Bayesian Counter-Apologetics: Ten Arguments for God Destroyed

Bayesian counter-apologetics is the method of using Bayesian logic to turn every argument for God into an argument against God, simply by understanding how the logic of evidence works, and then reintroducing all the evidence theists always leave out when they attempt to make an argument for God. Which reveals the fact that all arguments for God, are really just exercises in hiding evidence.

I’ll illustrate this here with a brief application to ten of the most common arguments for God; after a brief survey of the applicable principles of Bayesian logic. All of which will serve as a handy guide in general. But it will also prepare you for a critical review of a leading apologetics book I’m planning for later this month. I noticed, as I read through that book, that two common tricks are being pulled, over and over again, to scam their readers: leaving evidence out; and ignoring how the debate is actually about what best explains the evidence, reframing the debate instead as about something else. Just so much hand-waving to distract the reader from not noticing everything that’s been hidden from them. A pernicious form of lying.


Basic Framework

All arguments for and against God reduce to hypothesis testing: We have a body of evidence, and need merely ask, which hypothesis is made more probable by that evidence? And how much more probable? (See The Improbability of God.)

Even arguments for the logical impossibility of a God always end up there, since all that those arguments really do is demonstrate that certain definitions of God are logically impossible. And that leaves us with the logically possible definitions, the remaining “God hypotheses,” that don’t run into those problems of logical contradiction. Like a God we posit not as literally omnipotent (a claim that entails logical contradictions) but as only omnipotent in the sense of having all powers it is logically possible simultaneously to have. (See The Impossibility of God.)

Likewise, all arguments for the logical necessity of God. The ontological argument, for example, only really demonstrates that if an omni-God exists, then he is the greatest thing conceivable; but that cannot resolve the question of whether in fact the greatest thing we can think of actually exists. That remains an empirical hypothesis. Similarly, presuppositionalism really doesn’t get you to God by any non-circular route; in the end it just reduces to comparing empirical hypotheses for why the universe obeys logic and mathematics reliably enough for us to use logic and mathematics to understand it (hence, see my analysis of the Arguments from Reason and Mathematical Universe). And arguments like “you can’t say the universe is unjust or evil without a cosmic standard of justice or evil” always just collapse into comparing hypotheses for why the universe doesn’t behave the way we desire, and why we have that desire instead of some other—since justice and goodness are simply descriptions of things we like and want, nothing more; to argue they are more, is an empirical hypotheses that again has to be tested out against the evidence.

Skeptical theism is similarly taken down the moment you frame the question as a probability. Yet one must always frame questions of fact in terms of probability. You can’t just say God “might” have an excuse for all the evils he permits, and it “might” be the case that we can’t know what it is or even conceive of what it is, “therefore” he does have such an excuse. You have to ask if that’s probable, and how probable it is. And that always gets you back to hypothesis testing.

For example, why do you think the probability of there being an excuse is even as high as 50%, when you can’t even think of any possible excuse, and the entity claiming the excuse could have told you their excuse but refuses to do so when asked? You wouldn’t buy that defense from a criminal in a trial. If a common murderer just said, “you can’t claim to know I didn’t have a fully justified reason to kill them, therefore you cannot find me guilty,” you’d laugh in their face. Not because the premise is false—in fact it’s totally true: they may indeed have a totally justified excuse that you can’t think of and they may indeed also have some other totally justified excuse for not telling you what that excuse is, an excuse for not telling you that you also can’t think of and that they also can’t tell you. You laugh not because that’s not true. You laugh because that’s extraordinarily, indeed absurdly improbable.

And yet a God has infinitely more ability to effect his wishes without adverse consequence than any common criminal, and infinitely more ability to tell us why he can’t. So that a God would have such an excuse must necessarily be infinitely less likely than that any common criminal has such an excuse. And if you intend to gainsay that conclusion, you are back to hypothesis testing: you are stuck proposing some hypothesis, that we now have to test against the available evidence, before we can assert it likely. And if it’s not likely, neither is God. Right back where we started. (See Stephen Law’s Pandora’s Box.)

So after you dance pointlessly around the logical maypole, you will always end up comparing hypotheses against the evidence for them. And that necessarily throws you onto the horns of Bayesian logic.

Bayesian Logic

Bayes’ Theorem is a mathematical model of all correct empirical reasoning (see Proving History, Chapters 4 and 6). As such, you can use it to analyze the validity and soundness of any argument for matters of fact. You can learn more about Bayesian reasoning from my blog: in the resources I’ve provided under Bayes’ Theorem: Lust for Glory and in many other articles under the tag Bayes’ Theorem. But the simplest way to put it is this:

The Odds a Claim Is True = The Prior Odds on the Claim Being True x The Odds of the Evidence on the Claim Being True Rather Than False

The odds of a thing happening means the probability of a thing happening divided by the probability of it not happening. If something is 80% likely to be true, for example, then the odds are 80/20, or 4 to 1 odds that it’s true (and thus 4 to 1 against it being false). Odds are thus always some ratio of probabilities.

