The title of this article is a double entendre. I’m responding to a pretty good video by Emerson Green (“atheist, non-physicalist, and host” of the podcasts Counter Apologetics and Walden Pod) titled 5 Mistakes Atheists Make About Epistemology. In which Green aims to correct what he sees as the five most common mistakes atheists make regarding epistemology (as in the philosophy, or one’s theory, of knowledge). Here I will summarize his points, and correct some missteps. So, yeah. Green is correcting five mistakes; and I am correcting his correcting five mistakes.

But overall, my corrections are (mostly) trivial compared to his. Almost everything Green says is correct, and very worth the careful consideration of, by all atheists. I think his video should even be essential viewing and study for anyone who wants to be serious about what they should believe in regard to gods. The only “corrections” I have are mostly minor tweaks, really, which only enhance and build on his actual goal, which is to make atheists better at this, and thus less vulnerable to being pwned by more capable theists who actually understand the philosophy of epistemology better than them. You should reverse that order. You should endeavor to be better at it than they are. Green’s video, and my supplementation, will help.

General Overview

To be charitable (a point Green recommends in his closing, and I concur), when Green says these are five mistakes “atheists” make he does not mean “all atheists,” but just “some atheists.” But still, mistakes made often enough to become a problem that needs taking the trouble to produce a video discussion to correct. I agree. I’d probably rank most of these close to my top five as well (even if my exact top five would address some other issues; and as you’ll soon find out, I’d drop one of his five as really a non-problem); and I have seen atheists here and there make all of these mistakes, in some way. These five mistakes Green labels in the form of erroneous claims you often hear from atheist circles: “There’s no evidence for theism,” “Theism is unfalsifiable,” “Testimony is not evidence,” “Intuitions don’t matter,” and “Lacktheism [is real].” And in all this Green relies a lot on Michael Huemer’s Understanding Knowledge, a very good treatment of modern epistemology that I also recommend. Like any philosopher, I can’t say that “literally” everything Huemer says is correct (I haven’t systematically vetted his every statement), but he’s generally right, easy to read, and easy to vet.

There’s No Evidence for Theism

Green’s treatment of this claim is very good. There’s much of value the average atheist can learn from it. It falters only at the point of Ordinary Language Philosophy, such that there are two caveats I think any atheist should grasp in addition to what Green says, which corrects him on a few minor points.

The first of those caveats is that much of what Green is revealing in his discussion (but does not explicitly say) is that most debates over the existence of God aren’t really debates over the existence of God, but debates over the existence of the evidence that is claimed to exist for God. And this can be a helpful realization.

For a clear example, the Moral Argument for God is really an argument over whether the thing claimed to exist, that is proposed to be evidence for God, even exists. To illustrate what I mean, let’s walk it trough…

The Moral Argument proceeds as:

1. Moral facts would not exist if God did not.
2. Moral facts exist.
3. Therefore, God exists.

This is formally valid. But it is unsound, owing to a hidden equivocation fallacy: there is no sense of “Moral Facts” on which both premises are true. But the theist trades on switching the meaning from one premise to the next to get the argument to work.

Given the sense on which it’s true that “moral facts would not exist if God did not,” “moral facts” must consist of some sort of bizarre cosmic powers or substances, or ordinances from beyond. But there is no evidence any such things exist; so on this definition, Premise 1 is true, but Premise 2 is false (or at least, not established, and so neither can the conclusion be). Once this is noted, the theist will often resort to ad hominem or ad baculum or well-poisoning formulations of this argument to get everyone to think that if “moral facts” in their sense don’t exist, then no moral facts exist, and therefore (it is usually implied) everyone should fear atheists (and becoming atheists), as the godless must be teetering on the edge of sociopathic madness. This fear is then meant to persuade people to believe moral facts must exist in the sense they mean, and therefore the argument carries through. This affective fallacy then hides inside the premises meant to convince you Premise 2 is true, and so is invisible in the syllogism above.

This kind of semantic and emotional trickery is typical of God apologetics. In truth, moral facts can exist in ways wholly not requiring of a God (for example, see The Real Basis of a Moral World, or Shelley Kagan’s position in his debate with William Lane Craig, Is God Necessary for Morality?, or even Eric Wielenberg’s proposal that “moral facts are brute facts”), thus toppling Premise 1. Indeed the evidence also stacks in favor of only some such moral facts existing, also toppling Premise 2. But the point relevant here is that the evidence claimed to exist for God doesn’t in fact exist: it was a false claim all along. And so to expose this, and thus show this argument fails, you really have to get into debating the existence of this evidence, not of God. When atheists say “there is no evidence for God” they often mean this: that whenever theists present them with purported evidence for God, upon examination it turns out not to be. And “non-existent evidence that is merely claimed to exist” does not count as “evidence” in any such statement as “There is no evidence for God.”

What is useful about this is realizing the question is often not, “This evidence is not convincing enough to justify believing in God,” but actually, “This evidence you are claiming for God doesn’t even exist.” And often this is what atheists are actually talking about when they say “There is no evidence for God.” And they have good reason to believe this. Because it so routinely happens that evidence, once examined, falls apart. Green somewhat misses this, because he isn’t thinking about what the ordinary language meaning of such declarations is. And this follows even for the sense of “evidence” that Green correctly defines, which is Bayesian: a fact is not evidence for a thing unless it exists and its existence increases the probability of that thing (and, as he notes, the strength of any piece of evidence equals the degree to which it increases that probability).

Consider, for this point, a less obvious example: the Fine Tuning Argument. Usually the argument is that Fine Tuning (of the fundamental constants of physics, which includes the number of open dimensions and by extension the inevitable outcomes of those constants, like stellar physics) is more probable by design than by chance, and therefore this is evidence for a designer. Setting aside any specific definition of “designer,” in ordinary language any designer capable of that qualifies as “a god” in the global semantic sense (e.g. would the ancient Romans and Chinese have called it a god, even if it was just an alien or a supercomputer—yes). So, faced with such an argument, as an atheist who insists “it would be a miracle” if there were no evidence “for” anything (like God), Green might say, “Yes, this is evidence for God, in that it increases the probability of a God, but it still does not increase it enough to make ‘God exists’ more probable than ‘God does not exist’,” and so it does not successfully prove “God exists.” Which is a logically coherent position. Green is right that atheists don’t need to deny any evidence for God exists; it’s perfectly possible for evidence to exist for a thing and be insufficient to warrant believing that thing (as with the “single random universe” model I discuss in A Hidden Fallacy in the Fine Tuning Argument).

