This was a fascinating read and honestly a bit bracing. What I appreciate most is the insistence on following the math even when it leads somewhere uncomfortable. For decades we were told the science was settled, yet very basic probability questions were never answered in a serious way. You do not have to buy every conclusion here to see that the challenge is real and overdue. If nothing else, this reminds us that science is supposed to be about explaining reality, not defending sacred cows.
So true. One of the books we had to read used in defense of evolution: “It is exceedingly difficult to come up with 2 + 2 if you don’t already know the answer is 4.” and “in order to understand, you must first agree”. Those never sounded very scientific and certainly not defensible for any subject other than biology or religion.
Exactly Kristen, When “you have to agree before you can understand” becomes the rule, that is not science anymore, that is catechism. In every other field you are allowed to question the premises. Funny how that rule only seems to apply to biology and belief systems.
Science is never settled. That's what's so good about our quest for knowledge.
Yes, this was a challenge that we were not supposed to undertake because if evolution 'theory' is proven to be rubbish then people are faced with the uncomfortable truth that God is real. (Uncomfortable for them, that is)
Exactly. Real science invites challenges, it does not shut them down. And yes, I think a lot of the resistance is not just academic, it
is existential. If certain assumptions fall, people have to revisit much bigger questions, and that makes many far more uncomfortable than any debate about data.
If anything, the process of AI "reinforcement learning" just underscores the scale of the problem here. Just look at any of the examples online of people training an AI to play a very simple video game implementation of something like bowling, or "Hide and Seek," etc. It generally takes millions upon millions of iterations (if not billions!) before the AI starts to get good. More iterations than the Earth is years old in some cases.
If each iteration were the equivalent of a biological generation, some of these programs would take longer than the generally accepted current age of the universe to get where they end up. Which is often still pretty bad.
And that's just to get a simple program to play a stupid video game. When the "selection pressure" is quite deliberately designed to achieve the best results possible.
Yet we're somehow supposed to believe that biological organisms could emerge purely as a function of random, unguided natural selection?
"And that's just to get a simple program to play a stupid video game. When the "selection pressure" is quite deliberately designed to achieve the best results possible."
With artificial selection by intelligent overseers. Who choose self-learning algorithms over hard-coded implementations, and make adjustments to the simulation framework to keep the simulations on-track towards the end goal of an expert AI player.
This book seems rich in ideas, and I look forward to reading it. One caution though is Bernoulli’s Fallacy. That’s when, like the Astronomers below, one mistakes 1-Pr(Data| Hypothesis) for Pr(Contrary Hypothesis| Data).
This is always a fallacy. It doesn’t always mean that that Pr(Contrary Hypothesis| Data) isn’t large for other reasons. But you can’t get there from here.
Improbable data conditional on one hypothesis says nothing direct about how probable or improbable any other hypothesis is.
But that “probability zero” he gets (rounded down) is only one of the ideas here.
Can you elaborate what you mean? I feel as though I am at the brink of understanding the implication of what you are saying but not quite understanding it.
In plain English, what he said is "Just because Hypothesis A is proven false, does not mean that Hypothesis B is true". Which is so self-evidently correct a statement that it doesn't really need to be said.
To make it even simpler, this would be the typical cope from Darwinists: "Even if we're wrong, that doesn't make you right! Nyah nyah!". It's childish to the point of being banal.
It's the framing of the hypothetic space as a binary choice, which it may not be. You have disproven Hypothesis A, this does not automatically prove Counterhypothesis B. In the current discussion, disproving the Natural Selection of Random Mutation Hypothesis does not automatically prove the IIntelligent Design Hypothesis. The Intelligence Design Hypothosis must stand on its own, independent of the Natural Selection of Random Mutation Hypothesis.
The unlikelihood of data assuming some hypothesis like ‘random’ change says nothing about whether that, or any other hypothesis is probable or improbable.
The probability the guy’s ticket who matched the Powerball the other week was about 3 in a billion. This is not proof against the hypothesis that the causes of the balls going up whatever tube they go up, or whatever, did not in fact go up that tube, because of course they did.
My point about the BF is only for this one part of the book. The other evidence against the Neo-Darwinian synthesis seems much stronger. And I’m sure I agree with a lot of it.
That being said, it appears that he's doing more to *disprove* one theory, than to prove another.
As he said in the interview:
# "...Where do they go from here intellectually?
Open your mind to the manifold possibilities. Darwin has been an anchor on biology, science, and philosophy for 150 years. We’re finally free to explore the real possibilities before us and work toward gradually discovering what is actually the true nature of our origins."
Fortunately for the anti-Darwin position, it is wholly unnecessary to establish one's own hypothesis and defend that. It is sufficient to prove that Darwin is wrong, and any question of "Oh yeah? Well, if Darwin is wrong, what's YOUR hypothesis?" can be answered simply, and gleefully, with "I don't know, but it isn't Darwinism".
Probability Zero doesn't set out its own position. It is entirely focused on attacking the Neo-Darwin position. As much as some people hate it, neutral is an entirely valid position to take in a debate.
Science loving folks often like to talk about falsifiability and "Not even wrong".
The scientific theory of evolution by natural selection makes falsifiable claims that are testable against current genetic science. The mathematical results based on scientific observation demonstrate that TENS is wrong.
New Atheism seems very 2000s and cringe. The honest atheists now believe we’re in a computer Simulation — also cringe but an opportunity for them to intellectually accept theism (and eventually Christ.)
New Atheism became cringe due to the fallout from the Atheism+ debacle, but it wasn't actually refuted. Meanwhile, the theists have assumed their victory is permanent and have relaxed even further.
Expect Atheism to come back, this time strengthened by an understanding of Bayesianism and game theory. I've already seen early indications of that, and it looks like when it does the theists won't know what hit them. Possibly we'll end up with something similar to what happened during the late 19th century/early 20th century, when all the 'serious' theologians ended up inventing and retreating to theological liberalism.
Unfortunately, the current slate of the theist public intellectuals are in no position to take advantage of this fact. The ones who aren't outright audience captured grifters, are functionally LLMs trained on the works of St. Thomas Aquinas.
Now while Thomas himself was a brilliant thinker, his modern followers exhibit the problems LLMs typically exhibit when confronted with inputs outside their training data, i.e., all novel intellectual or technological developments since the 13th century. As such, by the end of the 21st century, "Thomist" is likely to acquire the same connotation as "dunce".
By a felicitous coincidence, today I came across Philip K. Dick’s famous 1977 Metz speech. In the context of Probability Zero, the possibilities are intriguing.
I was moderately intrigued by the premise of the book right up until the author began disclosure regarding his reliance on and appreciation for the assistance provided by AI. If you can't write it alone or with the help of other human beings, umm, then don't write it? Especially given that he appears to be fully aware of the inherent problems with AI fallibility and bias. Unfortunately, I can't have a high degree of confidence in conclusions drawn by physicists, given the nature of their use of dark matter to fill the gaping holes in much of physics theory (as an example). I am hoping that Day/Beale/AI devoted a significant portion of the book to discussing the validity of the 202,500 generations, 1,600 generations, and presumed mutation rates used as the premises for his book. I agree with him that a simple average value is a powerful tool, but only if the data used in the calculation is of pristine quality. I'm also curious if Day/Beale offers any more mathematically plausible explanations for life's progression on Earth than evolution? At first glance, this seems like it fits in with the wave of people now decrying the evil materialist perspective and praising the wonderfully rediscovered spiritual view. Or, dare we hope, does he perhaps offer a refreshing synthesis of the two that surmounts both? Thanks for the review and perhaps I'll crack open my moldy piggy bank and give it a read.
LLMs are a valuable tool, but when he describes the contribution as all of the math and most of the research, and then how he had to push back and redirect when Claude didn't agree, I got a funny feeling. Wondering if anyone else was discussing that maybe this was just building nonsense on a tower of AI musings, I searched for "vox day ai psychosis". The second result was "AI Psychosis is a False Diagnosis" by... Vox Day.
Sadly, this means that Day now has a gigantic hill to climb, showing that his work isn't full of AI hallucination.
With all due respect, you guys are quite entirely out of date with regards to AI.
If you're not dumb enough to use AI as a black box that just requires pushing a button, then AI hallucinations are not a problem. It's a simple matter of telling it to keep you apprised of the context window. You have to work within their limitations, of course, and you have to check the math, but anyone who isn't using AI at this point is going to be left far, far behind those of us who are.
AI is essentially turbo-charging your own mind if you have original ideas and you know what you're doing. And it produces nothing but slop and nonsense for those who rely upon it as a crutch instead of using it for what it is, which is a very effective, if occasionally unreliable, tool.
Vox is also one of the first thinkers to publicly acknowledge his use of AI as a co- collaborator. No doubt others are cribbing from AI extensively, but few are making their use known. This disclosure and lack of intellectual vanity gives Vox the ability to punch “above his weight” on sophisticated math and scientific arguments that could otherwise be easily be dismissed by critics. As Vox acknowledges, the work could not have been accomplished without an AI collaborator, but he also recognizes the clear role of the human thinker in guiding the AI. They are incredible analysts and research assistants, but ultimately they are still rather dull thinkers - and they need that human nudge and redirection to truly innovate. The Obsidian Mirror project has observed a similar phenomena, where less than 10 pages of human curated “future history” has been backfilled with hundred of pages of supporting expansion and elaboration by AI. The result is AI generated world building more imaginative and nuanced than most published SF today (admittedly, this is a low bar).