So what Bayes’ Theorem says is that the odds that some claim is true are always the product of two other probability ratios: the “prior odds” (the “prior probability” the claim is true, divided by the “prior probability” the claim is false); and what is called the likelihood ratio or “the Bayes Factor,” which is the probability of the evidence if the claim is true, divided by the probability of that same evidence if the claim is false. Multiplied together, these two ratios give us the odds of a claim being true, which you can translate into a probability that the claim is true. There are other ways to formulate and work Bayes’ Theorem. I discuss them and work through examples in the subject of history in my book Proving History.

We are already doing this every time we reason or argue about any matter of fact. We are always reaching conclusions by assuming we know the prior odds of a claim, and assuming we know how much more likely the evidence is if the claim is true than if it’s false (or vice versa), and assuming we know what the logical consequences are of each of these assumptions. We just don’t realize this is the logic we are applying to the case.

Consequently, we are highly prone to unsound reasoning. We might be making unwarranted assumptions about the prior odds or the odds of the evidence; and we might be affirming an invalid probability even if our assumptions on both are correct. We might incorrectly conclude something is “very likely,” when in fact even on our own assumptions we should conclude it’s not. And we might incorrectly act as though the prior odds are respectably high when in fact they aren’t; or as though the evidence is just as we should expect when in fact it isn’t.

So we really should get good at understanding and checking our logic when we reason about facts, and recognize we are making assumptions about probabilities; take seriously the need to justify those assumptions; and make sure our conclusions actually follow from them.

Overall, the things we should learn from understanding this mathematical model for empirical reasoning are these:

Prior Probability: What has typically been the case before? In other words, is the claim being defended typical, or unusual, or even highly unusual, or indeed otherwise unprecedented? The more unusual, the lower its prior probability. And the lower its prior probability, the more evidence you need to believe it—according to the Rule of Evidence…

Rule of Evidence: How likely is all the evidence if the claim is true? How likely is all that same evidence if the claim is false? The difference between those two probabilities is the strength of the evidence for or against the claim—its strength for the claim, if the evidence is more likely if the claim is true than if it’s false; or against the claim, if the evidence is more likely if the claim is false than if it’s true.

For instance, if some evidence e is 1% expected on theism and 100% expected on atheism, then e entails a Bayes Factor favoring atheism of 100 to 1. In other words, that evidence then makes atheism one hundred times more likely than theism. And that’s if atheism and theism start out equally likely. It’s even more than that if atheism was already more likely. For instance, if atheism starts out twice as likely as theism, adding this e then makes atheism two hundred times more likely than theism (since 2/1 x 100/1 = 200/1).

Occam’s Razor: Occam’s Razor, simply stated, is “entities must not be multiplied beyond necessity” (entia non sunt multiplicanda praeter necessitatem). In terms of the logic of evidence, this translates to: if you must add an element to a claim in order to make that claim fit the evidence better (in other words, to make the evidence more likely on that claim than the evidence would be on that claim without that added element), then the prior probability of the claim is reduced by the probability of that element being true. And if you must add many elements, then all their probabilities multiply to reduce the prior probability of the claim.

There is no logically valid way to avoid this.

For example, if the evidence is unlikely on claim A, but becomes likely again if we add element X to A (so our claim now becomes X+A), and we have no reason to believe X is likely or unlikely, then the probability of X is 50% (or 0.50), and the prior probability of A is then reduced by 50%. In other words, P(A) becomes P(A) x P(X) = P(A) x 0.50. The prior probability is thus halved, just by claiming X.

This means that adding excuses for why a claim doesn’t fit the evidence usually does not rescue that claim, but makes that claim less probable, not more probable.

If, however, X is demonstrably true (if, for instance, we have plenty of reason to believe it is indeed true), then adding it to A does not significantly reduce the prior probability of A. For example, if P(X) = 99.99% (or 0.9999), then P(A) becomes P(A) x 0.9999, which is pretty well close to the original probability of A. On the other hand, if X is demonstrably very unlikely (if, for instance, we have plenty of reason to believe it is not true—or if in fact it almost never is, and no good reason can be adduced to think it would be), then adding it to A renders A very unlikely as well. For example, if P(X) = 1% (or 0.01), then P(A) becomes P(A) x 0.01, which is effectively reducing the probability of A by a factor of a hundred, making A a hundred times less likely than if we didn’t add excuse X.

The Cost of Making Excuses

Anytime someone (and that includes you) tries to “make excuses” for why the evidence does not come out as expected on what they are claiming, don’t let them get away with doing this at no cost (and that means also don’t let yourself get away with this, either). To be logically consistent, they must multiply the probability of what they were claiming by the probability of what they are now claiming. So, now, the prior probability of the claim itself (e.g., “God exists”) must be multiplied by the probability of each new excuse being true (e.g., “God has a reason, consistent with everything else we are claiming about him, for being completely silent in our current conversation”).