The problem is, though, that this is all incorrect at the very first premise: in actual fact Fine Tuning is less probable if God exists and thus is actually evidence against God (see Why the Fine Tuning Argument Proves God Does Not Exist and Three Common Confusions of Creationists). Once again, when we fully examine the evidence, it falls apart. Pretty much all apologetics works this way: when you look closely at any evidence offered for God’s existence, it often flips the other way, into evidence against God’s existence; and even at best, it washes out as neither, and thus is no longer really evidence, even by Green’s definition. It is actually extraordinarily difficult to find any evidence that survives this analysis; and this is why atheists could be justified in saying “There is no evidence for God.” Green overlooks this because of one simple semantic move that he makes, which may be internally coherent but again collides with ordinary language: he keeps counting “understated evidence” as evidence. And that is the second defect in his presentation.

Green correctly defines “understated evidence” as evidence that, by leaving key details out, looks like evidence (as in, its existence as described increases the probability of the claim), but isn’t really evidence (when one puts the omitted data back in, it no longer increases the probability of the claim). He gives the example of someone accused of stabbing someone to death, and a prosecutor presenting evidence that they bought a knife the day of the murder—but omits the detail that it was a butter knife, and therefore forensically not the knife used in the murder. Green keeps calling this “evidence.” But in ordinary language, this is not evidence. When atheists say “There is no evidence for God,” they usually mean there is no real evidence for God, and thus they are not counting “understated evidence” as evidence, but in fact ruling it out as not really evidence at all.

So Green is wrong to take them to task for this. Their semantic usage is entirely coherent, and arguably formally more correct than his. Because Bayes’ Theorem entails all probabilities must be stated as given all available knowledge: evidence (e) and background knowledge (b) must be complete; e and b must sum to all available knowledge. If you leave information out, your equation no longer describes reality. And any statement of the evidence that does not describe reality should not be counted as evidence. This is simply another example of evidence not really existing. It is lying by omission, which is as much lying as lying by commission. And lies, by definition, are not evidence of what they purport. Even if you want to be charitable and call the omission an accident, the same follows: false descriptions of reality are never evidence of anything they purport to indicate about reality. Green is simply wrong.

Now, maybe there is something Green has in mind that avoids these defects in his argument, some “evidence” that really exists, and really does increase the probability of God (even if too minutely to be persuasive). But he never gives any example. So I am still not persuaded any evidence for theism exists. Once we exclude “understated evidence” as not really evidence, I don’t think there is any evidence for theism. Green thinks this would require a miracle. But he never establishes why. Obviously most things won’t have any evidence for them. This condition therefore can never be described as miraculous; it is, rather, quite typical. Which brings us back to the first point: really all Green is teaching us is that many arguments over whether God exists are really arguments over whether the evidence exists that is purported to make God more likely, which includes not just whether the “fact” claimed exists (though often even that turns out to be untrue), but whether that fact’s existence actually increases the probability of God (which is a more subtle defect).

Nevertheless, Green’s discussion teaches several valuable lessons: atheists do not need it to be the case that there is no evidence for God (there can be evidence for a thing that is still inadequate to make that thing likely; many false claims do have “evidence” for them, yet remain false—many even demonstrably false). Atheists do need to have a better grasp of what evidence is—what makes it “evidence.” And he is right that they need a proper Bayesian conception of this, as that is the only coherent definition of evidence that fully explains why evidence has any pertinence at all (which is that it increases the probability of a thing) and how it obtains that pertinence (what has to be the case for some fact to increase the probability of a claim, and how we know when it does so a lot or only a little). Atheists will be much better arguing their case if they grasp these essentials (see my article The Scary Truth about Critical Thinking and my list of examples in Bayesian Counter-Apologetics).

I also think that, in principle, Green’s advice to revise “There is no evidence for God” to “There is no good evidence for God” is okay as far as it goes. That can reframe a debate as to what counts as good evidence and why, and if you aren’t yet as sure (as I am) that there really is no real evidence for God, it can position you in a more easily defended space. Whereas I am well versed enough to know that I can actually defend the claim “There is no evidence for God,” and I don’t want to give the theist false hope that evidence exists when it doesn’t—as that will entice them to the fallacy that anything for which evidence exists can be believed.

Adding the word “good” might make clear you are not impressed by “understated evidence” even if that is for some reason still to be counted as “evidence,” in the hopes that this will prevent an apologist running a semantic game on you and start listing “understated evidence” as counter-examples to “no evidence,” requiring you to digress on why you don’t consider that to be real evidence (and neither should they). But this will happen even if you put the word “good” in, because any debate about what counts as “good” ends up in the same place. So there really is no net benefit. Either way, you’ll be stuck debating what counts as “good evidence.” And this is generally where you want any argument to land anyway. Hence what you should really take from Green’s overall lesson is that atheists should make most debates over God into debates over the existence—or strength—of the evidence.

Theism is Unfalsifiable

Green also handles this reasonably well, and again only drops the ball at the point of ordinary language philosophy. When atheists say “God is unfalsifiable” sometimes they might be behaving as Green is concerned about, but in my experience, they usually mean: after all is said and done; not “right out of the gate.” And in this respect Green’s treatment seems not at all aware of this being an actual serious position in formal philosophy, called theological noncognitivism. He erroneously thinks atheists who say this are referring to Popperian philosophy of science, and thus burns a lot of time talking about how theism doesn’t have to be a scientific claim to be justified, just like most of our beliefs aren’t. Which is true. Hardly any of our beliefs rest on scientific evidence, or scientific certainty, and atheists do need to confront this epistemological fact more coherently (for example, see my article History as a Science, and my discussion of the ladder of methods of knowing in Sense and Goodness without God II.3, pp. 49–62).

That’s worth doing—I do think there are atheists who make this mistake, of thinking, or saying unthinkingly, that only scientific knowledge is credible (an easily refuted position, as Green aptly makes clear). I do think the notion that God, as a scientific theory, performs too poorly to take seriously (that’s why it remains a fringe, even crank, position in every peer-reviewed science). But one should not confuse that with “belief in God is not justified.” Just eliminating God as a scientific fact is not enough. There is a lot more work left to do before you can say, “There is no adequate reason to believe in God,” much less, “God is unfalsifiable.” Particularly given all the gerrymandered ways to define God that escape scientific review. Though this is where the notion comes from that God is unfalsifiable. And I think this is the one point at which Green drops the ball.