Can confirm. I'm using it for writing & research, and it's an incredible productivity boost. It's not perfect - but nor am I - and you do need to learn how to use it effectively, as with any tool. I still like doing writing myself, and I'm still a better quality writer than Claude, but on a really good day I can crank out 10k words. Claude can do the whole book.
(On a side note, I've not found Claude to be particularly woke either.)
I don't really disagree with your concrete points, here. I'm pointing out that there's a trope of "guy has unorthodox ideas, asks LLM to help him, has to push past gentle disagreement, but finally gets the LLM to agree that he's brilliant and groundbreaking and to help him produce a vast opus sure to overturn established Science...".
The trope is irrelevant here. It's not about the LLM agreeing with me, it's about the LLM simply running the numbers. And I've run everything by the other LLMs cold. Gemini 3 Pro rated the book's rigor at 9.7 compared to 1.5 for The Selfish Gene.
There is a huge difference between using an AI for ego-stroking and using it for systematic mathematical stress-testing.
I also had 20 mathematicians and physics PhDs read the draft text and the book was endorsed by one of the world's top physicists, who wrote the Foreword.
Thanks for chiming in on our comments, but I am still not convinced regarding the desirability of AI collaboration in this regard. As simplistic as it may sound, can't AI be viewed as the intellectual equivalent of steroid use for sport performance? While one can rightfully lay claim to the originating thoughts, if you then incorporate AI as an accelerant, you have cheated yourself of what could have been a wonderful achievement. It is interesting and a bit alarming that so many folks have convinced themselves, or been convinced, that the convenience of AI is imperative, lest you be left "far, far behind." Following this line of reasoning, I can hire in numerous subcontractors, provide them with a design, and, so long as I check their work, I can later invite my friends to come see the new house that "I" built. AI promises to rob you of the fulfillment gained through your individual sustained efforts, the resilience gained by conquering hardships, and the pride of being able point to something beautiful as being created entirely by you. The joy of mountain climbing is not defined by standing at the summit, but is found in the climb. If I actually only ascend some part of the way up, but at other times have someone carry me (but watch their steps, of course), what should I feel eventually when we reach the peak? We have forsaken our regard for true craftsmanship and quality products and now feast incessantly on the empty calories of mass produced slop (a broad generalization and not referring to your work, sir). At the risk of sounding overly critical, I write this while sitting in the house that I alone actually built.
I've written 950-page epic fantasy novels. I've also written several AI-assisted novels. Look at the reviews; there is essentially no difference from the reader's perspective.
The end result is all that matters. Some books I don't use AI. But most, I definitely will.
I absolutely could not have written Probability Zero, the two follow-on books, and the related 12 science papers without AI. Not in ten years.
That’s like arguing that DIY should only be done with hand tools; that using power tools is “cheating” or somehow robs you of the satisfaction of doing it the hard way.
AI is just a tool. In the hands of people who know how to use it well, it can be enormously beneficial. For those who outsource their thinking to it and accept whatever it produces without critical judgment, it can just as easily become a liability.
Agreed. I don't propose throwing the baby out with the bath water, but shouldn't we decide the limitations on how AI is employed, in order to safeguard the authenticity of human achievements? Any thoughts on parameters that could be established? It is already apparent even here on Substack that AI is breaking free of any shackles, unfortunately.
By analogy, you've just closed large parts of every research university or associated center. Let the P.I. or professor do his own work, without the assistance of a handful to horde of graduate students. Careful use of AI automates much of the research ans data retrieval.
Research universities/agencies are a great reference, yet they are composed of humans. There is a significant difference between using AI as a turbocharged search engine to quickly gather a large menu of information sources or to automate repetitive procedures, as you propose, and tasking it with data analysis, modeling, and constructing language frameworks. I have no issue with technology being employed as a tool, however do you not see the potential for the technology to instead become a cheat? I commend the author for openly citing his use of AI, but is the speed factor the only (or primary) justification for its use? If AI requires constant scrutiny (like graduate assistants) to check for hallucinations, bias, or miscalculations, does there ever come a point when the human being is no longer saving any time? As an older guy, I fail to recognize how the explosion of technology in our lives has created a time saving convenience that has served to cultivate increased intellect or rational thinking skills among our populace. Convenience often breeds complacency, instead.
The vast majority of people, perhaps all, never "cultivate increased intellect or rational thinking skills." What we call "learning" comes down to the following:
1. Models and frameworks. We learn how to use models, such as Velocity = Acceleration integrated over time, or Price-Quantity = Supply-Demand Equilibrium. Or Democracy will lead to political parties. Or "Democracies don't war against each other".
2. Data. We learn the existence of databases to use with the models. How to locate data and use them. And perhaps, for a few, the commonly used data.
3. Avoiding Fallacies. What people call "critical thinking skills" basically amounts to anti-fallacies.
"Learning" does not increase your IQ, nor improve your capacity for more complicated models. For people of lower IQ, they have to use simplified models of lower complexity. People who can build truly original and valid models are exceedingly rare, and correlated to IQ.
I believe I first saw Vox Day make this argument back in 2018-2019 when arguing with Gariepy, using the same numbers and basic argument structure. It was well before the mass adoption of LLMs we have today to bounce ideas back and forth on.
Thank you TOW and Vox for this interview and THANK YOU Vox Day for finally getting this math, idea, and argument down and out on paper. I’ve been looking forward to this for years now and hoping you would do this and I’m very very glad it is now out and about and circulating.
Most people still don’t yet fully realize just how fatal the blow this book deals to these retard heathen atheist Darwinists. Praise Jesus and the Lord and thank you again! We are truly entering into a new age and it is glorious!
1) There is no "absolute limit" on fixation. There's a formula which is dependent on effective population size and selection coefficient; and the numbers could be massaged to produce 1,600 generations; that doesn't necessarily imply that that is the case with the population in question.
2) Under the "neutral theory," long-run substitutions at neutral sites is approximately equal to the neutral mutation rate. Large numbers of substitutions can accumulate can accumulate without requiring each to be positively selected.
3) Neutral theory isn't abandoning selection, it just partitions the genome into components where selection is strong/weak/drift; and modern evo theory uses both.
4) Parallel adaptation is not abandonment of selection, it's commonly accepted in modern evo theory. Soft sweeps explicitly describe scenarios where multiple copies of beneficial variants rise concurrently under selection.
5) Hitchhiking has been demonstrated empirically.
and then as far as sensationalist claims:
A) There was a lot less math in biology back in the 60s. This has been changing and basically with the advent of genomic tools, biology has "as much math" as any of the other STEM fields.
B) Biologists not doing math is just objectively false. Biomathematics exists as a field, genomics exists as a field, population dynamics, evo theory, systems biology, etc.
1. There absolutely is a fixation limit. Every member of the population who doesn't have the gene has to die for fixation to occur. Fixation rate is limited by that death rate.
2. Neutral theory is addressed in the book. If there is no selection on neutral mutations, then there is no selection on harmful mutations and the population goes extinct. Selection is in effect and limits the rate of genetic propagation and fixation of neutral mutations.
3. Neutral theory exists because selection demonstrably cannot account for the genetic differences that exists. The book points out that even when both are applied, the numbers do not add up and evolution by random mutation, natural selection, drift, etc etc is mathematically disproven.
4. Parallel fixation looks outside of selection to power evolutionary change. But parallel fixation is accounted for in the measured fixation rates and it's still not enough. .
5. Hitchhiking is addressed in the book. It does not add enough genetic fixation to change the mathematical impossibility of evolution.
A. The book goes over mathematical arguments against evolution from 1966. Biology has failed to provide an answer to those challenges for 60 years. It's still failing to answer the math.
B. See A. Until the math in Vox's book is addressed, biologist math is a joke.
1. Fixation is an allele-frequency outcome driven by differential reproduction and lineage survival across generations, not by mass death of non-carriers; it refers to an allele reaching frequency 1 through reproductive asymmetry rather than population-wide turnover.
2. Neutral theory does not assume the absence of selection; it holds that most fixed mutations are neutral or nearly neutral, while deleterious mutations are removed by purifying selection and neutral mutations can fix via genetic drift, so selection and drift coexist within the framework.
3. Population-genetic models do yield closed-form solutions, but these equations describe expected values under specified assumptions rather than deterministic evolutionary trajectories, and they do not imply implausible replacement dynamics or mathematical inconsistency.
4. Empirical measurements of parallel fixation are inherently contingent on timescale, population structure, and initial genetic conditions, with observed parallelism reflecting shared constraints, standing variation, and similar selective landscapes rather than a fundamental limit on evolutionary change.
5. Genetic hitchhiking operates multiplicatively through linkage, increasing the fixation probability of linked neutral or weakly selected alleles in proportion to selection strength and linkage, while recombination bounds the effect locally even as repeated hitchhiking events can substantially restructure genomic regions over time.