For example, when God, as merely defined, entails far more obvious evidence of his existence—such as a far different looking universe, a far better governed universe, far more (and far more consistent and universal) communications from God, and so on—such that the actual universe we find ourselves in is actually improbable if God exists, then even if we were to say the prior probability of such a God is 50%, the posterior probability (the final probability) that God exists will be less than 50%. Because all that evidence is then less probable than it would be if there is no God—since the absence of God perfectly predicts all the appearance of an absent God. (Which is the real reason atheism is justified.)

A believer cannot have this, of course. So they will try to make excuses for why God causes all the evidence to look exactly like the evidence would look if God didn’t exist. But those excuses are not known to be true. There is no evidence they are true. So at best they are only 50% likely to be true or false. And some might not even be that likely; if, for example, the excuses make no sense for an invincible superhero, then they make no sense for a god, either, and are therefore much less likely to be true than 50%.

And if they need more than one such excuse, these excuses multiply: for instance, if the believer needs to make three different assertions to get the evidence to be likely if God exists, and the believer can adduce no real evidence any of those assertions is true (nor can you adduce any evidence they are false), then they are each 50% likely to be true, but their conjunction (all of them being true at the same time) is far less than that. It is, in fact, 0.5 x 0.5 x 0.5 = 0.125 or 12.5% likely. This must now be multiplied by God’s original prior probability without these excuses, so we end up starting with a God whose prior probability of existing is, let’s say, 50%, but end up with a God whose prior probability of existing is 6.25% (0.125 x 0.50 = 0.0625). Even if that gets the evidence to be 100% expected if God exists, it doesn’t help. Because that same evidence is already 100% expected if he doesn’t exist, so the evidence causes no change to the probability that God exists. He simply then remains 6% likely. Which means likely not. So adding those excuses just ended up making God less likely, not more.

And of course assuming a 50% prior probability for just a basic god is already grossly illogical, since “supernatural beings” are not typical but extremely rare—in fact unprecedented so far, and therefore, at the start of any honest calculation, they are the least likely explanation of anything. That which you’ve never confirmed before ever to be causing or explaining anything, is the last thing likely to explain anything else you encounter. Honestly, that should be the first thing you ever learn from Bayesian logic.

Lambasting the Cherry Pickers

A believer will then try to cherry pick evidence they claim is unlikely unless there is a “God.” Don’t let them cheat like that. They have to account for all the evidence, not their own biased selection of evidence. They want to focus on fine tuning, for example, and ignore this universe’s extraordinary hostility to life. They want to focus on a single instance of someone being miraculously cured, and ignore the millions more who died horribly of that or other ailments. They want to focus on an unsourced legend about Jesus curing a single leper, and ignore the tens of thousands of children around him in Judea who died of cholera and other diseases during his ministry (indeed, horrifically, 50% of all children born then died of some illness before reaching the age of 5). They want to focus on unsourced legends that Jesus could turn people into disciples with a single sentence, while ignoring the fact that in the same legend he didn’t know about germs, and consequently advised people that they didn’t have to wash their hands and dishes before eating (Mark 7).

And so on.

Always remember to look at what all the evidence is, and then ask how likely that evidence is if there is a God…and how likely it is if there isn’t. The ratio between those two probabilities is how much the evidence counts for or against the existence of God. If the evidence is less than 1% expected with God, and essentially 100% without, then that evidence makes the existence of God a hundred times less likely (as that makes 100/1).

Not only point this out, but make an issue of it: call specific attention to the fact that they are doing this.

Point out to them explicitly that they have illogically cherry picked the evidence and had to ignore evidence to get the result they wanted, and that when we stop ignoring that evidence, we get the opposite result. They will then of course resort to the tactic of excuse-making. In which case, see my point above about the costs of making excuses. And again call specific attention to the fact that they are adding more improbable claims in order to rescue an already-improbable claim—and that this cannot logically succeed, but in fact makes their God even less likely.

Ten Arguments for God

Applying these principles to the ten most common arguments for God gets you these results:

(1) The Cosmological Argument: “Everything that begins has a cause” and “all existence began” and “only disembodied minds can precede the beginning of time” are all hypotheses. Not one of them ever proven likely. We don’t know if time is the sort of thing that can even have a cause; the notion is not even intelligible. If it began, time in fact seems necessarily causeless, since a cause is by definition what precedes an effect in time. Many other things may well be causeless, too. We only know how things we’ve seen in this universe, within time behave. We cannot infer from that how things behave outside this universe, or outside time.

Similarly, we only know this universe began. But we have no evidence that this universe is everything that exists (and theism already presupposes that it is not), or that time itself began with our universe. And we don’t even have any evidence that disembodied minds can exist, much less that they could exist before time began, any more than anything else could. And if we suppose God created time simultaneously with the beginning of time rather than ever existing before time began, then anything could do that, even something embodied, or mindless.

In other words, reduced to hypotheses, cosmological arguments get us nowhere, other than up the ass of random guessers pretending to be scientists, without a single iota of relevant data. Except that the only causes we’ve ever confirmed for anything for hundreds of years now have been godless physics. Which leaves us with extremely high prior odds that that’s what it is all the way down the line. Only evidence can change that conclusion.