If you go back to the seminal work that launched the modern philosophical position of theological noncognitivism, you’d discover a detail Green skips over (possibly for lack of knowing it): Antony Flew’s famous paper “Theology and Falsification” published in 1955 (which—fun fact—I helped get republished online, at the Secular Web, when the internet was in its infancy). Flew arrives at the position that God is an unfalsifiable proposition not by the way Green thinks—he didn’t simply argue “that’s how everyone always defines God” or any such notion. Flew arrives at it by rhetorical iteration: his point is that although everyone starts by proposing God in a falsifiable format, the moment you start challenging this with falsifying information, the theist retreats to an ever-more-unfalsifiable definition of God, until eventually they give up falsifiability altogether. Flew’s point was that belief in God is posed as falsifiable; but in actual psychological fact is un-falsifiable. And this can only be discovered by Socratic inquiry, whereby the theist, who starts at a falsifiable position, retreats to an unfalsifiable one when challenged.

Flew’s point was therefore that in actual fact theism is unfalsifiable. Its falsifiable forms are just facades intended to carry rhetorical effect, but they do not honestly represent any theist’s real belief. And Flew is right to a point: lots of theists (and I mean lots) cling only to an unfalsifiable God, and simply engage every effort to avoid admitting this; and for them all surface rhetoric in defense of God is thus an effort to dodge cognitive dissonance and not an honest accounting of what the theist really believes. Flew only went wrong (and all theological noncognitivists with him) in declaring this universally the case—rather than only the case for a subset of theists (however large or small).

Green’s closing point in his video about charity and when to give it up (which is when a theist tells you they reject your charitable or steel-manned take on their claim) I think is still a tad too gullible, because it assumes everyone is honest, even with themselves, about what they really believe and why they really believe it. But that’s not the case. A lot of people will not be that honest with you—especially the delusional. And this can explain why, after convoluted discourse after convoluted discourse, so many atheists start suspecting that theists don’t really believe in a falsifiable God. They only really believe in an unfalsifiable one. Because that’s what they keep finding. Looks like no white crows.

Green does mention this possibility, albeit late and briefly. But he neglects to consider whether this might in fact be what atheists actually mean in ordinary language by “God is not falsifiable”: that “every time I argue with a theist, that’s where we end up; ergo, I no longer believe it to be probable that any theist really believes in a falsifiable God,” which could be from their epistemic position a reasonable conclusion. Because they’ve met the facade, and only discover the truth after Socratic interrogation; and most theists are not available for that interrogation, or are highly skilled in thwarting or dodging it, so the existence of, for example, books defending a falsifiable theism no longer affords evidence of that being anyone’s real belief. Prior probability at that point favors it not being (even if only by a little, still by a preponderance). This can only be countered by meeting, and engaging extensively with, a sincere theist who maintains a falsifiable position under substantial challenge; and atheists rarely have that experience.

That said, though, I still concur with Green’s overall analysis, that this cannot describe “all” theists or even “all” theism, as even positions no one genuinely takes still exist as possible hypotheses in need of eliminating (that even believers avoid those versions, though, does count as evidence they are false; Flew’s paper launches with a detailed example of this very phenomenon). So regardless of whether a theist really believes in a falsifiable God when they present one (“fine tuning proves God exists” entails a falsifiable position: that if there were no fine-tuning, their case for God diminishes; or worse if fine tuning turns out to prove God does not exist!), they are still presenting one, and that “version” of God is falsifiable. Likewise, the existence of atheists who were once believers also proves God is a falsifiable proposition: as clearly, for them, enough evidence arose to falsify it.

Green demonstrates that there are theists (even if not very common) who admit there is evidence against God’s existence, they just think it’s insufficient to overcome what they believe is evidence for God’s existence. There are, true, theists who present, say, a design argument for God, and then are met with a counterexample proving their evidence in fact to be “understated evidence,” such as the Argument from Evil, and then respond by making up epicycles of excuses to deny that that feature is not predicted by their design hypothesis (indeed it’s even counter-predicted by it). But there are also theists who admit this counter-evidence does reduce the probability of their hypothesis, or genuinely don’t understand why it does. Such theists do exist. Where they will falter is, say, in their gross under-estimating of how much such counter-evidence reduces the probability of their God. But that then is not an attempt to render their theory immune to evidence in the sense of unfalsifiable, but rather an attempt to render their theory immune to evidence in some more convoluted way.

Flew might have said there was no difference in practice; refusing to admit the strength of evidence against them is just another way of making a theory unfalsifiable, one that maintains falsifiability in appearance, but denies it in reality. That, too, may be the case. But I doubt it is always the case; and even when it is the case, there still remains the falsifiable form of the proposition—because, obviously, if you’re an atheist, you have concluded it has been falsified.

So in this technical sense, Green is correct: it is not true that “God is unfalsifiable” in the sense that “Every possible God is unfalsifiable.” Obviously most Gods are falsifiable, whether any theist really believes in any of those or not (and it is credible to think some do). Green is only incorrect in assuming that when any atheist says “God is unfalsifiable,” they must mean this. They usually don’t. They usually mean something closer to what Flew meant: that in actual practice, no one really believes in any falsifiable God, despite whatever pose they take. And atheists who say this often are saying it as an honest reaction to their own empirically updated priors: they have yet to meet a theist whose belief survived interrogation without retreating to an unfalsifiable position. I would only say their experience has been too limited; I’ve met counter-examples. Otherwise, it remains the case that all atheists who were once believers are counter-examples to Green’s understanding of the statement (they clearly regarded God as falsifiable); but if what an atheist means by “God is unfalsifiable” is not that, but the other thing (that no one remains a believer in a falsifiable God after substantial challenge), then the existence of the deconverted is no longer a counter-example.

Still, either way, I think one lesson we can take from Green’s analysis is that atheists would do better to qualify their declaration as, “What believers really mean by God is often unfalsifiable,” to clarify they are not talking about “all possible gods” nor “all possible believers.”