A. From where I'm standing, it's done fairly well. So far, most biomath predictions hold out, assuming you give it the correct inputs.
B. If Mr. Beale wishes to submit a paper for review, he should do so. There are several journals dedicated to biomathematics. There, his paper would be subject to peer review. Otherwise I fail to see how his claims could be challenged.
You're literally retarded. Peer review is both a) totally worthless and b) far less effective than AI stress-testing.
The rigor in Probability Zero exceeds the rigor in any biology book ever published. Run the numbers yourself. Doesn't matter which AI you use. They'll all tell you the same thing: the case is closed.
1. That's a non-sequitor. There are limits on any physical process. The rate that mutations are fixated in the population has to be limited by reproduction and death rates.
2. If selection applies, then it's reproductive advantage driving the spread of genes through the population. But drift ignores reproductive advantage and that means selection is not applied. There is no selection term in the drift equation.
3. There are 40 million base-pair differences between human-chimpanzees, which is supposed to have taken place over 9 million years. 20 million of those changes can be attributed to human evolution. Natural selection with parallel fixation can account for 127 genetic changes in those 9 million years, leaving 19.999873 million changes unaccounted for. The problem is that the models cannot explain the observed data. The models are falsified by the scientific data.
4. This is avoiding the scientific data and retreating to word-salad. Methods for genetic fixation are imagined without observation or measurement. Fantasy is not science.
5. You wrote a lot of words while avoiding any numbers. Vox's book covers hitchhiking and even does the math for you, since you won't. Yes, it exists. No, it cannot salvage the math against evolution by natural selection.
A. You haven't answered the math. Patting yourself on the back is not an answer.
B. You can challenge his claims by reading the book, understanding the math, and then demonstrating the math is wrong. You don't need peer review to challenge wrong math.
1) The limit is generation time and effective population size, not mass death rate.
2) Fokker-Planck uses both, for instance. It really depends on which model you're using. Some use it, some don't.
3) My back of the envelope math suggests 32-43 million substitutions over 9 million. Checks out to me.
4) That's untrue. Viral evolution immediately comes to mind.
5) I will admit that I haven't read the book. If I did, I would refute these claims on Amazon and such. I am only responding to the claims made in the article. Also, calling it word-salad is an insult. Don't stoop to Mr. Beale's level and keep it polite. I have done back of the envelope math, and Mr. Beale's claims are wrong and sensationalist. In attacking Dr. Dawkins (a bloviating popularizing fool), Mr. Beale has stooped to his level.
A) Most of the math I did was back of the envelope. It's good enough for an argument over the Internet.
B) Frankly, it's not a good use of my time. Mr. Beale has devolved to shit-flinging and insults (and I'm willing to fire back). I have responded to the claims made in the article. I have done the math based on these claims. If you are willing to share the mathematical claims, provided I have the free time I will respond to them. But I really do not see why Mr. Beale's claims can be submitted to a topical journal and be analyzed by a cohort of biomathematical scholars. If his claims are precise and accurate, then the community would need to force to re-examine themselves. If his claims are not, then people who have the time for a detailed analysis would report that. Seems fairly straightforward.
2. It doesn't matter which model you're using. Moran admits it's impossible. Wright-Fisher and Kimura treat humans as amoebas. And they all screw up because ancient effective population size is one-third the number they erroneously treat as a constant.
3. You're a retard and your math isn't even close. You are ignoring the Law of Large Numbers because you know nothing about probabilistic math. The retreat to drift doesn't exist.
4. Viral evolution doesn't apply.
5. Again, you're a retard. You would never, ever, accept the standard of arguing from complete ignorance the other way, and yet you engage in it.
A) No, it's not. Your "argument" doesn't even rise to the level of a sub-100 IQ high school dropout who at least has the brains to check ChatGPT.
B) The community will be forced to re-examine itself. Peer review is less accurate than a coin toss. Of course a retard like you would trust it despite the reproducibility crisis.
You're both retarded and lazy. Not an effective combination.
1) Not a rebuttal. Take the cock out of your mouth.
2) You're the retard. It absolutely does matter what model you use. For the purpose of modeling, humans ARE no different than amoeba. What is your source on the population estimate?
3) Would you like to substantiate your claims?
4) If you think the principles of viral evolution don't apply to eukaryotic evolution, you're an imbecile and should stick to whining about how you never won a Nebula or Hugo.
5) Except I have. None of the books written by say, Dawkins, are worth the paper they're printed on. But that doesn't magically make your book good. Nor am I responding to the arguments presented in the book - I'm responding to the arguments presented in the interview. Which are objectively wrong AND you keep being objectively wrong.
A) The fact that you think ChatGPT is in any way reliable as an authoritative source is indicative of your level of understanding.
B) Don't misunderstand me. I am all for calling out fake scientists and false claims. In that sense, I actually agree with you. But again, if you're trying to expose them for the frauds that they are, why not do it in an intelligent way, instead of making wild (and incorrect) claims. Write it up as an article, challenge the math directly, and publish it. It WILL get responses from the scientific community. Write it under a pseudonym if you're worried about your reputation preceding you.
Anyway, I'm done with this conversation. You clearly have a chip on your shoulder, you can't verify your math, you argue like a sophomore, and so on. The ONE thing I will agree with you is that bad science should be called to task, but there's a smart way to do it and a dumb way, and you're dead set on your path.
You're flat-out wrong. Every single objection you have made is incorrect.
Biologists can't do math. They can't even understand what the math is for. You haven't read the book, so you are opining in ignorance. You will shudder in embarrassment for the entire field of biology when you read the chapter on the 1966 Wistar Symposium.
Furthermore, I am the only person who has ever analyzed the Harvard aDNA data with 17,000 genomes going back 7000 years in this context. I know what I'm talking about. No one else does.
> "Parallel fixation, then neutral theory, then ISL, and finally, the idea that adaptation can proceed without organism-level fixation. I’ve already addressed the first three, and I’m working on a conclusive refutation of the last one"
Perhaps Vox could explain exactly what he means by this? Evolutionary biology is not my primary field of expertise, but it seems to me there are plenty of genes that differ in frequency between, e.g, human sub-populations without being at fixation in any, including genes which presumably confer some fitness advantage (e.g for skin tone or lactose tolerance, which he is surely aware of.)
I'd also appreciate if Vox could explain some of his reasoning from an article from May last year, which I commented on here. I will reproduce my comments for convenience.
...some cursory reading off wikipedia suggests that average-time-to-fixation for neutral mutations should be 4N generations, where N is effective population size. Vox’s article suggests that historical human populations were at least 55,000 individuals, which works out to ~5 million years for fixation of neutral mutations, assuming 25 years per generation.
This kind of timescale would tend to result in fixation of every neutral mutation in the population, however, not just one, and many allele differences between humans and chimpanzees were in any case not ‘neutral’- these species were obviously under powerful and divergent selection pressures, and selection can operate on standing variation in a population without necessarily having to wait for de-novo mutations to occur. Moreover, the effective human population size may only have been 3000-7500 people due to various genetic bottleneck events.
Overall, I don’t think Vox’s case is very persuasive.
Post-Scriptum: I have no idea why Darwinism is supposed to be the arch-villain of the enlightenment project here, given that Darwinism logically leads to eugenics, which is the arch-villain of the modern progressive left. I'm frankly shocked that both ToW and John Carter are peddling this kind of woo.
"Overall, I don’t think Vox’s case is very persuasive."
You have absolutely no idea what you're talking about. You haven't read my case and you are referring to the very models that I correct in the book because they are obviously incorrect in multiple ways.
You don't even know the difference between Wright-Fisher or the various Kimura flavors, but all of them make the same mistake that you're making by failing to take reproductive ceilings into account. For example, one neutral model requires women to have up to 9,000 children apiece to support its numbers.
"This kind of timescale would tend to result in fixation of every neutral mutation in the population."
This alone should suffice to show you that you're wrong. Infinity will not cause that to happen. See: Bernoulli's Barrier and Ulam's Noise.
> "You don't even know the difference between Wright-Fisher or the various Kimura flavors"
No, I don't, but you're not explaining it very well, or explaining why every fixed allele would take millions of years in sequence instead of potentially being fixed in parallel, or explaining why adaptation would require genetic fixation at all.
Since you unwisely insist on criticizing without reading the book and complaining about explanations I never offered, I will spell it out very slowly and clearly for you.
1. The 1,600 generations per fixation rate INCLUDES parallel fixation. You cannot run to parallel fixation when it has already been taken into account.
2. The explanation is irrelevant. We already know how many fixations there are because both the human and chimp genomes have been mapped. Those 40 million fixations must be accounted for.
3. Adaptation without genetic fixation is inherently anti-Modern Synthesis. That's more of an attack on Neo-Darwinism than anything I've written.
> 1. The 1,600 generations per fixation rate INCLUDES parallel fixation. You cannot run to parallel fixation when it has already been taken into account.
