Objections to Genetic Entropy

"Yes, there is a disturbingly high mutation rate in humans, which would lead to a too-high mutational load." — Joe Felsenstein, population geneticist and critic of genetic entropy¹

Every objection below is stated the way its best advocate would state it, not the way it is easiest to answer.  Each gets a verdict:

• Fatal — the argument does not survive this.
• Serious — a real problem that limits the claim.
• Partial — right about something, wrong about the conclusion.
• Weak — a fair point that does not do much damage.
• Confused — rests on a mistake.

Where the verdict is Fatal or Serious, that is a concession, and the main article is written to accommodate it.

About the premises and the parameters

1. "The genome is mostly junk, so U is small"

The objection.  Only 8–10% of the human genome is under purifying selection.  With 70 mutations per generation, that is fewer than 7 harmful mutations per person, and at the low end of the estimates fewer than 2.  Larry Moran states it flatly: "Since only 10% of our genome is functional DNA... we now know that those arguments were correct and 90% of our genome is junk."²

Response.  This is the load-bearing disagreement and it deserves to be treated as such rather than waved away.  Two things are worth saying.

First, the constraint estimates have been rising, not falling: 4.2% from 29 mammals in 2011, 8.2% from a turnover-corrected model in 2014, 10.7% from 240 mammals in 2023.^{3 4 5}  The last figure needs a caveat that cuts against this article:  10.7% is the share of the genome inside constrained elements, and the same paper's count of individually constrained bases is 3.3%, which gives U ≈ 2.25 rather than 7.3.  The element figure counts unconstrained bases inside elements and the single-base test is conservative, so the per-base fraction most likely lies between the two.  ENCODE's own leadership, writing after the 2012 controversy, put the union of estimates at 16–26%.⁶  The main article uses 8.2% — a mainstream number, not an ENCODE number — and gets U ≈ 5.6.

Second, there is a circularity worth naming, and it is not a creationist reading of the history.  The genetic load argument was used to argue for junk DNA.  Susumu Ohno, in the 1972 paper that coined the term, makes the inference in a single word:

"In fact, there seems to be a strict upper limit for the number of gene loci which we can afford to keep in our genome.  Consequently, only a fraction of our DNA appears to function as genes."⁷

His ceiling is U = 1.0, and the functional fraction he derives from it is 6%.⁷  Joe Felsenstein's textbook states the same logic thirty years later: "Thus the mutational load argument seems to give weight to the notion that this DNA is nonspecific in sequence."⁸  Using junk DNA to dissolve the load, when the load was the reason for believing in junk DNA, is circular.  That does not make the low estimates wrong — comparative genomics is independent evidence — but it does mean the two claims cannot both be used to support each other.

Verdict: Serious.  If the functional fraction really is 8%, U is around 5 and everything in the main article follows at that value.  If it is 2%, the effect is small enough to ignore.  The argument scales linearly with a number that is genuinely disputed.

2. "Only 0.4% to 0.7% of the genome is under strong selection"

The objection.  A direct within-human measurement across 71,702 genomes found only 0.4–0.7% of the genome "ultraselected," implying 0.26–0.51 strongly deleterious mutations per generation.⁹  That is an order of magnitude below even the low estimates.

Response.  Correct, and the number belongs in any honest account.  But "ultraselection" in that paper means near-total absence even among rare variants, which requires a large |s|.  Genetic entropy is specifically a claim about mutations below the threshold that method can detect.  The measurement is real and it constrains the strongly harmful class only.

Verdict: Partial.

3. "Your distribution of fitness effects is assumed, not measured"

The objection.  The gamma distribution with shape 0.2 is a curve fitted to data, and the fit is dominated by mutations selection can see.  The shape of the tail below 1/(2N_e) is extrapolation.

Response.  True, and this is one of the two or three most important limitations of the whole subject.  The main article's headline number moves by a factor of about two across the plausible range of the mean, and the shape parameter matters more than the mean.  What can be said is that the direction does not change: every measured distribution in every organism is skewed toward harm, and the extrapolation is into the region where mutations are hardest to remove.

Verdict: Serious.

4. "Below the drift barrier, mutations are as likely to help as harm"

The objection.  This is the strongest live objection and it goes to the heart of the argument.  Stephen Schaffner of the Broad Institute puts it this way: a genome produced by mutation and selection is one arbitrary draw from an enormous set of selectively equivalent configurations.  Within the effectively-neutral zone there is no privileged "correct" state, so drift among near-neutral variants has no direction.  The assumption that near-neutral mutations are systematically harmful is a consequence of assuming a designed, optimal starting genome — which is what the argument is trying to prove.

Response.  The circularity charge has force and should be conceded as a matter of logic: you cannot assume an optimal starting point and then present its decay as evidence for the optimal starting point.

But the premise can be argued directly rather than assumed.  Fisher's geometric model — which is where this objection comes from in the first place — guarantees an excess of harmful mutations at every effect size.  What it also says is that the excess shrinks toward an even split as effects get smaller, and small effects are exactly what is at issue.