Testimony Is Not Evidence

Green’s analysis of testimony is a good corrective to a lot of mistaken attitudes, although he seems a bit wobbly on the probabilistic reason why testimony is trustworthy when it is (and therefore why, mathematically, it isn’t trustworthy when it isn’t). See my recent discussion of this very point in Daniel Bonevac’s Bayesian Argument for Miracles.

The reason testimony in some formats (e.g. mundane institutional records) and circumstances (e.g. scientists documenting observations; historians openly using methods known to be reliable; testimonies under oath) and sources (e.g. witnesses who have been well-vetted as reliable) is more reliable than others (e.g. religious propaganda vs. disinterested bystander accounts; vague or unvettable hearsay; witnesses whose trustworthiness is unknown or, worse, already documented to be unreliable) is that “reliable” testimony is less likely to be generated by error or deceit (even if only by a bit) whereas “unreliable” testimony has a well-established track record of more likely being generated by error or deceit. Hence, not all testimony is equal. Some testimony isn’t evidence.

You can think of it this way: Why do we not have this account from a reliable source?

If Jesus murdered two thousand pigs, it is not probable that no one would have reported such an astonishing collapse of a whole region’s entire pork industry, and that we only hear of it via anonymous hearsay half a century later. A story such as that is usually produced by mythmaking, not by having happened (establishing a low prior probability), but also, the form in which the account survives confirms that fact (establishing a negative likelihood ratio), as that is exactly what we would have on the mythmaking hypothesis, but is at least slightly (if not in fact very much) less likely to be what we would have if it really happened. And once you document dozens of such things in the same text, thus from the same “source” (like the unknown author of Mark), their rep is destroyed (see Craig vs. Law on the Argument from Contamination and Is the Principle of Contamination Invalid?).

This is because once all that evidence gets fed in and repeatedly updates your priors, your background knowledge at that point establishes that that author usually fabricates, which means there can never be any testimony from them that is more likely to be true than false—unless you have some very particular evidence otherwise. Which is the point of the “method of criteria” applied to the Gospels, which has its own fatal methodological problems as I and others have documented (see Proving History). But this doesn’t mean such conditions can’t exist. It just means, as a matter of contingent happenstance, it doesn’t exist for the Gospels.

Christians hate being told this. But alas, it’s objectively, epistemically the case that the Gospels are shit as evidence—compared to a hundred other kinds of testimonial evidence that could exist, and does exist for quite a lot of things (even ancient things). This dissonance is why Christians fight so hard for us to regard the Gospels as “just like” researched biographies of the era: they want to contaminate their mythology with the superior epistemic qualities of that genre, for exactly this reason. In truth, those qualities are not so great anyway (ancient biography is not well trusted by experts); but they also simply don’t exist for the Gospels at all (per both Burridge and Walsh). This is actually the mainstream position in the particular fields of classics and ancient history (see On the Historicity of Jesus, Chapter 10, and Elements 44–48, Chapter 5).

For example, I find in On the Historicity of Jesus that no story in the Gospels has better than a 50/50 chance of being true (and many have less) without independent external corroboration (and for stories about Jesus, there isn’t any). That didn’t have to be the case. But that it is the case argues against the Gospels being historical reports at all. The peculiar nature of that source thus does not produce favorable priors or likelihoods. Whereas when we get to something closer to real testimony (like the letters of Paul), Christians get nothing clear enough to work with—which is also weird, and weird is just another word for “infrequent,” which is just another word for “improbable.” But it’s not weird if the claims in question are false; what we have is then exactly expected. So the likelihoods go poorly even there. And yet that’s because we are trusting Paul’s testimony! When he only ever says Jesus was seen in revelations and not dinner parties, it’s more likely that’s true (or at least true that that’s what he was going around saying).

This is why Christian apologists have to tell lies or twist truths to try and “rehabilitate” the reliability-status of their sources (see, e.g., Did the Apostles Die for a Lie? and There Are No Undesigned Coincidences and Establishing the Biblical Literalism of Early Christians, etc.). Because that would up the likelihoods of their content on the hypothesis of it having actually happened; otherwise their contents are at best just as likely whether they happened or not, and at worse, less likely. We can sometimes do better with the letters of Paul, but not to results Christians want (see, for example, Then He Appeared to Over Five Hundred Brethren at Once! and Resurrection: Faith or Fact?). Testimony can also of course be countermanded by other testimony, as well as by its own internal suspect content, like unexpected omissions of information, which being “unexpected” lowers the likelihood (see, for example, How To Fabricate History: The Example of Eusebius on Alexandrian Christianity and The Rain Miracle of Marcus Aurelius: A Case Study in Christian Lies). But even our knowledge today that eyewitness testimony is unreliable (and hearsay even more so) is itself based on testimony—of the scientists and researchers who observed and documented all the data this conclusion depends upon.

Hence apart from the muddle about why, everything Green says remains true: it is almost always testimony we are basing all our beliefs on; so, obviously, it cannot be the case that “all testimony is unreliable.” Indeed, all prior probabilities are the outcome of previous likelihood ratios (actually or de facto), so even the fact that miracles (for example) have a low prior probability is built on the evidence of testimony (see Naturalism Is Not an Axiom of the Sciences but a Conclusion of Them and How Not to Be a Doofus). For example, in the ancient Rain Miracle case (and I’ll assume you’re familiar; if not, you can go back and catch up there if I lose you on the details here), all the evidence exposing the Christian story bogus still consists of other people’s testimony; even the archaeological inscription corroborating the pagan account still just contains a person’s testimony (the sorcerer Harnouphis confessing his religious beliefs and presence in that region at the time). It’s just that there are attributes of these testimonies that cause them to have more favorable likelihood ratios.

For example, it is extremely unlikely they could have been forged or coordinated against the Christian account; and they agree with other known details in a way the Christian account doesn’t, e.g. it is extremely unlikely Marcus Aurelius commissioned a legion staffed entirely by Christians, but entirely expected that it would have a pagan sorcerer in its retinue—again, facts we know only from testimony, as to the context and conditions and attitudes of that imperial court, but that’s still testimony extremely unlikely to have been doctored or altered to conceal a completely different state of affairs (whereas the Christian testimony has no such virtues; it’s actually quite likely to be faked). As another example, we have an inscription attesting to the existence of the same “Thundering Legion” already a century before Aurelius, which leaves the Christian testimony that that legion won that name for the Aurelian Rain Miracle highly unlikely to be true. But that inscription is still just a recorded human testimony. It’s just that it can’t likely have been forged or coordinated centuries in advance to impeach a later Christian tale.