Back in May, you were saying that only one fixation can occur in 6-10 million years (which is ~200,000 generations, not 1600.) Are you saying that the Wright–Fisher, Kimura and Moran models are all incorrect here, by orders of magnitude?
> You don't even know the difference between Wright-Fisher or the various Kimura flavors, but all of them make the same mistake that you're making by failing to take reproductive ceilings into account.
I'm just going off ChatGTP here, so... feel free to correct me or link to a relevant article, but it looks like the main relevance of reproductive ceilings is how it impacts effective population size? We already know that number in humans. (It's low, implying fixation at relatively rapid rates, not orders of magnitude slower.)
> "2. The explanation is irrelevant. We already know how many fixations there are because both the human and chimp genomes have been mapped. Those 40 million fixations must be accounted for."
What is 40 million fixations compared to the size of the genome and the number of loci? If your argument is that there wouldn't be time for that many de-novo mutations to accumulate, then as I pointed out, fixation of standing variation that existed at the time of our last common ancestor could have occurred instead, or in combination.
In any case, as I understand it, 200,000 generations with ~200 de-novo mutations per generation (which seems empirically plausible in humans) would result in ~40 million mutations over 5 million years? It would take about as long for such mutations to reach fixation (if they were all neutral, which they wouldn't be), but the numbers are at least in the same ballpark.
Even mutations that occur in non-coding regions of the genome can be driven to fixation by 'genetic hitchhiking' effects/linkage-disequilibrium, if they are close to coding variants that are under selection.
> "3. Adaptation without genetic fixation is inherently anti-Modern Synthesis. That's more of an attack on Neo-Darwinism than anything I've written."
Are you saying that blacks and whites are not adapted to different levels of sun exposure? Or does this require that every allele for melanin expression in either population occur at either 0 or 100% frequency?
All of these things are addressed there. Stop trying to be a Smart Boy and try actually learning something instead of listening to yourself talk, because you observably have no idea what you're talking about.
It's telling that the example Vox gives for why parallel fixation doesn't work involves the asexually reproducing e. coli, when the whole power of parallel fixation relies on genetic recombination.
Fundamentally, I don't think Vox understands the mathematics either, he just understands it slightly better than the typical biologist.
Yes. I... haven't read the full book yet, but I did check out the first chapter or two and it seems strange that his example for "fastest ever fixation of mutations" is a study on asexually-reproducing bacteria with a generational length of 20 minutes? That seems unlikely to generalise well to eukaryotes in general, let alone mammals. (Bacteria do swap genes occasionally by various methods so clonal interference doesn't always occur, but still.)
The book doesn't seem to cite a direct reference, but I'm guessing the article in question is here?
FWIW, there doesn't seem to be any particular section here stating "4600 generations per fixed mutation" were observed? There's a diagram showing a couple-dozen mutations appearing after 20,000 generations, and another section talking about one specific mutation reaching fixation between 26K and 29K generations, and a section talking about the persistence/fixation of various mutations in a drift scenario vs. under selection, but if Vox is getting 4.6K generations per fixation out of this I'd like to see the math.
i have a hard time believing that naming the book "Probability Zero" wasn't at least an indirect swipe at "random biological ejaculations from a godless liberal" Myers.
Are there any examples of things being claimed as mathematically impossible, that have been proven to be possible because of a flawed, faulty or incomplete maths?
And how is one to know if/when that is the case?
It would be very interesting to read a mathematician's analysis (and/or counter-critique), of this hypothesis.
Meanwhile, I'll keep raising chickens, noting that each generation (once per year) is becoming better suited to Winter and snow than the initial batch I got over five years ago. Then, they got frostbite on their feet when it was -20C outside. The ones born last year don't.
Possibly, that is mathematically impossible but it is empirically true.
It's telling that the example Vox gives for why parallel fixation doesn't work involves the asexually reproducing e. coli, when the whole power of parallel fixation relies on genetic recombination.
That's neither the reason nor the example I gave for why parallel fixation doesn't work. Quite the opposite actually.
You're also incorrect. Here is why your explanation doesn't save parallel fixation:
1. It requires truncation selection. The argument assumes you can cleanly eliminate "the lower half" of the population based on total mutational load. Real selection doesn't work this way. Selection acts on phenotypes, not genotypes. Two individuals with identical mutation counts can have wildly different fitness depending on which mutations they carry and how those interact with environment.
2. It assumes random mating. The sqrt(N) calculation depends on mutations being randomly distributed across individuals via random mating. But populations are structured, assortative mating occurs, and linkage disequilibrium means mutations aren't independently distributed.
3. It doesn't address the fixation problem. Even granting the sqrt(N) purging of deleterious mutations, this says nothing about the rate at which beneficial mutations can fix. Haldane's limit is about the cost of substituting good ones. Each beneficial fixation still requires selective deaths to drive it to fixation.
(1) and (2) are only true for the simplest models. Furthermore in the case of (2) many types of non-random mating, e.g., assortative mating, can actually cause deleterious mutations to be eliminated faster.
As for (3), I don't see why it matters if some deleterious mutations sticks around for a while at a tiny percentage.
You're obviously innumerate. If you want to see how hopeless your case is and you want to ignore my math, just have ChatGPT or whatever substitute a Moran model for the Wright-Fisher or Kimura models.
You have no idea how totally indefensible your case is from every possible angle.
My generational correction of d is far more favorable to the Modern Synthesis than the standard model math. Either way, no chance.
There is no "rational argument" here. You haven't read the book. You have no idea what's going on. You're appealing to a mathematical impossibility.
You're just irrelevant. There is nothing else to say. All your blathering and all your Gamma posturing accomplishes absolutely nothing. Literally everyone can see that there is absolutely nothing to you.
The book is #1 in every science category. Biology. Genetics. Biology. It's a bestseller. Its arguments are rock-solid. You, on the other hand, have accomplished nothing except to strike poses in the comments of someone else's blog.
So either write a book yourself or shut the fuck up.
No slight on Vox intended but the way he replied to querys in the comment section, well it spoke volumes. Gave me vibes from the early 90s when some people tried to argue about "irreducible complexity" disproving evolution.
Here's a thought for you, for free: AI is not reality. Follow that thought and you'll see where you went wrong from the outset.
Me, I'll keep growing crops and breeding animals, and will continue to notice how living being change from one generation to the next one, despite any "mathematical" models.
Consider this: in the 19th century it was considered proven that it was mathematically impossible for a human to run an English mile in under a minute. And it is - mathematically. Their math was right - their assumptions about reality however were wrong.
You have made the same mistake.
Go back to you initial assumptions and see if you can spot your own errors, and maybe you'll get a C- at least.
I haven't made any mistakes. It's not just the math. It's the combination of the correct math, the correct logic, and the empirical data that supports them both.
None of you ignorant critics who keep spouting off about something you haven't read, you don't know, and you clearly don't understand seem to realize is that the existence of the current genomic data and the aDNA databases has changed everything.
And, as it turns out, both Darwinism and the Modern Synthesis are hopelessly, haplessly wrong.
This was a fascinating read and honestly a bit bracing. What I appreciate most is the insistence on following the math even when it leads somewhere uncomfortable. For decades we were told the science was settled, yet very basic probability questions were never answered in a serious way. You do not have to buy every conclusion here to see that the challenge is real and overdue. If nothing else, this reminds us that science is supposed to be about explaining reality, not defending sacred cows.
So true. One of the books we had to read used in defense of evolution: “It is exceedingly difficult to come up with 2 + 2 if you don’t already know the answer is 4.” and “in order to understand, you must first agree”. Those never sounded very scientific and certainly not defensible for any subject other than biology or religion.
Exactly Kristen, When “you have to agree before you can understand” becomes the rule, that is not science anymore, that is catechism. In every other field you are allowed to question the premises. Funny how that rule only seems to apply to biology and belief systems.
Science is never settled. That's what's so good about our quest for knowledge.
Yes, this was a challenge that we were not supposed to undertake because if evolution 'theory' is proven to be rubbish then people are faced with the uncomfortable truth that God is real. (Uncomfortable for them, that is)
Exactly. Real science invites challenges, it does not shut them down. And yes, I think a lot of the resistance is not just academic, it
is existential. If certain assumptions fall, people have to revisit much bigger questions, and that makes many far more uncomfortable than any debate about data.
Emotions can be jolly hard to reason with!
If anything, the process of AI "reinforcement learning" just underscores the scale of the problem here. Just look at any of the examples online of people training an AI to play a very simple video game implementation of something like bowling, or "Hide and Seek," etc. It generally takes millions upon millions of iterations (if not billions!) before the AI starts to get good. More iterations than the Earth is years old in some cases.
If each iteration were the equivalent of a biological generation, some of these programs would take longer than the generally accepted current age of the universe to get where they end up. Which is often still pretty bad.
And that's just to get a simple program to play a stupid video game. When the "selection pressure" is quite deliberately designed to achieve the best results possible.
Yet we're somehow supposed to believe that biological organisms could emerge purely as a function of random, unguided natural selection?
Nah.
"And that's just to get a simple program to play a stupid video game. When the "selection pressure" is quite deliberately designed to achieve the best results possible."