(2) (3) Arguments from Design (Fine Tuning & Biogenesis): I show how to turn these arguments on their head as arguments against the existence of God in my chapter on design arguments in The End of Christianity. In short, if (a) we exist and (b) God did not design the universe, then (c) we should expect to observe several things, and lo and behold, those are exactly the things we observe; yet we do not expect to observe those things if God did design the universe. By definition that which is expected on x is probable on x; that which is unexpected on x is improbable on x. So if the evidence is probable if God does not exist and improbable if God exists, then that evidence argues against God, not for God.

As I explain in Merry Chrismas, God is Still a Delusion and 20 Questions, the only way we could exist without a God is by an extremely improbable chemical accident, and the only way an extremely improbable chemical accident is likely to occur is in a universe that’s vastly old and vastly large; so atheism predicts a vastly old and large universe; theism does not (without fabricating excuses—a bankrupt procedure, as I already explained above).

Similarly, the only way we could exist without a God is by an extremely long process of evolution by natural selection, beginning from a single molecule, through hundreds of millions of years of single cells, through hundreds of millions of years of cooperating cells, to hundreds of millions of years of multicellular organisms; so atheism predicts essentially that; theism does not (without, again, piling on excuses).

Likewise, if chance produced this universe, we should expect it to be only barely conducive to life, not almost entirely lethal to it (as in fact it is), since there are vastly more ways to get those universes by chance selection, than to get a universe perfectly suited to life throughout (indeed, among all possible universes that can be chosen at random, barely conducive universes exceed perfectly suited universes by countlessly many trillions to one). Design predicts exactly the opposite (again, without a parade of convenient excuses).

Even if we grant fine tuning exists, there are two ways it can happen: chance accident, or intelligent design. And what theists don’t want to admit, is that all the evidence actually points to chance accident. Quite simply, the universe and the history we observe is 100% expected to look that way if chance accident caused it; but its looking that way is not at all probable on design. So here we find that not only do the prior odds strongly support atheism (since, as I just mentioned previously and will further explain in a moment, all the way up to now we’ve only ever found natural and chance causes of anything), but from the evidence of life and the cosmos, the Bayes Factor also strongly establishes atheism. (In fact, all this is far better evidence for multiverse theory than for monotheism.)

(4) Argument from Consciousness: I also dispatch this in Bayesian terms in TEC (pp. 298-302). Theists try to focus just on the fact that conscious phenomena are weird and not yet scientifically explained, “therefore” God is the best explanation of it. But that’s a non sequitur. When we don’t know an explanation, the most likely explanation will be the one that has most commonly succeeded before when we thought something couldn’t be explained. And that’s always turned out to be physics, not God. Prior odds thus strongly favor physics, not theism, for anything as yet unexplained. We need evidence to conclude otherwise. And that’s where theists try to ignore all the pertinent evidence. When we bring all that ignored evidence back in, atheism, not theism, ends up most likely.

For example, that we need brains to generate conscious phenomena is quite unexpected if God exists. Because if God exists, disembodied minds can exist, and are the best minds to have, therefore we should also have disembodied minds. Indeed, there is no inherent reason it would even occur to a god to make our minds out of brains at all (without, again, a pile of convenient excuses). Whereas if God does not exist, the only way minds could exist is as the output of a complex physical machine that evolved slowly by natural selection over hundreds of millions of years from ultra-simple worm-brains to fish-brains, lizard-brains, mammal-brains, monkey-brains, ape-brains, hominid-brains, and eventually human brains. Just as we observe.

Therefore, the fact that thought is dependent on complex evolved brains, which are physical machines, and which also inefficiently exhaust oxygen and energy, and place us in needless risk of injury and death, and intellectual malfunction, due to their delicate vulnerability and badly organized structure, is exactly what we expect if there is no God, but not at all what we expect if there is. The Bayes Factor once again supports atheism, not theism. (For a formalization of this point, see AMBD. I also discuss six points in its favor in Sense and Goodness without God.)

(5) Argument from Reason: I also cover this in TEC (ibid.), and elsewhere I have exhaustively refuted every version of it. But it all reduces to a simple Bayesian case against God: if God did not design us, our innate reasoning abilities should be shoddy and ad hoc and only ever improved upon by what are in essence culturally (not biologically) installed software patches (like the scientific method, logic and mathematics, and so on), which corrected our reasoning abilities only after thousands of years of humans trying out different fixes, fixes that were only discovered through human trial and error, and not communicated in any divine revelation or scripture. But if God did design us, our brains should have worked properly from the start and required no software patches, much less software patches that took thousands of years to figure out, and are completely missing from all supposed communications from God.

Thus, observation confirms that the actual evidence of human reason is far more probable if God did not exist than if he does. Thus, even the Christian’s own Argument from Reason argues that God does not exist, rather than that he does. Because once again, when we bring in all the evidence, the Bayes Factor strongly supports atheism.

(6) Argument from Religious Experience: This goes the same way. By falsely selecting and ignoring evidence, the believer tries to turn their religious experience into evidence for (their own) God, while ignoring everyone else’s religious experience that contradicts theirs. Hence when we actually bring back in all the evidence, the conclusion goes the other way. (This is superbly argued from every angle and against nearly every possible excuse by John Loftus in The Outsider Test for Faith, with an excellent response to his remaining critics in The Christian Delusion, Chapter 4.)