So in no way can one say “Testimony is not Evidence.” Testimony is by far the bulk of all evidence we ever have for anything. Science. History. Geography. The location of the DMV. Where the nearest loo is when you ask a bystander. It’s all just human testimony (see History as a Science). The distinction that has to be made is, rather, the likelihood of a certain testimony, appearing in a certain form, in a certain context: some testimony contains content or arises in contexts that make it more expected as an error or fabrication than reliable reporting; and some testimony contains content or arises in contexts that make it less expected as an error or fabrication than reliable reporting; and some, neither. Indeed, our entire navigation of modern media depends on these distinctions having to be made (see A Vital Primer on Media Literacy and No, Christianity Was Not Invented in the 4th Century).

Green eventually gets stuck in the weeds of untangling coherentism vs. foundationalism over this, which I think is irrelevant. He eventually comes to the same conclusion, that it is irrelevant, but he doesn’t articulate why, mathematically, that distinction makes no difference. The answer is that it all works out the same either way, once we realize the question of whether, and how much, to trust testimony is a function of the content and conditions of that testimony, and not a function of it merely being testimony. This does mean it is possible for human testimony alone to be sufficient to establish a miracle has occurred, even if it just so happens that it never yet has (as I just explained in detail in Daniel Bonevac’s Bayesian Argument for Miracles)—and atheists don’t like admitting this, but only because they are not analyzing why this seems fishy to them.

The reason it seems fishy to them is that the kinds of testimony that would have swayed a rational observer to such conclusions have never occurred, while what we get instead is a lot of lying and gullibility; and this has correctly updated our priors to simply not trust testimony when it is to the unbelievable and serves all the convenient needs of the testifier. But it is easy to confuse that with the principle “all testimony is untrustworthy.” Incorrect. Christian testimony is untrustworthy specifically owing to the way history went: if Christians hadn’t piled up centuries of lying and gullibility, but had piled up instead the kinds of testimony we have for other weird facts of science and history, we’d be in an entirely different epistemic position with regard to their testimony. It’s important not to lose sight of that. There are reasons Christian miracle testimony is unbelievable now—reasons other than merely that it is testimony. Green is right to correct atheists who make the mistake of not understanding this.

The Trick Here

Atheists who say “testimony is not evidence” might be reacting to what is actually a commonplace apologetical trick of “changing their hypothesis” mid-equation. They’ll start with “Jesus rose from the dead” and halfway through, when evidence doesn’t look to be going their way, they’ll suddenly convert that hypothesis into “Jesus rose from the dead and wanted to hide from the authorities” and all manner of other “add-ons” designed to get testimony that actually has a low likelihood on their hypothesis to have a high likelihood so it can finally be counted as evidence (see, for example, my entire debate with Jonathan Sheffield, “Shouldn’t the Romans Have Refuted the Resurrection of Jesus?”). But this is logically invalid. In any single run of Bayesian analysis, h (the hypothesis) must be the same; b (background knowledge) must be the same; e (the evidence to be accounted for) must be the same. You can’t swap in and out different ones in different places in the equation.

There are two ways to do it properly: scrap the whole equation and start over from scratch with a new h, with all add-ons attached. Christians don’t want to do this because then it means they have to downscore their prior probability, usually by more than any gain they could then achieve from the thereby-reinterpreted evidence, by taking into account the prior probability of each add-on (see The Cost of Making Excuses). But we can accommodate them. They can keep the stripped-down hypothesis and the original prior probability and the original equation. But then they have to “buy” this privilege back by moving that same probability-loss into the likelihoods: if the add-on (like “Jesus wanted to hide from the authorities”) is fully predicted by the hypothesis (“Jesus rose from the dead”) then it has no effect, the likelihood isn’t substantially reduced; but if it isn’t predicted by that hypothesis, whatever probability it has commutes to the likelihood. Because now it’s not a part of your hypothesis; it’s a part of the evidence.

That means if the evidence is “Jesus hid from the authorities,” and your hypothesis doesn’t contain any prediction that he’d do that, then the likelihood of that evidence is the probability of that observation on your hypothesis—the likelihood therefore includes the probability that Jesus would hide from the authorities. Which is at best 50/50, and therefore halved, if there is no evidence for or against Jesus wanting to do that or it is not deductively entailed by the hypothesis itself; and less than 50/50 if there is evidence against it or it is counter-entailed by the hypothesis. Leaving this out and just listing “some fanatical followers and one fanatical foe claim to have seen Jesus after he died” as “the evidence” is then “understated evidence.” It’s the prosecutor’s butter knife.

A full accounting of the evidence (also known as “reality”) includes “no one else saw him,” which is unlikely on the hypothesis that he really survived his death. Not impossible, remember, just less likely; whereas it is 100% expected if he didn’t survive his death and certain motivated people just imagined having seen him—or pretended to. Once again, when you put data back in that Christians leave out, their “evidence” for their position ends up reversing into evidence against their position. This is why it matters that the content of Paul’s testimony is also more expected if it was imagined than if real—because an eyewitness, and contemporary tradent, outweighs anonymous double hearsay decades later, yet his testimony is more likely to exist if it’s false than if true.

This is how testimony actually works: sometimes it can be evidence against what it purports; sometimes it’s evidence for what it purports, but only weakly; and sometimes, strongly; and sometimes, none of the above. It all depends on the details. Like an example Green mentioned: if you’ve never been to Japan, your belief in its existence relies on testimony; as does your belief that, by contrast, Atlantis does not. So think about what it is about the testimonies in each case that makes this evidence sufficient to warrant those beliefs—the number of witnesses implicated, the processes by which their claims were vetted in getting to you (e.g. how encyclopedias are procedurally, officially and unofficially, generated and fact-checked), and what would have to be the case if the claim were instead false. If the circumstances are less probable on it being false than if it were simply correct, then testimony is evidence, because it increases the probability of the claim being true, even if by a tiny amount—and even if not by a sufficient amount. Again, the prior probability of a claim can be so low that even a millionfold increase in its probability leaves it well below 1% likely to be true. So admitting “testimony is evidence” does not constitute admitting it’s true. Being evidence and being true are not the same thing; and I think that’s the principal confusion Green is here taking to task.