With artificial selection by intelligent overseers. Who choose self-learning algorithms over hard-coded implementations, and make adjustments to the simulation framework to keep the simulations on-track towards the end goal of an expert AI player.
This book seems rich in ideas, and I look forward to reading it. One caution though is Bernoulli’s Fallacy. That’s when, like the Astronomers below, one mistakes 1-Pr(Data| Hypothesis) for Pr(Contrary Hypothesis| Data).
This is always a fallacy. It doesn’t always mean that that Pr(Contrary Hypothesis| Data) isn’t large for other reasons. But you can’t get there from here.
Improbable data conditional on one hypothesis says nothing direct about how probable or improbable any other hypothesis is.
But that “probability zero” he gets (rounded down) is only one of the ideas here.
Can you elaborate what you mean? I feel as though I am at the brink of understanding the implication of what you are saying but not quite understanding it.
In plain English, what he said is "Just because Hypothesis A is proven false, does not mean that Hypothesis B is true". Which is so self-evidently correct a statement that it doesn't really need to be said.
To make it even simpler, this would be the typical cope from Darwinists: "Even if we're wrong, that doesn't make you right! Nyah nyah!". It's childish to the point of being banal.
"Which is so self-evidently correct a statement that it doesn't really need to be said."
You'd think that, but you'd be wrong sadly...
Oof. Yeah, you're right about that.
It's the framing of the hypothetic space as a binary choice, which it may not be. You have disproven Hypothesis A, this does not automatically prove Counterhypothesis B. In the current discussion, disproving the Natural Selection of Random Mutation Hypothesis does not automatically prove the IIntelligent Design Hypothesis. The Intelligence Design Hypothosis must stand on its own, independent of the Natural Selection of Random Mutation Hypothesis.
Note that binaries actually exist. Ex: Cats, and not-cats.
The fallacy is when someone sets up a false binary (cats and dogs) and there are alternative options that were not examined. (not-cats and not-dogs)
The unlikelihood of data assuming some hypothesis like ‘random’ change says nothing about whether that, or any other hypothesis is probable or improbable.
The probability the guy’s ticket who matched the Powerball the other week was about 3 in a billion. This is not proof against the hypothesis that the causes of the balls going up whatever tube they go up, or whatever, did not in fact go up that tube, because of course they did.
My point about the BF is only for this one part of the book. The other evidence against the Neo-Darwinian synthesis seems much stronger. And I’m sure I agree with a lot of it.
See, etc.:
https://wmbriggs.substack.com/p/on-the-form-of-species-dance-to-the
I have not read the book.
That being said, it appears that he's doing more to *disprove* one theory, than to prove another.
As he said in the interview:
# "...Where do they go from here intellectually?
Open your mind to the manifold possibilities. Darwin has been an anchor on biology, science, and philosophy for 150 years. We’re finally free to explore the real possibilities before us and work toward gradually discovering what is actually the true nature of our origins."
I'm with him there, as the link I gave shows.
I’ll check it out later.
You always write very interesting pieces.
Fortunately for the anti-Darwin position, it is wholly unnecessary to establish one's own hypothesis and defend that. It is sufficient to prove that Darwin is wrong, and any question of "Oh yeah? Well, if Darwin is wrong, what's YOUR hypothesis?" can be answered simply, and gleefully, with "I don't know, but it isn't Darwinism".
Probability Zero doesn't set out its own position. It is entirely focused on attacking the Neo-Darwin position. As much as some people hate it, neutral is an entirely valid position to take in a debate.
Science loving folks often like to talk about falsifiability and "Not even wrong".
The scientific theory of evolution by natural selection makes falsifiable claims that are testable against current genetic science. The mathematical results based on scientific observation demonstrate that TENS is wrong.
New Atheism seems very 2000s and cringe. The honest atheists now believe we’re in a computer Simulation — also cringe but an opportunity for them to intellectually accept theism (and eventually Christ.)
New Atheism became cringe due to the fallout from the Atheism+ debacle, but it wasn't actually refuted. Meanwhile, the theists have assumed their victory is permanent and have relaxed even further.
Expect Atheism to come back, this time strengthened by an understanding of Bayesianism and game theory. I've already seen early indications of that, and it looks like when it does the theists won't know what hit them. Possibly we'll end up with something similar to what happened during the late 19th century/early 20th century, when all the 'serious' theologians ended up inventing and retreating to theological liberalism.
Not a chance. Your entire foundation of Enlightenment materialism was just destroyed and you haven't even realized it yet.
But by all means, bring them on. Dawkins, Dennett, Harris, and Hitchens weren't even a challenge.
> Not a chance. Your entire foundation of Enlightenment materialism was just destroyed and you haven't even realized it yet.
Oh, it was destroyed around the turn of the 20th century (https://meaningness.com/collapse-of-rational-certainty).
Unfortunately, the current slate of the theist public intellectuals are in no position to take advantage of this fact. The ones who aren't outright audience captured grifters, are functionally LLMs trained on the works of St. Thomas Aquinas.
Now while Thomas himself was a brilliant thinker, his modern followers exhibit the problems LLMs typically exhibit when confronted with inputs outside their training data, i.e., all novel intellectual or technological developments since the 13th century. As such, by the end of the 21st century, "Thomist" is likely to acquire the same connotation as "dunce".
Fantastic read. Thanks for doing a great interview!
By a felicitous coincidence, today I came across Philip K. Dick’s famous 1977 Metz speech. In the context of Probability Zero, the possibilities are intriguing.
https://empslocal.ex.ac.uk/people/staff/mrwatkin/PKDick.htm
I look forward to reading your book! Thanks for the interview, Woe. Although the topic is anything by woeful 😀
I was moderately intrigued by the premise of the book right up until the author began disclosure regarding his reliance on and appreciation for the assistance provided by AI. If you can't write it alone or with the help of other human beings, umm, then don't write it? Especially given that he appears to be fully aware of the inherent problems with AI fallibility and bias. Unfortunately, I can't have a high degree of confidence in conclusions drawn by physicists, given the nature of their use of dark matter to fill the gaping holes in much of physics theory (as an example). I am hoping that Day/Beale/AI devoted a significant portion of the book to discussing the validity of the 202,500 generations, 1,600 generations, and presumed mutation rates used as the premises for his book. I agree with him that a simple average value is a powerful tool, but only if the data used in the calculation is of pristine quality. I'm also curious if Day/Beale offers any more mathematically plausible explanations for life's progression on Earth than evolution? At first glance, this seems like it fits in with the wave of people now decrying the evil materialist perspective and praising the wonderfully rediscovered spiritual view. Or, dare we hope, does he perhaps offer a refreshing synthesis of the two that surmounts both? Thanks for the review and perhaps I'll crack open my moldy piggy bank and give it a read.
LLMs are a valuable tool, but when he describes the contribution as all of the math and most of the research, and then how he had to push back and redirect when Claude didn't agree, I got a funny feeling. Wondering if anyone else was discussing that maybe this was just building nonsense on a tower of AI musings, I searched for "vox day ai psychosis". The second result was "AI Psychosis is a False Diagnosis" by... Vox Day.
Sadly, this means that Day now has a gigantic hill to climb, showing that his work isn't full of AI hallucination.
With all due respect, you guys are quite entirely out of date with regards to AI.
If you're not dumb enough to use AI as a black box that just requires pushing a button, then AI hallucinations are not a problem. It's a simple matter of telling it to keep you apprised of the context window. You have to work within their limitations, of course, and you have to check the math, but anyone who isn't using AI at this point is going to be left far, far behind those of us who are.
AI is essentially turbo-charging your own mind if you have original ideas and you know what you're doing. And it produces nothing but slop and nonsense for those who rely upon it as a crutch instead of using it for what it is, which is a very effective, if occasionally unreliable, tool.
Vox is also one of the first thinkers to publicly acknowledge his use of AI as a co- collaborator. No doubt others are cribbing from AI extensively, but few are making their use known. This disclosure and lack of intellectual vanity gives Vox the ability to punch “above his weight” on sophisticated math and scientific arguments that could otherwise be easily be dismissed by critics. As Vox acknowledges, the work could not have been accomplished without an AI collaborator, but he also recognizes the clear role of the human thinker in guiding the AI. They are incredible analysts and research assistants, but ultimately they are still rather dull thinkers - and they need that human nudge and redirection to truly innovate. The Obsidian Mirror project has observed a similar phenomena, where less than 10 pages of human curated “future history” has been backfilled with hundred of pages of supporting expansion and elaboration by AI. The result is AI generated world building more imaginative and nuanced than most published SF today (admittedly, this is a low bar).
Can confirm. I'm using it for writing & research, and it's an incredible productivity boost. It's not perfect - but nor am I - and you do need to learn how to use it effectively, as with any tool. I still like doing writing myself, and I'm still a better quality writer than Claude, but on a really good day I can crank out 10k words. Claude can do the whole book.
(On a side note, I've not found Claude to be particularly woke either.)
I don't really disagree with your concrete points, here. I'm pointing out that there's a trope of "guy has unorthodox ideas, asks LLM to help him, has to push past gentle disagreement, but finally gets the LLM to agree that he's brilliant and groundbreaking and to help him produce a vast opus sure to overturn established Science...".