We have evidence of divine communications going back tens of thousands of years (in shamanic cave art, the crafting of religious icons, ritual burials, and eventually shrines, temples, and actual writing, on stone and clay, then parchment, papyrus and paper). Theism without added excuses predicts that all communications from the divine would be consistently the same at all times in history and across all geographical regions, and presciently in line with the true facts of the world and human existence, right from the start. Atheism predicts, instead, that these communications will be pervasively inconsistent across time and space, and full of factual errors about the world and human existence, exactly matching the ignorance of the culture “experiencing the divine” at that time. And guess what? We observe exactly what atheism predicts; not at all what theism predicts. And again, adding excuses for that, only makes theism even more improbable.

Thus, the actual evidence of religious experience is highly probable on atheism, and highly improbable on theism. This disparity in probabilities entails the evidence of religious experience argues God does not exist, not that he does.

(7) Argument from Miracles: This works the same way, too. Atheism predicts random good luck and bad luck will be observed, and therefore anything we can confirm happened that seems miraculous will be physically explicable (because, not really miraculous) and rare (because, random). Without a parade of excuses, theism predicts miracles will be commonplace and physically inexplicable (e.g. Christian healing wings in hospitals would exist where amputees have their limbs restored by prayer, or anything like that; yet we observe not a single thing like that). Likewise, atheism predicts the only miracle claims that will “survive scrutiny,” are claims that are never reliably investigated; and that every time a miracle claim gets proper scrutiny, it dissolves. And lo and behold, that is also what we see. Thus, again, what we observe is exactly what is expected on atheism, not at all what we expect on theism. So even the evidence of miracles refutes theism and confirms atheism.

(8) The Moral Argument: If atheism is true, it is still true that: (a) we all want to live in a just and kind and honest world, which desire is sufficient reason for us to try and create one (basically, if you don’t want the world to be amoral, then you already have sufficient reason to be moral); (b) we are social animals, and social animals need to be just and kind and honest to work together well, and they need to work together well to optimize survival and realize their goals (indeed, one need only compare moral societies with immoral societies to see the difference, which observation is more than sufficient reason to be moral); and (c) more and deeper joy and satisfaction comes from feeling compassion with others (and thus sharing their joys) and loving truth (loving falsity and falsehood, by contrast, will always result in embracing self-defeating or self-frustrating behaviors; while compassion is necessary to vicariously experience the joy and happiness of others).

Thus, atheism predicts three motivating reasons for people to develop a common morality centered around compassion, honesty, justice, and cooperation. But more importantly, atheism predicts that moral rules will only come from human beings, and thus will begin deeply flawed, and will be improved by experiment (after empirically observing the social discomfort and dissatisfaction and waste that comes from flawed moral systems), and consequently that will happen only slowly over thousands of years. And that is exactly what we observe. Just look at the examples of slavery and the subordination of women in Bible.

By contrast, theism predicts a universe directly governed by justice-laws, or a kind and just stewardship, or the enacting and teaching of divine justice and mercy, everywhere, from the start. But we observe no such laws built into the universe, and no stewards or law enforcers but us, and no perfect moral code has existed anywhere throughout history; the best moralities have always just slowly evolved from human trial and error (see Pinker’s Better Angels of Our Nature and Shermer’s The Moral Arc). Thus, the evidence of human morality (it’s starting abysmal and being slowly improved by humans over thousands of years in the direction that would make their societies better for them) is evidence against God, not evidence for God.

(9) Argument from Meaning of Life: “It would be better if I had a million dollars; therefore I have a million dollars” is not even a logically valid argument to start with. So the only way to get from “life must have some meaning” to “therefore God exists” is with two hypotheses: that life does have some meaning; and that only a god could provide it. But there is no evidence that second hypothesis is true—we readily and easily assign meaning to things all the time, by ourselves, with no help from anyone. And if you define “meaning” as “cosmic external meaning,” and not “what we as individuals value about our lives and the lives of others,” in an attempt to get that second hypothesis to be true, there is then no evidence the first hypothesis is true. Either way, you can’t get to the conclusion.

All the evidence of history and science weighs heavily for the conclusion that we are mortal, and that we actually value our lives because of that, and not because we are immortal—which would actually render this life cheap as dirt (since death would cost us nothing, and life is better and vastly longer on the other side of it). Life would still be valuable if we are immortal, but not because we are immortal. It only has value because it can be lived. Which even a mortal can do.

In fact, the Prior Odds and all Bayes Factors render only one conclusion probable for those who want to live forever: only future  human-made technology is likely to get you that outcome. In the meantime, life only has meaning because you value it, and because of the things you value about it. It’s meaning comes from you. That being so, does not increase the probability of a god one whit. To the contrary, that we are mortal, and throughout history have always invented our own meaning for life, and always different people have valued different things about it, is exactly what we expect if there is no god. Whereas, excuses aside, it’s not all all what we expect if there is a god.