What Green could have added is that, unless you port your “theory rescuing” assumptions into its prior probability and thus redo all the math from scratch (as you otherwise must), those assumptions impact your likelihoods. And that can turn what you thought was evidence, into something that isn’t—or worse, into something that is now evidence against your hypothesis. If your theory only makes the results observed probable if you add an assumption that is not itself predicted by your theory, the probability of that assumption on your theory then becomes (or multiplies by) the probability of the results observed. Because then you aren’t just using your hypothesis to predict, say, “some people would see Jesus alive again,” but “some people would see Jesus alive again and most other people, especially the ones who would actually have been the most valuable witnesses, wouldn’t,” and you are probably doing that through an intermediary assumption, such that what you are really predicting is “Jesus would appear to a small number of fanatics and not want to appear to anyone else,” and so the likelihood of the evidence is the probability of the conjunction of both of those things. So now you have to ask, “On my hypothesis, how likely is it that Jesus wanted to hide from the authorities (and every other quality witness)?” Because that probability has to be multiplied against the remaining probability of the other testimonies.

This is why “I saw God in an ecstatic state” is not evidence for God, because we know typically that’s how people see phantoms, not realities. A real God would not probably choose a method of appearing that is identical to false appearances. Yes, they “could” choose it, but it isn’t probable; that is not something predicted to be the case 100% of the time (whereas it is if it was a phantom and not real). And if you add some feature to God, like some complex motive, that entails it would happen 100% of the time, you are now moving the goal post back to the probability that God has that feature. But, does your original theory predict with 100% deductive certainty that he would have that feature? And if not, what then is the probability he would have it? That then commutes to the probability of the observed result. If it doesn’t tank your priors, it tanks your likelihoods.

So there is no way to escape this outcome: given our actual background knowledge of the world and human history and experience (almost all based, as Green points out, on testimony—just, a lot of it, and of very good quality), a statement like “I saw God in an ecstatic state” simply will always be more probable if you didn’t really see God than if you did, even if by a small amount, but probably by a very large one (how often, after all, is someone who says that correct?). The same goes for all the usual modes of claiming to see God: “This unreliable fable says someone saw God” is not probably what would exist if God exists, but is probably what would exist if he didn’t. Likewise “I felt God in my heart,” “I heard hundreds of people saw God in the sky once,” and so on. We only find examples of good testimony to the existence of gods in fiction (from the Marvel Universe to American Gods, Time Bandits, Constantine, Dogma, Bruce Almighty, Lucifer, etc.). That nothing the like ever happens in reality is indeed evidence against the existence of gods. Which means had it occurred in reality, per Green, it would be evidence for the existence of gods.

Thus, to say “But God wants to hide and always look like a false god” in order to get “I saw God in an ecstatic state” or “This unreliable fable claims people had dinner with God” to be just as probable if God exists than if he doesn’t (or, more implausibly, to get it to be more probable somehow), will never work on existing background knowledge of how the world works, because “God exists” does not entail “God has that specific motive,” so the improbability that God has that specific motive multiplies by any other expectancy to produce the likelihood, e.g. if the evidence is 100% expected with that assumption, but that assumption is only 50% likely to be true (a 50/50 chance God has that specific motive), then the theory without that assumption only predicts the observation 50% of the time, because now what you are predicting is two things: “this observation and the existence of this motive to produce this observation.” Whereas the contrary hypothesis (that none of that is true) requires no added assumption; it simply predicts the observation 100% of the time, all on its own.

This would resolve Green’s concerns about underdetermination. The reason we prefer known (documented) explanations, e.g. hallucination or lying, to unknown or convoluted explanations (conspiracy theories, superhuman agencies, etc.), is that the latter are posits without evidence and thus are improbable, and their improbability thus commutes to all predictions (or to the prior; the mathematical effect is the same); whereas the former are posits well established by evidence and thus are probable. It is in our background knowledge that “claims of alien abduction” and “hypnopompic hallucination” correlate very strongly, and thus their co-appearance is highly probable, in precisely the way “aliens actually abducted you” is not. Particularly when we add background data such as that, ethnographically, alien abduction claims simply replaced faerie abduction claims precisely when the cultural plausibilities changed (once faeries became ridiculous and aliens seemingly plausible—when really, they are both pretty much the same). This “updating” of delusions in reaction to the changing tides in what is regarded as culturally plausible is well explored for the category of psychosomatic illnesses (which ones are popular changes as technology and culture change) in From Paralysis to Fatigue by Edward Shorter. And background knowledge affects all likelihoods.

Even in mundane respects this is true. Underdetermination is not a widespread problem (in science or any field) because knowledge of possible hypotheses is a feature of background knowledge and thus the discovery of a new hypothesis can update all your probabilities, and because prior probabilities can often quickly adjudicate between competing explanations without anything more ado (see Proving History, index, “Underdetermination”). And when neither is the case, we already admit when two or several hypotheses can’t be decided between on extant evidence. Which is more likely true is then simply unknown. That’s called agnosticism. I just don’t think we have this problem vis gods—which gets to Green’s fifth point (below). But first now to his fourth…

Intuitions Don’t Matter

Green performs worst on this point. And I think that’s largely because he doesn’t seem up on any of the scientific literature on intuition. There is quite a lot—indeed, already twenty years ago (see my bibliographies in Sense and Goodness without God, index, “intuition”); even more so now (a decade after Kahneman’s Thinking Fast and Slow became a standard in the field). The biggest problem here is that Green presents no examples. He gives no examples of what he thinks atheists are talking about when they say “Intuitions Don’t Matter,” nor any examples of what their opponents claim vis-a-vis intuition. And philosophy suffers when you jump directly to the abstract rather than starting with real-world particulars and building your abstractions from those. He just engages in some vague hand-waving about “intuition,” a concept he never even defines.