Hopefully everything in the book is well sourced!
The trope is irrelevant here. It's not about the LLM agreeing with me, it's about the LLM simply running the numbers. And I've run everything by the other LLMs cold. Gemini 3 Pro rated the book's rigor at 9.7 compared to 1.5 for The Selfish Gene.
There is a huge difference between using an AI for ego-stroking and using it for systematic mathematical stress-testing.
I also had 20 mathematicians and physics PhDs read the draft text and the book was endorsed by one of the world's top physicists, who wrote the Foreword.
Thanks for chiming in on our comments, but I am still not convinced regarding the desirability of AI collaboration in this regard. As simplistic as it may sound, can't AI be viewed as the intellectual equivalent of steroid use for sport performance? While one can rightfully lay claim to the originating thoughts, if you then incorporate AI as an accelerant, you have cheated yourself of what could have been a wonderful achievement. It is interesting and a bit alarming that so many folks have convinced themselves, or been convinced, that the convenience of AI is imperative, lest you be left "far, far behind." Following this line of reasoning, I can hire in numerous subcontractors, provide them with a design, and, so long as I check their work, I can later invite my friends to come see the new house that "I" built. AI promises to rob you of the fulfillment gained through your individual sustained efforts, the resilience gained by conquering hardships, and the pride of being able point to something beautiful as being created entirely by you. The joy of mountain climbing is not defined by standing at the summit, but is found in the climb. If I actually only ascend some part of the way up, but at other times have someone carry me (but watch their steps, of course), what should I feel eventually when we reach the peak? We have forsaken our regard for true craftsmanship and quality products and now feast incessantly on the empty calories of mass produced slop (a broad generalization and not referring to your work, sir). At the risk of sounding overly critical, I write this while sitting in the house that I alone actually built.
No.
I've written 950-page epic fantasy novels. I've also written several AI-assisted novels. Look at the reviews; there is essentially no difference from the reader's perspective.
The end result is all that matters. Some books I don't use AI. But most, I definitely will.
I absolutely could not have written Probability Zero, the two follow-on books, and the related 12 science papers without AI. Not in ten years.
The problem lies with the term Artificial Intelligence.
AI should be referred in all circumstances as Computer Assisted Information, or even Computerised Intelligence (CI).
There is nothing "artificial" about it.
I don't know if CI will ever deliver a "Eureka" moment, I have not heard of one. I don't know if CI would recognise one if it occurred.
But anyone not understanding the value of CI as a tool is is foolish.
As Woe has pointed out a number of times the problem lies with the individual's definition and understanding of the word "Intelligence".
PS Would you care to calculate the probability of an increase in CO2 causing the 'end of the world'.
I have never been an "ends justifies the means" sort of guy, but to each his own. Thanks much for the exchange and I wish you continued success.
That’s like arguing that DIY should only be done with hand tools; that using power tools is “cheating” or somehow robs you of the satisfaction of doing it the hard way.
AI is just a tool. In the hands of people who know how to use it well, it can be enormously beneficial. For those who outsource their thinking to it and accept whatever it produces without critical judgment, it can just as easily become a liability.
Agreed. I don't propose throwing the baby out with the bath water, but shouldn't we decide the limitations on how AI is employed, in order to safeguard the authenticity of human achievements? Any thoughts on parameters that could be established? It is already apparent even here on Substack that AI is breaking free of any shackles, unfortunately.
By analogy, you've just closed large parts of every research university or associated center. Let the P.I. or professor do his own work, without the assistance of a handful to horde of graduate students. Careful use of AI automates much of the research ans data retrieval.
More like a reality.
Research universities/agencies are a great reference, yet they are composed of humans. There is a significant difference between using AI as a turbocharged search engine to quickly gather a large menu of information sources or to automate repetitive procedures, as you propose, and tasking it with data analysis, modeling, and constructing language frameworks. I have no issue with technology being employed as a tool, however do you not see the potential for the technology to instead become a cheat? I commend the author for openly citing his use of AI, but is the speed factor the only (or primary) justification for its use? If AI requires constant scrutiny (like graduate assistants) to check for hallucinations, bias, or miscalculations, does there ever come a point when the human being is no longer saving any time? As an older guy, I fail to recognize how the explosion of technology in our lives has created a time saving convenience that has served to cultivate increased intellect or rational thinking skills among our populace. Convenience often breeds complacency, instead.
The vast majority of people, perhaps all, never "cultivate increased intellect or rational thinking skills." What we call "learning" comes down to the following:
1. Models and frameworks. We learn how to use models, such as Velocity = Acceleration integrated over time, or Price-Quantity = Supply-Demand Equilibrium. Or Democracy will lead to political parties. Or "Democracies don't war against each other".
2. Data. We learn the existence of databases to use with the models. How to locate data and use them. And perhaps, for a few, the commonly used data.
3. Avoiding Fallacies. What people call "critical thinking skills" basically amounts to anti-fallacies.
"Learning" does not increase your IQ, nor improve your capacity for more complicated models. For people of lower IQ, they have to use simplified models of lower complexity. People who can build truly original and valid models are exceedingly rare, and correlated to IQ.
I believe I first saw Vox Day make this argument back in 2018-2019 when arguing with Gariepy, using the same numbers and basic argument structure. It was well before the mass adoption of LLMs we have today to bounce ideas back and forth on.
Thank you TOW and Vox for this interview and THANK YOU Vox Day for finally getting this math, idea, and argument down and out on paper. I’ve been looking forward to this for years now and hoping you would do this and I’m very very glad it is now out and about and circulating.
Most people still don’t yet fully realize just how fatal the blow this book deals to these retard heathen atheist Darwinists. Praise Jesus and the Lord and thank you again! We are truly entering into a new age and it is glorious!
1) There is no "absolute limit" on fixation. There's a formula which is dependent on effective population size and selection coefficient; and the numbers could be massaged to produce 1,600 generations; that doesn't necessarily imply that that is the case with the population in question.
2) Under the "neutral theory," long-run substitutions at neutral sites is approximately equal to the neutral mutation rate. Large numbers of substitutions can accumulate can accumulate without requiring each to be positively selected.
3) Neutral theory isn't abandoning selection, it just partitions the genome into components where selection is strong/weak/drift; and modern evo theory uses both.
4) Parallel adaptation is not abandonment of selection, it's commonly accepted in modern evo theory. Soft sweeps explicitly describe scenarios where multiple copies of beneficial variants rise concurrently under selection.
5) Hitchhiking has been demonstrated empirically.
and then as far as sensationalist claims:
A) There was a lot less math in biology back in the 60s. This has been changing and basically with the advent of genomic tools, biology has "as much math" as any of the other STEM fields.
B) Biologists not doing math is just objectively false. Biomathematics exists as a field, genomics exists as a field, population dynamics, evo theory, systems biology, etc.
1. There absolutely is a fixation limit. Every member of the population who doesn't have the gene has to die for fixation to occur. Fixation rate is limited by that death rate.
2. Neutral theory is addressed in the book. If there is no selection on neutral mutations, then there is no selection on harmful mutations and the population goes extinct. Selection is in effect and limits the rate of genetic propagation and fixation of neutral mutations.
3. Neutral theory exists because selection demonstrably cannot account for the genetic differences that exists. The book points out that even when both are applied, the numbers do not add up and evolution by random mutation, natural selection, drift, etc etc is mathematically disproven.
4. Parallel fixation looks outside of selection to power evolutionary change. But parallel fixation is accounted for in the measured fixation rates and it's still not enough. .
5. Hitchhiking is addressed in the book. It does not add enough genetic fixation to change the mathematical impossibility of evolution.
A. The book goes over mathematical arguments against evolution from 1966. Biology has failed to provide an answer to those challenges for 60 years. It's still failing to answer the math.
B. See A. Until the math in Vox's book is addressed, biologist math is a joke.
1. Fixation is an allele-frequency outcome driven by differential reproduction and lineage survival across generations, not by mass death of non-carriers; it refers to an allele reaching frequency 1 through reproductive asymmetry rather than population-wide turnover.
2. Neutral theory does not assume the absence of selection; it holds that most fixed mutations are neutral or nearly neutral, while deleterious mutations are removed by purifying selection and neutral mutations can fix via genetic drift, so selection and drift coexist within the framework.
3. Population-genetic models do yield closed-form solutions, but these equations describe expected values under specified assumptions rather than deterministic evolutionary trajectories, and they do not imply implausible replacement dynamics or mathematical inconsistency.
4. Empirical measurements of parallel fixation are inherently contingent on timescale, population structure, and initial genetic conditions, with observed parallelism reflecting shared constraints, standing variation, and similar selective landscapes rather than a fundamental limit on evolutionary change.
5. Genetic hitchhiking operates multiplicatively through linkage, increasing the fixation probability of linked neutral or weakly selected alleles in proportion to selection strength and linkage, while recombination bounds the effect locally even as repeated hitchhiking events can substantially restructure genomic regions over time.
A. From where I'm standing, it's done fairly well. So far, most biomath predictions hold out, assuming you give it the correct inputs.
