administration.
I think the more plausible explanation for why Earman is elevated by apologists is because he provides a simple solution to a problem they cannot solve and cannot be bothered to solve. None of this is to say that Hume is perfect, but that he is far closer to the truth than his critics portray. To the extent that Hume is wrong, it’s because it’s difficult to update his probability into Bayesian terms without revision. In fact, one of the best Hume scholars is Peter Millican, who has a lot of interesting criticisms of Hume’s maxim on miracles on this very point. I’ll hopefully post that in the near future.
But in the meantime, miracle apologists should not cite John Earman.
This is a huge point, I’m not sure if I’m going to cover it all in this post, but if not, I will in a near-future post.
Yes, I’m showing my cards early here, are you surprised, dear reader???
The wording of this sentence is lifted directly from page 38 of Vanderburgh’s article on the subject. I couldn’t find Coleman’s article in free PDF form.
Vanderburgh disagrees with Coleman that it’s accurate to call the non-Pascalian probability as Baconian for historical reasons, as it precedes Bacon for centuries, so I’m just going to refer to this kind of probability as non-Pascalian.
That’s another post that I’m excited to write soon.
Some scholars such as Millican disagree with me and have interesting things to say about this, but I still think it at least appears like a double count.
And Hume says they often aren’t even a probability.
I thought this said Ehrman for a second and I was like when did Ehrman rebut Hume, I thought they played for the same team 😂