Because Green evidently doesn’t have a science-based model of what intuition is or how it works, he can’t articulate when we should and when we shouldn’t trust intuition, and how much. As I explained in Sense and Goodness (and as any scientific expert in the field today would tell you) “intuition” just means “skill,” and relates to what we call noncognitive knowledge (for some examples, see my discussions in The Mind Is a Process Not an Object, “The Argument from Plus or Quus,” and “The Trouble with Mary”). It will be reliable only in proportion to what empirical past experience it is built upon—and even then it will typically never be as reliable as a well-run rational-empirical method. An experienced firefighter will have good intuitions about what to do in a burning building; but getting trained up in a century of accumulated evidence-based firefighting studies will help them even more; and they can only know their intuitions are reliable in that subject after their repeated empirical success. But someone who has never experienced the formation of a bunch of different universes will have less reliable intuitions about what’s possible or likely in the domain of cosmology; and someone with no formal experience or training in the science of cosmology will perform even worse.

But this is all moot when we’re talking about what to believe. Intuition is great for abduction. It’s shit as a method of deduction or induction. That’s why we had to invent formal logics, mathematics, science, and critical thinking: because human intuition is catastrophically unreliable (see The Argument from Reason). Intuition can be helpful for adducing what theories to test or how to test them. It is not helpful for ultimately deciding what theories are true. Any conclusion arrived at by intuition is empirically only a hypothesis. You still need a logical or evidence-based process to prove it likely. The firefighter does that by acting on their intuition and facing the results. That this experiment is often fatal is why we pay them well. This is in no way a model for building a religion or philosophy of life. We don’t need to make an “immediate decision,” come what may, when doing those things. We have plenty of time to do them in a properly sound and more reliable way.

It’s thus inexplicable to me why Green mistakenly claims we don’t have any “non-intuitive” way of knowing things like “the conjunction fallacy is a fallacy.” This is false. Not only can this be (and has been) proved empirically—in both the real world and in the theatre of the mind—but it can even be formally proved axiomatically. One can then know it by testimony (“all mathematicians who have studied the matter, and all well-vetted mathematical studies of the matter, report this is the case”), if one doesn’t want to take the trouble to know it directly, though that’s easy enough to do as well. The most useful thing about most math is that you can empirically re-prove any of its experimental claims at near zero cost with just a pad and a pencil. It sounds like Green has maybe confused whether it is a fallacy with whether it is the fallacy people are committing to in certain experimental studies (these are completely different questions). But that would be a strange mistake to make.

Maybe Green means that, while all formal proofs in mathematics can be known non-intuitively, the axioms of mathematics from which all those proofs derive are, supposedly, only known intuitively. But this is false. Every single axiom in every single axiomatic system of mathematics is either tautologically true (and therefore cannot be meaningfully false), or is an empirically testable proposition and is known precisely by surviving such tests: the probability of the resulting evidence is then correctly deemed astronomically small on any other conclusion than that the axiom is true.

Even the coherence of robust axiomatic systems is not known “intuitively” but by such empirical reasoning. For example, people often go on about Gödel proving we can’t know certain axiomatic systems are internally coherent, but that only relates to deductive proof—all he claimed was that we could not deductively prove this; we still conclude they probably are coherent on empirical grounds. And now there are axiomatic systems that defeat Gödel, and in fact how they do so is itself evidence for the conclusion Gödel tried proving unknowable: his result depended entirely on maintaining the need of a merely procedural axiom, like The Power Set Axiom, which can be dispensed with and the same results achieved in other ways by self-verifying systems, albeit by such tedious means that we find it far faster and easier to just use the Power Axiom. This suggests Gödel’s conclusion was a consequence of the incompleteness of a common short-cut, and not the actual incoherence of any axiomatic system. Which is itself evidence for the conclusion that such axiomatic systems are internally coherent (and we don’t need a formal proof to believe this probable: see my discussion of this condition in The God Impossible).

The most commonly used axiomatic system is ZFC. Some of its axioms are functional tautologies, e.g. the Axiom of Extensionality is simply a definition of equality: we are merely deciding what we shall mean by that term in any equation or proposition, so there isn’t any meaningful sense in which this axiom can be “false” within that system (on the semantics of stipulations and other propositions, see my discussion in Sense and Goodness II.2, pp. 27–48); whereas dispensing with that axiom has the consequence that you either always end up with it again, or simply can’t talk about anything. That axiom is thus simply a mechanism for running analyses, without which (or a derived equivalent) you can’t analyze anything. Any attempt to claim analysis is therefore fictional is then refuted by its vast empirical success.

Meanwhile, some of the other axioms in ZFC are fully empirical hypotheses, like the Axiom of Infinity: we cannot know for sure that there are an infinite number of possible whole numbers (we can’t sit down and count them all out or write them all down; and there are more limited axiomatic systems that deny it as an axiom), but we can adduce strong empirical evidence that there are (by reductio, assuming the opposite and evaluating its likelihood on available evidence, e.g. every time we try thinking of a highest number we always fail; we can give good reasons why we should expect to fail; and we can observe the uncanny empirical success of transfinite mathematics, e.g. calculus). Mathematics simply has no real use for “intuitionism.” Any claim can always be proved empirically or functionally, or lacks sufficient reason to be sure it’s true. And that’s that. (On my ontology of mathematics, see All Godless Universes Are Mathematical, The Ontology of Logic, and Defining Naturalism II.)

Intuitionism is only useful as a heuristic for adducing hypotheses, and the urgent deployment of physical and cognitive skills. Because “intuition” isn’t magic; it’s just a computer engaging in trained pattern-matching and a creative exploration of possibility-space. Experienced guessing. Nothing more. Whether its results are true or not still has to be proved—and that always means ultimately with evidence (this was in fact the point of my first-ever peer-reviewed article). This does mean it would be false to say, universally, “Intuitions Don’t Matter” (they have a functional use), but as Green never gave any examples of atheists saying this, I can’t evaluate what he is talking about. What are those atheists denying that he is trying to rehabilitate? It doesn’t sound like “abductive utility” is what he is trying to recover for it. It sounds like there is some sort of “intuitionist proposition” theists are citing in defense of their belief in God that atheists are rolling their eyes at. But, like, what?

Racking my brain trying to think up an example, all I could come up with is the example of how William Lane Craig gets cornered into admitting that his claim that “actual infinities are impossible” is itself false (there is no deductive proof of any logical contradiction in actual infinities, and real mathematicians generally conclude they are logically possible), and thus retreats to claiming that somehow they are nevertheless “metaphysically impossible,” and when it is pointed out that that phrase is nonsense (it literally has no relevant meaning here, and thus confers no actual impossibility on anything), he appeals to “intuition.” Somehow he just “knows in his gut” that actual infinities are impossible, even though he can literally provide no evidence whatsoever that they are.