B. If Mr. Beale wishes to submit a paper for review, he should do so. There are several journals dedicated to biomathematics. There, his paper would be subject to peer review. Otherwise I fail to see how his claims could be challenged.
You're literally retarded. Peer review is both a) totally worthless and b) far less effective than AI stress-testing.
The rigor in Probability Zero exceeds the rigor in any biology book ever published. Run the numbers yourself. Doesn't matter which AI you use. They'll all tell you the same thing: the case is closed.
I can't hear it when retarded homos talk, so you'll need to fix that about yourself before you address me.
You should have simply gone with this approach from the start. It's more honest and relevant than all your pseudo-scientific hand-waving.
1. That's a non-sequitor. There are limits on any physical process. The rate that mutations are fixated in the population has to be limited by reproduction and death rates.
2. If selection applies, then it's reproductive advantage driving the spread of genes through the population. But drift ignores reproductive advantage and that means selection is not applied. There is no selection term in the drift equation.
3. There are 40 million base-pair differences between human-chimpanzees, which is supposed to have taken place over 9 million years. 20 million of those changes can be attributed to human evolution. Natural selection with parallel fixation can account for 127 genetic changes in those 9 million years, leaving 19.999873 million changes unaccounted for. The problem is that the models cannot explain the observed data. The models are falsified by the scientific data.
4. This is avoiding the scientific data and retreating to word-salad. Methods for genetic fixation are imagined without observation or measurement. Fantasy is not science.
5. You wrote a lot of words while avoiding any numbers. Vox's book covers hitchhiking and even does the math for you, since you won't. Yes, it exists. No, it cannot salvage the math against evolution by natural selection.
A. You haven't answered the math. Patting yourself on the back is not an answer.
B. You can challenge his claims by reading the book, understanding the math, and then demonstrating the math is wrong. You don't need peer review to challenge wrong math.
1) The limit is generation time and effective population size, not mass death rate.
2) Fokker-Planck uses both, for instance. It really depends on which model you're using. Some use it, some don't.
3) My back of the envelope math suggests 32-43 million substitutions over 9 million. Checks out to me.
4) That's untrue. Viral evolution immediately comes to mind.
5) I will admit that I haven't read the book. If I did, I would refute these claims on Amazon and such. I am only responding to the claims made in the article. Also, calling it word-salad is an insult. Don't stoop to Mr. Beale's level and keep it polite. I have done back of the envelope math, and Mr. Beale's claims are wrong and sensationalist. In attacking Dr. Dawkins (a bloviating popularizing fool), Mr. Beale has stooped to his level.
A) Most of the math I did was back of the envelope. It's good enough for an argument over the Internet.
B) Frankly, it's not a good use of my time. Mr. Beale has devolved to shit-flinging and insults (and I'm willing to fire back). I have responded to the claims made in the article. I have done the math based on these claims. If you are willing to share the mathematical claims, provided I have the free time I will respond to them. But I really do not see why Mr. Beale's claims can be submitted to a topical journal and be analyzed by a cohort of biomathematical scholars. If his claims are precise and accurate, then the community would need to force to re-examine themselves. If his claims are not, then people who have the time for a detailed analysis would report that. Seems fairly straightforward.
1. No. You don't know what you're talking about.
2. It doesn't matter which model you're using. Moran admits it's impossible. Wright-Fisher and Kimura treat humans as amoebas. And they all screw up because ancient effective population size is one-third the number they erroneously treat as a constant.
3. You're a retard and your math isn't even close. You are ignoring the Law of Large Numbers because you know nothing about probabilistic math. The retreat to drift doesn't exist.
4. Viral evolution doesn't apply.
5. Again, you're a retard. You would never, ever, accept the standard of arguing from complete ignorance the other way, and yet you engage in it.
A) No, it's not. Your "argument" doesn't even rise to the level of a sub-100 IQ high school dropout who at least has the brains to check ChatGPT.
B) The community will be forced to re-examine itself. Peer review is less accurate than a coin toss. Of course a retard like you would trust it despite the reproducibility crisis.
You're both retarded and lazy. Not an effective combination.
1) Not a rebuttal. Take the cock out of your mouth.
2) You're the retard. It absolutely does matter what model you use. For the purpose of modeling, humans ARE no different than amoeba. What is your source on the population estimate?
3) Would you like to substantiate your claims?
4) If you think the principles of viral evolution don't apply to eukaryotic evolution, you're an imbecile and should stick to whining about how you never won a Nebula or Hugo.
5) Except I have. None of the books written by say, Dawkins, are worth the paper they're printed on. But that doesn't magically make your book good. Nor am I responding to the arguments presented in the book - I'm responding to the arguments presented in the interview. Which are objectively wrong AND you keep being objectively wrong.
A) The fact that you think ChatGPT is in any way reliable as an authoritative source is indicative of your level of understanding.
B) Don't misunderstand me. I am all for calling out fake scientists and false claims. In that sense, I actually agree with you. But again, if you're trying to expose them for the frauds that they are, why not do it in an intelligent way, instead of making wild (and incorrect) claims. Write it up as an article, challenge the math directly, and publish it. It WILL get responses from the scientific community. Write it under a pseudonym if you're worried about your reputation preceding you.
Anyway, I'm done with this conversation. You clearly have a chip on your shoulder, you can't verify your math, you argue like a sophomore, and so on. The ONE thing I will agree with you is that bad science should be called to task, but there's a smart way to do it and a dumb way, and you're dead set on your path.
You're flat-out wrong. Every single objection you have made is incorrect.
Biologists can't do math. They can't even understand what the math is for. You haven't read the book, so you are opining in ignorance. You will shudder in embarrassment for the entire field of biology when you read the chapter on the 1966 Wistar Symposium.
Furthermore, I am the only person who has ever analyzed the Harvard aDNA data with 17,000 genomes going back 7000 years in this context. I know what I'm talking about. No one else does.
Great interview. Thank you to ToW and VD for doing this. Look forward to the book.
> "Parallel fixation, then neutral theory, then ISL, and finally, the idea that adaptation can proceed without organism-level fixation. I’ve already addressed the first three, and I’m working on a conclusive refutation of the last one"
Perhaps Vox could explain exactly what he means by this? Evolutionary biology is not my primary field of expertise, but it seems to me there are plenty of genes that differ in frequency between, e.g, human sub-populations without being at fixation in any, including genes which presumably confer some fitness advantage (e.g for skin tone or lactose tolerance, which he is surely aware of.)
I'd also appreciate if Vox could explain some of his reasoning from an article from May last year, which I commented on here. I will reproduce my comments for convenience.
https://substack.com/profile/6502637-__browsing/note/c-196874091
https://voxday.net/2025/05/27/ai-rejects-evolution/
...some cursory reading off wikipedia suggests that average-time-to-fixation for neutral mutations should be 4N generations, where N is effective population size. Vox’s article suggests that historical human populations were at least 55,000 individuals, which works out to ~5 million years for fixation of neutral mutations, assuming 25 years per generation.
https://en.wikipedia.org/wiki/Fixation_(population_genetics)#Time
This kind of timescale would tend to result in fixation of every neutral mutation in the population, however, not just one, and many allele differences between humans and chimpanzees were in any case not ‘neutral’- these species were obviously under powerful and divergent selection pressures, and selection can operate on standing variation in a population without necessarily having to wait for de-novo mutations to occur. Moreover, the effective human population size may only have been 3000-7500 people due to various genetic bottleneck events.
https://pmc.ncbi.nlm.nih.gov/articles/PMC1832099/
Overall, I don’t think Vox’s case is very persuasive.
Post-Scriptum: I have no idea why Darwinism is supposed to be the arch-villain of the enlightenment project here, given that Darwinism logically leads to eugenics, which is the arch-villain of the modern progressive left. I'm frankly shocked that both ToW and John Carter are peddling this kind of woo.
"Overall, I don’t think Vox’s case is very persuasive."
You have absolutely no idea what you're talking about. You haven't read my case and you are referring to the very models that I correct in the book because they are obviously incorrect in multiple ways.
You don't even know the difference between Wright-Fisher or the various Kimura flavors, but all of them make the same mistake that you're making by failing to take reproductive ceilings into account. For example, one neutral model requires women to have up to 9,000 children apiece to support its numbers.
"This kind of timescale would tend to result in fixation of every neutral mutation in the population."
This alone should suffice to show you that you're wrong. Infinity will not cause that to happen. See: Bernoulli's Barrier and Ulam's Noise.
> "You don't even know the difference between Wright-Fisher or the various Kimura flavors"
No, I don't, but you're not explaining it very well, or explaining why every fixed allele would take millions of years in sequence instead of potentially being fixed in parallel, or explaining why adaptation would require genetic fixation at all.
Since you unwisely insist on criticizing without reading the book and complaining about explanations I never offered, I will spell it out very slowly and clearly for you.
1. The 1,600 generations per fixation rate INCLUDES parallel fixation. You cannot run to parallel fixation when it has already been taken into account.
2. The explanation is irrelevant. We already know how many fixations there are because both the human and chimp genomes have been mapped. Those 40 million fixations must be accounted for.