(10) Argument from Superman: Every religion has its own Superman argument. Moroni, Jesus, Mohammed, Moses, Buddha, even Lao Tzu, are all claimed to have proved their religious teachings supernaturally true by miraculous demonstrations of their power. “Our Superman exists; therefore our God exists.” All these arguments collapse the same way: when you put all the evidence back in, the Bayes Factor and Prior Odds both guarantee they are all just made up stories. And not being true, they fail as arguments. A real God would not produce stories that look just like they were made up, and then present no adequate evidence for them being true. I illustrate the Bayesian logic of this in detail for Christianity in The Christian Delusion (“Why the Resurrection Is Unbelievable”) and even more so in The End of Christianity (“Christianity’s Success Was Not Incredible”).


You can see by now how any argument for God can be turned around into an argument against God by (a) including all the evidence the theist is conspicuously ignoring and then (b) showing how this entails a strong Bayes Factor against the existence of God (or, or also, a strong Prior Odds against). Theism is built on hiding evidence. Hiding the evidence of history, that makes gods the least likely explanation of anything, and then hiding the specific evidence that refutes each and every reason to believe in God. Bayesian counter-apologetics exposes and corrects all this.


    1. Marc Miller January 10, 2017, 6:43 pm

      Imagine the financial benefits of buying up churches as historic buildings and turning them into apartments, museums, meeting halls, etc…

  1. Justin Legault January 11, 2017, 7:58 am

    Great article to help with counter-apologetics. On an unrelated note, I’ve always wondered about the passage in Isaiah 7:14. Apart from the virgin mistranslation in Almah/Betulah. Does the “Immanuel” argument hold water? If the messiah was supposed to be called Immanuel then does it not refute “Jesus” as the messiah?

    Thanks 🙂


    1. Immanuel means (as even Matthew makes a point of telling us) “God is with us”; so Matthew has Jesus called that at the start of his Gospel (1:23), then has Jesus declare himself that at the end of his Gospel (“I am with you,” 28:20). (See OHJ, p. 464.)

      In other words, Jesus was called “God is with us.” He is called that by Christians at every eucharist ritual, by the declaration that Jesus is with them. Matthew is thus cleverly playing on the words and their meaning, to make Christian ritual itself a fulfillment of prophecy.

      1. Justin January 11, 2017, 6:14 pm

        Ahh that clarifies alot, thank you! Im currently reading OHJ, at page 230 so far. Such a massive book! Thanks!!!

  2. Very clear and useful information, per usual! I really enjoy your writing because while being informative, it always retains a good sense of humor as well, thus being entertaining also! Keep up the good work. You are greatly appreciated!

  3. Someone asked why I only show the improbability of a generic theism and not the whole specific biblical sin-story-based Christian worldview. The answer is because the probability that some god exists is logically necessarily greater than the probability that a specific god exists. So if even a generic god doesn’t exist, necessarily a specific one doesn’t either.

    The only way to escape that outcome is to argue that the specific additions to God you propose make that God more likely than a God without those additions. Which is exactly what I explain are “convenient excuses” that reduce the probability that God exists, rather than increase it. So asking about a more specific God always makes things worse, not better. Unless you have evidence for your specific additions. Exactly as I explain in the article.

    And the fact is, there is less evidence that the Bible is a reliable book and that its specific worldview makes any sense, than there is for a more generic God, not the other way around. In fact, the evidence is even more strongly against that God than a generic God. Genesis simply fails as science. The Exodus never happened. Deuteronomy and Leviticus attribute absolutely evil actions and judgments and laws to its depicted God. And the worldview of Adamic sin makes no sense of natural evils and indeed not even of human evils, since a compassionate God would protect people from evils of both kinds, not gleefully let them suffer them.

    Now, if you wish to propose your God is evil and that we should bow to that evil God, you would actually have a claim more probably true than the entire Christian religion. But alas, it still is probably false, for many of the reasons just enumerated in my article.

    1. Richard – Would I be correct in assuming the main reason you don’t argue against any particular religion’s “god” is because of Occam’s Razor? For example, the god of the bible has so many assigned improbable attributes that the probability of his existence would be essentially zero?

      1. That’s also true. It’s the argument made by John Loftus in “Christianity Is Wildly Improbable,” Ch. 3 of The End of Christianity.

        I usually allow for a minimally defined God, such that it is the least a God is proposed to be, that anyone actually is willing to believe in. Even that God is super complex (in terms of the specified complexity of any hypothesis affirming his existence). But so is any god—even gods who are much simpler than the ones people insist on believing in (apart from semantically meaningless gods, e.g. god as “a brain’s feeling of awe in the face of existence” and such). Ironically, simpler gods, whose prior probability is therefore still much greater than the minimally-defined-actually-believed in gods, most everyone is sure don’t exist—including all Christians and Muslims, for example; which betrays the truth that their belief rests more on a desire that a certain god exist, than on actual evidence that any god does.