I see no merit in this to defend, so I hope this isn’t the kind of argument Green means to rehabilitate. Craig is simply bluffing (see my thorough adjudication of this, in the hands of one of his acolytes, in my Debate with Wallace Marshall). Intuitions, precisely in such matters (beyond human experience and contradicting of formal experts), are so routinely wrong that appealing to it here adds zero to the probability of Craig’s conclusion being true (for other examples, see The God Impossible, Three Common Confusions of Creationists, Everything You Need to Know about Coincidences; and On Hosing Thought Experiments, More Hosing of Thought Experiments, and Another Failed Thought Experiment).

That sounds like “Intuitions Don’t Matter” to me. If this is what the atheists Green is talking about are saying, they’re right—and Green is wrong. Otherwise, are there some limited senses in which intuitions can “matter”? Yes. But so far as I can tell, none are of any help to theists. In the end, I don’t see any big problem among atheists to correct here.

Lacktheism Is Real

Everything Green says here is correct, though it is mostly just a covert dig at one delusional atheist influencer, Steve McRae, who’s kind of the functional equivalent of an atheist flat-earther in this debate (see Who Is an Atheist? and Misunderstanding the Burden of Proof). McRae never updates his conclusions in light of evidence, and routinely will be gotten to agree with something (after hours of pulling teeth to get there), and then forget he agreed with it and go back to making the same claims he admitted were false the very next day. Which kind of suggests we are not dealing with a rational agent there. But his “position” is that there are three, not just two, possible states to be in with respect to gods: atheism, theism, and lacktheism. Somehow it is possible to merely “lack” belief in God, but that doesn’t make you an atheist, because that must entail denying God exists, which is a position requiring evidence, and one doesn’t need evidence to merely “lack” theism. This is nonsensical poppycock in almost every conceivable respect. It’s wrong on etymological history, real-world linguistic conventions, formal philosophical accounts, and basic logic. But I’ve already thoroughly covered that in the two articles linked above. No need to go round that again.

There are other atheists who do this, though they are usually less committed to defending it in any substantive way, and usually just call themselves agnostics and say they prefer not to use the word “atheist” only for fear that people will mistake that as indicating a stronger position than they hold, which is an entirely reasonable pose to take. One can admit that technically all agnostics who lack belief in gods are also atheists, by many a widely extant linguistic convention and countless stipulated ones, just as doubtful believers are agnostic theists, and at the same time have good reasons to avoid using that identifier in discourse, when you lack time and gumption to explain exactly what your degree of unbelief is. Saying “I’m an agnostic” simply covers that, with adequate efficiency. Debating whether that means you are “also” an atheist is just a stupid semantic game with zero utility—unless you have a legitimate reason for it, e.g. if you want to know how many “atheists” there are specifically in the sense of “people who don’t believe in God,” you had better design your polling instrument to account for agnostic unbelievers as well (likewise apatheists and the like). Otherwise you’re going to undercount.

Beyond that, the epistemology is simple: you always have some belief. Just as Green says. Merely lacking belief in God is a belief. This becomes clear when you replace “belief” or “knowledge” with, instead, simply assigning an epistemic probability to the proposition, i.e. the probability you would be correct to say God exists (or the converse of the probability that you would be correct to say God does not exist, which is the same thing). Then you will realize you have a belief—in a probability (or probability range)—and that you must provide justification for that belief. “Why are you assigning that probability, in the face of all the purported evidence as yet available to you?” You always have to give a reason. Even if it’s “I have no information making that more likely than not” or “I have never thought of that existing before.” That’s still a reason. It’s still an accounting of the state of evidence warranting your belief-state in that probability or range of probabilities (see, again, Who Is an Atheist? and Misunderstanding the Burden of Proof; no, seriously, if you don’t believe me, please read Who Is an Atheist? and Misunderstanding the Burden of Proof).

This is because beliefs are always just confidences in cognitive models, not propositions. Propositions only articulate in a linguistic code the cognitive models you believe in (or don’t). If gods don’t exist in your cognitive model of the universe, then you lack belief in them, and are therefore an atheist by many a widespread definition in use. Because your model automatically entails a certain (low) probability of them, whether you are aware of this or not—you live your life accordingly; this belief-state thus has a causal effect on your behavior. Talking about noncognitive agents (fetuses, cats, rocks) is just stupid semantic rigmarole at this point; yes, only cognitive agents have beliefs in this specific sense. No one is going around insisting triangles are atheists.

And even when atheists go around saying babies are atheists, they don’t mean they are hard atheists, only that they lack a belief; that they are, presumably, “agnostics,” in the precise and broad sense. They lack knowledge, and in consequence, lack belief, which translates to: their model of the universe entails a low probability for that entity, whether they can articulate that belief-state or not. Ironically, though, this is actually false. Child development studies suggest no babies are atheists; all babies believe in gods (in any globally relevant sense of the word), e.g. they start out believing they are god (newborns appear to be solipsistic), and soon believe their parents are gods, because they believe they cause every pain they suffer, that they cause the rain and thunder, that they can cause food or toys to materialize or disintegrate at will, that in fact they control literally every aspect of life and experience. In other words, gods. It takes years for babies to disabuse themselves of this false belief and start to realize there are no gods at all—neither themselves, nor their parents, nor (if they escape indoctrination) anyone else.

But the short of it is, most debates over “who is an atheist” or whether one can be “not an atheist” and still lack belief in any gods are just semantic wheel-spinning, a waste of everyone’s time, serving no useful purpose. The only thing that really matters is what probability you would assign to God and why, and whether that’s a good reason—regardless of whether you have thought about any of this or not. Almost everything else about this is smoke and mirrors, a fabulous dancing monkey better off strangled dead by common sense and a healthy respect for sensible time management.

Conclusion

I mostly agree with Emerson Green. And I highly recommend this video as fundamental. Even where it is wrong it exercises your critical faculties in productive ways. And if you fear being misled or disinformed, my article here will help prevent that. I don’t think Green makes any mistakes not here addressed and resolved, and with one exception (his odd discussion of intuition), even after correction, he ends up right, insofar as what he is attempting to correct could warrant it, in exactly the ways he recommends.