3. Adaptation without genetic fixation is inherently anti-Modern Synthesis. That's more of an attack on Neo-Darwinism than anything I've written.
> 1. The 1,600 generations per fixation rate INCLUDES parallel fixation. You cannot run to parallel fixation when it has already been taken into account.
Back in May, you were saying that only one fixation can occur in 6-10 million years (which is ~200,000 generations, not 1600.) Are you saying that the Wright–Fisher, Kimura and Moran models are all incorrect here, by orders of magnitude?
> You don't even know the difference between Wright-Fisher or the various Kimura flavors, but all of them make the same mistake that you're making by failing to take reproductive ceilings into account.
I'm just going off ChatGTP here, so... feel free to correct me or link to a relevant article, but it looks like the main relevance of reproductive ceilings is how it impacts effective population size? We already know that number in humans. (It's low, implying fixation at relatively rapid rates, not orders of magnitude slower.)
> "2. The explanation is irrelevant. We already know how many fixations there are because both the human and chimp genomes have been mapped. Those 40 million fixations must be accounted for."
What is 40 million fixations compared to the size of the genome and the number of loci? If your argument is that there wouldn't be time for that many de-novo mutations to accumulate, then as I pointed out, fixation of standing variation that existed at the time of our last common ancestor could have occurred instead, or in combination.
In any case, as I understand it, 200,000 generations with ~200 de-novo mutations per generation (which seems empirically plausible in humans) would result in ~40 million mutations over 5 million years? It would take about as long for such mutations to reach fixation (if they were all neutral, which they wouldn't be), but the numbers are at least in the same ballpark.
https://www.nature.com/articles/s41586-025-08922-2
Even mutations that occur in non-coding regions of the genome can be driven to fixation by 'genetic hitchhiking' effects/linkage-disequilibrium, if they are close to coding variants that are under selection.
> "3. Adaptation without genetic fixation is inherently anti-Modern Synthesis. That's more of an attack on Neo-Darwinism than anything I've written."
Are you saying that blacks and whites are not adapted to different levels of sun exposure? Or does this require that every allele for melanin expression in either population occur at either 0 or 100% frequency?
Read. The. Book.
All of these things are addressed there. Stop trying to be a Smart Boy and try actually learning something instead of listening to yourself talk, because you observably have no idea what you're talking about.
Just so you know- "Buy My Book" as a sum-total argument is not a great way to persuade me to actually buy your book.
Vox is wrong about parallel fixation. The post below has a good explanation.
https://www.lesswrong.com/w/speed-limit-and-complexity-bound-for-evolution-post-summary
It's telling that the example Vox gives for why parallel fixation doesn't work involves the asexually reproducing e. coli, when the whole power of parallel fixation relies on genetic recombination.
Fundamentally, I don't think Vox understands the mathematics either, he just understands it slightly better than the typical biologist.
You're completely wrong and you're relying upon a recanted study with multiple flaws that couldn't salvage parallel fixation anyhow.
You didn't even understand my case against parallel fixation.
Yes. I... haven't read the full book yet, but I did check out the first chapter or two and it seems strange that his example for "fastest ever fixation of mutations" is a study on asexually-reproducing bacteria with a generational length of 20 minutes? That seems unlikely to generalise well to eukaryotes in general, let alone mammals. (Bacteria do swap genes occasionally by various methods so clonal interference doesn't always occur, but still.)
The book doesn't seem to cite a direct reference, but I'm guessing the article in question is here?
https://sci-hub.se/https://www.nature.com/articles/nature08480
FWIW, there doesn't seem to be any particular section here stating "4600 generations per fixed mutation" were observed? There's a diagram showing a couple-dozen mutations appearing after 20,000 generations, and another section talking about one specific mutation reaching fixation between 26K and 29K generations, and a section talking about the persistence/fixation of various mutations in a drift scenario vs. under selection, but if Vox is getting 4.6K generations per fixation out of this I'd like to see the math.
i have a hard time believing that naming the book "Probability Zero" wasn't at least an indirect swipe at "random biological ejaculations from a godless liberal" Myers.
You can bet it’s going to eat away at old Paul Zachary on the inside, either. Surely he’ll think the same thing, true or not.
Vox be vindictive like that. he likes living rent free in their heads.
What is the methodology to do a text interview like this?
Maybe a live google doc? I imagine both of you read at a fast pace.
I wrote up the intro essay and then we just traded emails with the questions. Then I polished it up a bit to make myself sound smarter.
Based
I like Vox’s title, but I think “On the Origin of Specious” ain’t bad either
Are there any examples of things being claimed as mathematically impossible, that have been proven to be possible because of a flawed, faulty or incomplete maths?
And how is one to know if/when that is the case?
It would be very interesting to read a mathematician's analysis (and/or counter-critique), of this hypothesis.
Meanwhile, I'll keep raising chickens, noting that each generation (once per year) is becoming better suited to Winter and snow than the initial batch I got over five years ago. Then, they got frostbite on their feet when it was -20C outside. The ones born last year don't.
Possibly, that is mathematically impossible but it is empirically true.
Vox is wrong about parallel fixation. The post below has a good explanation.
https://www.lesswrong.com/w/speed-limit-and-complexity-bound-for-evolution-post-summary
It's telling that the example Vox gives for why parallel fixation doesn't work involves the asexually reproducing e. coli, when the whole power of parallel fixation relies on genetic recombination.
That's neither the reason nor the example I gave for why parallel fixation doesn't work. Quite the opposite actually.
You're also incorrect. Here is why your explanation doesn't save parallel fixation:
1. It requires truncation selection. The argument assumes you can cleanly eliminate "the lower half" of the population based on total mutational load. Real selection doesn't work this way. Selection acts on phenotypes, not genotypes. Two individuals with identical mutation counts can have wildly different fitness depending on which mutations they carry and how those interact with environment.
2. It assumes random mating. The sqrt(N) calculation depends on mutations being randomly distributed across individuals via random mating. But populations are structured, assortative mating occurs, and linkage disequilibrium means mutations aren't independently distributed.
3. It doesn't address the fixation problem. Even granting the sqrt(N) purging of deleterious mutations, this says nothing about the rate at which beneficial mutations can fix. Haldane's limit is about the cost of substituting good ones. Each beneficial fixation still requires selective deaths to drive it to fixation.
(1) and (2) are only true for the simplest models. Furthermore in the case of (2) many types of non-random mating, e.g., assortative mating, can actually cause deleterious mutations to be eliminated faster.
As for (3), I don't see why it matters if some deleterious mutations sticks around for a while at a tiny percentage.
You're obviously innumerate. If you want to see how hopeless your case is and you want to ignore my math, just have ChatGPT or whatever substitute a Moran model for the Wright-Fisher or Kimura models.
You have no idea how totally indefensible your case is from every possible angle.
My generational correction of d is far more favorable to the Modern Synthesis than the standard model math. Either way, no chance.
And now you drop all pretense of engaging in rational argument and resort to insults.
Given your position at this point, can't say I blame you.
There is no "rational argument" here. You haven't read the book. You have no idea what's going on. You're appealing to a mathematical impossibility.
You're just irrelevant. There is nothing else to say. All your blathering and all your Gamma posturing accomplishes absolutely nothing. Literally everyone can see that there is absolutely nothing to you.
The book is #1 in every science category. Biology. Genetics. Biology. It's a bestseller. Its arguments are rock-solid. You, on the other hand, have accomplished nothing except to strike poses in the comments of someone else's blog.
So either write a book yourself or shut the fuck up.
I'll have to check it out.
No slight on Vox intended but the way he replied to querys in the comment section, well it spoke volumes. Gave me vibes from the early 90s when some people tried to argue about "irreducible complexity" disproving evolution.
I have no respect or regard for retards who try to argue in self-admitted ignorance.
Literally no one who has read the book - and there are thousands now - have mustered one single serious critique of anything in it.
Nor will they. Because they'll have to come up with something that five different AIs didn't pick up in multiple rounds of Red Team Stress Testing.
Here's a thought for you, for free: AI is not reality. Follow that thought and you'll see where you went wrong from the outset.
Me, I'll keep growing crops and breeding animals, and will continue to notice how living being change from one generation to the next one, despite any "mathematical" models.
Consider this: in the 19th century it was considered proven that it was mathematically impossible for a human to run an English mile in under a minute. And it is - mathematically. Their math was right - their assumptions about reality however were wrong.
You have made the same mistake.
Go back to you initial assumptions and see if you can spot your own errors, and maybe you'll get a C- at least.
I haven't made any mistakes. It's not just the math. It's the combination of the correct math, the correct logic, and the empirical data that supports them both.
None of you ignorant critics who keep spouting off about something you haven't read, you don't know, and you clearly don't understand seem to realize is that the existence of the current genomic data and the aDNA databases has changed everything.
And, as it turns out, both Darwinism and the Modern Synthesis are hopelessly, haplessly wrong.
This web site provides an intriguing hypothesis and agree that the modern synthesis has problems:
https://www.macroevolution.net/human-origins.html
The author is well published in the hybridization literature.