  4. Hi Richard. How would you respond to someone who responds that since Eden both reason and (human and non-human) nature is fallen in some sense, and therefore would exhibit some of the characteristics you see as evidence in favour of atheism (or at least not for theism)? Regarding religious experience, someone might respond that demons may sometimes masquerade as angels of light, or perhaps even other gods, in order to deceive us (an argument with an ancient pedigree as you’ve pointed out in other writings). Personally I’ve heard both responses. I’d love to hear your approach.

    1. The second claim runs into the bull’s horns of: How do you know which ones are the real angels and which the fake? (Loftus’s Outsider Test problem.) But also, it would be a failed prediction, because a real God would not allow people to be deceived using the very same channel by which he reveals himself. You have to fabricate a litany of unevidenced and improbable excuses to get a God who would do such an absurd, self-defeating, and stupid thing. So the Bayes Factor kills that hypothesis dead. Unless you gerrymander it, and then the Prior Odds kill it dead. Whereas the natural godless explanation fits perfectly, and thus has a high Bayes Factor; and since all past phenomena have turned out natural as well, the Prior Odds also highly favor it.

      Which gets to your first question, the Eden nonsense. See my comment above on why attaching the entire collection of improbable “hypothesis extensions” that is the “traditional biblical worldview” onto your God, builds an extraordinarily elaborate and highly specified hypothesis that by that fact alone starts out with an absurdly low Prior Odds. Because worse than that already being the most improbable series of unevidenced excuses, it makes a huge number of predictions falsified by the evidence (the Eden story and timeline is abundantly refuted by all pertinent scientific evidence). So it also has an extremely low Bayes Factor. And to explain all that evidence away requires creating a new series of even more improbable excuses, thus trading that poor Bayes Factor for an even poorer Prior Odds. There simply is no logical escape.

  5. Someone asked “if it is the case that our reasoning is ad hoc, so too is the ‘cultural software’ our reasoning created. How do you get out of that dilemma?”

    The answer is that the cultural software isn’t ad hoc. It was intelligently designed, and empirically tested. More specifically, it underwent selection effects for a truth-finding process specifically, unlike our brains, which only needed to be “good enough” to increase differential reproductive success.

    Countless other ways of knowing were culturally engineered and tested, but only the ones that worked became cultural universals, and allowed the advance of a technological civilization, ultimately even landing men on the moon. Capabilities we were not biologically selected for. But when truth (measured by success) is consciously targeted for selection, it can take but tens of thousands of years of random trying to hit on the correct method. The biological lever was conscious awareness: that one ability, which was naturally selected for its benefits to DRS, made it possible to use that awareness to intelligently select truth-finding procedures.

    We know we were successful because if we weren’t, the procedures would not have accomplished all that they have (e.g. landing men on the moon). Indeed, science continues to undergo selection effects grinding away error-prone procedures (e.g. the invention of the double-blind experimental model greatly improved the truth-finding success of the scientific method; and this is known because of its observable success, e.g. in predicting observations, which would not have happened had it not been an improvement in discovering the truth and avoiding error).

    Ordinary ad hoc observation is all that is needed to verify whether a procedure increased our ability to predict further observations. No further advanced intelligence is required.

  6. Jonathan May 18, 2017, 12:53 pm

    Excellent article, Richard! I did want to comment on one thing in this article though that sounds erroneous to me: “indeed, among all possible universes that can be chosen at random, barely conducive universes exceed perfectly suited universes by countlessly many trillions to one”. This seems like a very difficult statement to defend even though intuition seems to indicate that the general idea is true. Assuming an infinite set of possible universes, the idea of a ratio of size of one infinite subset to another infinite subset becomes undefined. Reasoning about infinities is full of logical pitfalls. I think the only valid argument that can be made along these lines is the argument Victor Stenger makes in God: The Failed Hypothesis that if a god created the universe, it could have chosen a universe where life could have arisen naturally such as the one we are in or it could have chosen one where life could have only arisen through supernatural means. This means that the fact that we find ourselves in a universe where life could have arisen naturally gives a higher consequent probability on atheism than on theism since the consequent for a universe allowing life through only natural means since on the theistic hypothesis there are additional possible universes we could have found ourselves in. Though of course given the infinite nature of the sets quantification is impossible.

    1. Undefinable is not the same thing as impossible. That’s why the method of exhaustion works. You can do the math for an ever increasing n, approaching infinity, and graph it. You can even do it with summing infinitesimals. It’s called calculus. Either way, the area under the graph is the proportional probability of an event falling on that point on the infinitely graphed line. It can thus easily and validly be shown mathematically that the probability of a perfectly suited universe if throwing a dart at the graph at random, is virtually zero. Exactly what it is is what can’t be defined (we do have a science of infinitesimals, but it isn’t developed to the point of solving that specific problem). But that it is less than any arbitrarily defined finite probability is necessarily the case.