The empirical situation is worse than that, and the article should not pretend otherwise.  No instrument routinely resolves the sign of a mutation in the zone the argument is about.  One method can, in narrow conditions — haplotype-based inference recovers sign in expanding populations, but only at 4N_e s of ±50 to ±100, which is |s| ≈ 10⁻³, about thirty times above the human drift barrier; in a constant-size population the same authors state that their method "cannot infer the sign of the selection coefficient."¹⁰  That is the honest version of the claim.  The best precision ever achieved on a single selection coefficient is about 2 × 10^-4, for three mutations in E. coli.  A typical distribution-of-effects study resolves 4 × 10^-3.  A deep mutational scan of the long-term E. coli experiment resolves 3 × 10^-2.  The human drift barrier is 4 × 10^-5 — five times finer than the best measurement ever made, and a hundred to seven hundred times finer than routine ones.

Deep mutational scanning does not rescue this.  Those studies find beneficial fractions above 1% among the mutations they can detect, and their own authors note that "fitness measurement errors are several orders of magnitude greater than the sensitivity of natural selection."

Felsenstein concedes the same gap from the other side: "it is very very hard to collect information on what fraction of those are deleterious" for effects below 1/(4N).¹

What can be said is that where the supply of beneficial mutations has been tracked as a population adapts, it contracts — from 6.8% to 3.2% over 15,000 generations in the long-term E. coli experiment, and to zero in high-fitness bacteriophage clones, which then declined under drift while low-fitness clones improved.

Verdict: Serious, and the single most important open question in the subject.

5. "Mutation rates were lower in the past"

The objection.  Modern human mutation rates reflect modern paternal ages and modern conditions.  Earlier populations with younger fathers had lower rates.

Response.  The paternal age effect is real and measured — about two extra mutations per year of the father's age.¹¹  Shifting the mean paternal age from 30 to 20 would cut the rate by roughly 20 mutations, about 30%.  That reduces U by 30%, which reduces the fixation rate by 30%.  It does not change the sign of anything.

Verdict: Weak.

About the population genetics

6. "The load formula compares everyone to a genotype that does not exist"

The objection.  1 − e^(−U) measures the population against a mutation-free individual who has never been born.  Under a relative-fitness model, Lesecque, Keightley and Eyre-Walker showed a species "could tolerate 10's or even 100's of new deleterious mutations per genome each generation."¹²  Charlesworth reached the same conclusion.¹³  Galeota-Sprung and colleagues applied it to the functional-fraction question and concluded that "an argument from mutational load does not appear to be particularly limiting on f."¹⁴

Response.  Conceded in full.  The load argument does not prove decline, and the main article does not use it as one.  The argument that survives is about fixation, which is untouched by this objection: a fixed mutation is carried by everyone, so there is no relative comparison left to make.

Verdict: Fatal — to the load argument specifically.

7. "Soft selection means nobody needs 800 children"

The objection.  Selection is largely competitive: a fixed number of slots filled by whoever does best on the day.  The absolute fitness of the population is irrelevant to who fills them.

Response.  Correct, and it is why the species is not extinct.  It is also why the degradation is invisible: soft selection determines who wins the competition, not how good the competitors are in absolute terms.  Eric Frost's analogy in an earlier discussion is apt — an accountant saying "we are $100,000 in debt but our competitors are $150,000 in debt, so we are ahead."

Verdict: Partial.  It answers "why are we alive."  It does not answer "is the genome degrading."

8. "Synergistic epistasis removes harmful mutations in bundles"

The objection.  If mutations are worse together than separately, selection can remove many at once, and the load collapses.  Kondrashov, Crow, and Eyre-Walker and Keightley all proposed it.  Sohail and colleagues reported evidence for it in humans and flies in Science in 2017.¹⁵

Response.  It reduces the load, which was never the surviving argument — and the empirical support for it is thin by the assessment of the person who measured the human mutation rate.  Keightley, in Genetics: "there is little empirical evidence that net synergistic epistasis for fitness is common."^{16 16}  Whether it removes nearly-neutral mutations depends on whether the interaction lifts their individual effects above 1/(2N_e), and that is an empirical question nobody has answered.  Note also that Keightley, one of the field's leading figures, has written that there is little empirical evidence that net synergistic epistasis for fitness is common.  The field is divided.

Verdict: Partial.

9. "Truncation selection is far more efficient than you allow"

The objection.  Crow showed that quasi-truncation selection — where individuals above a threshold survive and those below do not — removes mutations far more efficiently than independent selection, and he considered it realistic.¹⁷

Response.  Granted, and stronger than creationist sources usually admit.  Sanford's own group showed truncation halting mutation accumulation outright at ten new mutations per generation and six offspring per female, and quasi-truncation doing it too.¹⁸

The reply is not that truncation is weak.  It is that truncation selects on the total mutational burden, which is dominated by mutations of large effect, so a mutation contributing one part in a hundred thousand of that total does not decide which side of the threshold you fall on.