      Hence when I say the number of universes thus graphed that are barely conducive to life is “countlessly” more than the universes on that graph that are not, that is a perfectly valid, and logically and mathematically demonstrable fact. The “undefined” part is exactly what “countlessly many” means. But that we can’t define exactly what that means, does not make the statement untrue. It remains the case that it’s countlessly many more. And that entails the probability of a random selection of a point on a graph that’s in that infinitesimal domain, is itself infinitesimal: it has no finite value. But having no finite value, does not mean it can be “any” value. It cannot be above 1% for example. Nor in fact, as you can formally prove, can it be higher than any finite probability you pick. Which is what makes the exact probability undefinable: whatever it is, it’s some cardinality of infinitesimal. We don’t know which, because we lack the knowledge of how to calculate it. But it’s still an infinitesimal. And so we can perfectly well say that it is. It’s being undefinable has no effect on that fact.

      (And that’s even assuming the section representing such universes is an infinitesimal on the graph. If it were finite, it would be definable, and thus just as demonstrably small.)

      1. Jonathan May 21, 2017, 12:12 am

        It seems like I don’t fully understand how you view the set of all possible universes and the subsets of barely conducive universes and perfectly suited universes. I’m assuming we’re imagining a hypothetical hyperspace of dimensionality equal to the number of parameters for a universe and that each of these subsets would correspond to different regions or even perhaps sets of individual points within the hyperspace.

        It appears from your response that you think that the set of barely conducive universes is infinite and that the set of perfectly suited universes is infinite too, and herein lies the first problem problem. To get pedantic, it would technically only be correct to state that one set is uncountably many more than another if the smaller one were finite. So if nothing else the statement needs to be rephrased in terms of probability, to something like, “it is almost surely the case that we would get a barely conducive universe rather than a perfectly suitably universe” – see Wikipedia article for almost surely.

        Assuming the statement were better phrased to make a statement about probability, unless we assume that one of these sets takes up a finite volume within hyperspace while the other doesn’t or that their volumes are of different dimensionality, we can’t use dartboard reasoning to produce a meaningful answer as to the probability of getting one or the other. If both sets take up infinite volume the dartboard reasoning doesn’t work and at that point we face the fact that both sets are of the same cardinality, that cardinality being the cardinality of the reals since a hyperspace of a finite number of dimensions made of real value coordinates has the same cardinality as the cardinality of the reals. Assuming both subsets have the same volume (infinite) and the same cardinality, it’s hard to say anything at all useful about their relative probabilities.

        Just because we have Calculus and can take some limits doesn’t mean that all limits are defined. It’s not clear that infinitesimals or the method of exhaustion apply to this. I actually think that Measure Theory is what we need to address this question, and I’d guess it’s more than well enough developed to address this issue. This type of question doesn’t strike me as something that mathematicians wouldn’t have been thinking about in the 1800s.

        Anyway, maybe I’m being overly nitpicky and there is some way to save the spirit of the statement that I’m not thinking of, but I frequently hear incorrect things about infinities in debates related to religion on both sides of the debates and felt like commenting.

        1. You don’t seem to understand how transfinite sums work.

          There are infinitely many points in an inch. There are infinitely many points in two inches. Two inches is still twice the size of one inch. The fact that both sums are infinite (“uncountable”) does not magically make one inch equal two. This is exactly what calculus demonstrates (and it was demonstrated by Archimedes thousands of years before, and not just by his method of exhaustion, but his actual system for summing an infinite array of infinitesimals, the foundation of calculus).

          The same happens when you are measuring “outcome space” rather than “physical space.” And again you can show this by exhaustion: you can assuming the total number of possible universes is finite (not infinite) and calculate the ratio between barely conducive to conducive universes in that finite set. Then you can multiply the finite space of possible universes by a billion. And repeat the math. And by a billion again. And repeat the math. You can keep doing this as long as you want to, and watch where the trajectory is heading for the resulting ratio each time. Calculus then lets you repeat that procedure infinitely many times, by calculating where the trajectory of that ratio ends up at the limit (of infinite possible universes).

          That we cannot calculate what the exact ratio is “at infinity” has exactly no effect on this. We can still say the ratio approaches zero as n approaches infinity (as the number of possible universes approaches the limit, of infinite possible universes).

          You do understand all this, yes? Or do I need to show you a physical graph? And how probability is then calculated as an area on that graph? (As in all statistics, even standard bell curve statistics, which also assume tails of infinite extension, with a finite n standing in as a sample taken of a hypothetical infinite possibility space.)

  7. Jefferson June 21, 2017, 4:35 am

    “If something is 80% likely to be true, for example, then the odds are 80/20, or 4 to 1 odds that it’s true (and thus 4 to 1 against it being false). ”

    Correction: The odds are 5 to 1, because 1/4 = 25% and 1/5 = 20%.

    1. Actually, no.

      5:1 would require six parts (5 + 1 = 6). There are only 4 parts (80/20 reduces to 8/2 which reduces to 4/1, for five parts: 4 + 1 = 5).

      This is a common mistake. Converting odds to percentages confuses a lot of people this way. I agree it’s confusing, so I’m sympathetic when people err. But no, 80/20 is 4/1.

      To confirm, use this odds calculator.


Add a Comment (For Patrons & Select Persons Only)

This site uses Akismet to reduce spam. Learn how your comment data is processed.