And the answer turns on one measurable number.  In their own runs, quasi-truncation with a fitness heritability of 0.2 reached a stable equilibrium; at 0.02 the population went extinct.  Bonnet and colleagues measured heritability of fitness across 19 wild vertebrate populations at about 0.03.¹⁹  That sits at the losing end of their own bracket — though Mendel's "heritability" is defined by adding scaled noise to genotypic fitness, which is not the same quantity as a wild-population animal-model estimate, so the comparison is suggestive rather than decisive.

Verdict: Partial, and the most quantitatively live objection in this list.

10. "Complex traits are polygenic, so selection is effective even where it is weak on each allele"

The objection.  Zach Hancock names this as the strongest argument against genetic entropy, ²⁰ and Nick Barton makes it in PNAS: under the infinitesimal model a trait is the sum of thousands of tiny-effect alleles, selection acts on the sum, and it can hold the sum at its optimum even where no single allele beats drift.  Barton applies it directly to this argument: "under a model of stabilizing selection on very many traits, any change that degrades fitness can be compensated."²¹  Charlesworth's polygenic load expression, worked with human numbers, gives about 0.05 rather than 0.89.¹³

Response.  Granted in full for what it claims, and it claims less than it appears to.  Trait-level selection is real, powerful, and demonstrated by the entire history of breeding.  Complex traits genuinely are polygenic.  And selection on a sum genuinely is far more effective than selection on any single term.

But read the mechanism.  Barton's word is compensated, not repaired.  The infinitesimal model tolerates drift at individual sites precisely because the specific alleles do not matter — only the sum does.  So what it predicts is a stable phenotype with the sequences underneath it turning over, which is this article's Argument 3 stated by the other side and counted as good news.  It answers the load argument, which this article already concedes, and does not touch the accounting argument.

The genuine limit on it is architectural.  It is strong for traits that really are sums, and weak for the parts of the genome that are switch-like: a splice acceptor either reads GT or it does not, and there is no compensating allele elsewhere that restores a sum, because there is no sum.  How the genome divides between those two regimes is unmeasured by both sides.

Verdict: Serious, and unresolved.  It is the best objection on this page.

11. "Larger populations solve it"

The objection.  The blind spot is 1/(2N_e) wide, so a bigger population sees more.

Response.  It helps, and how much depends entirely on which population size you mean.

For the historical effective size, the answer is: less than you would think.  Raising it a hundredfold, from 1,000 to 100,000, cuts the fixation rate by about 60% — not by a factor of a hundred, because the substitution rate for effectively neutral mutations is independent of population size.  The human long-term figure is around 10,000, not 100,000.

For the present and future effective size, the objection is much stronger and this article concedes it.  Joe Felsenstein applies his own 1971 overlapping-generations correction and gets roughly the census size divided by three — about 2.7 billion — with the consequence that "the dynamics of any new mutations, from now on, would be nearly-neutral if the selection coefficients were less than +/- 0.0000000003."²²  That is a blind spot five orders of magnitude narrower than the one this argument works with.  For mutations arising today, the ratchet has all but stopped.

The reply is narrow and it should not be stretched.  The load already fixed was fixed at the small ancient size, and fixation takes on the order of 4N_e generations, so a population explosion roughly ten generations old has not yet had time to change what fixes.  The ratchet rate computed in this article is a claim about the past.  Anyone quoting it as a forecast about modern humanity is misusing it.

Verdict: Weak for the past, Serious for the future — and the article's claims are about the past.

12. "Back mutation gives an equilibrium, so there is no runaway"

The objection.  Sites are finite.  Broken sites mutate back.  There is a mutation-selection-drift equilibrium and the genome sits at it.

Response.  Entirely correct, and calculable.  For a population of effective size 10,000, the equilibrium has any site worth less than about 0.01% of fitness sitting in the wrong state most of the time, and reaching it takes on the order of 80 million generations.  The objection is true, and the equilibrium it points to is a genome that has stopped working.

Verdict: Partial.

13. "If everyone declines equally, selection sees nothing — but they do not decline equally"

The objection.  Dan Stern Cardinale's internal-contradiction argument.  Genetic entropy needs mutations that both escape purifying selection and collectively degrade reproductive output.  But every individual carries a different set, so the degradation differs between individuals — and if it differs, selection sees it.  The escape hatch, that everyone declines equally, requires every individual to have identical reproductive output, which is false.

Response.  This is a serious argument and it deserves a quantitative answer rather than an assertion.

The answer is that selection does see the variance — and the simulations in the main article let it.  In the Full kernel, an individual's chance of reproducing depends on its own total mutational burden, exactly as the objection demands.  The decline still happens, because what selection can act on is the total, and the response to selection on any individual contributor is proportional to that contributor's effect.  A mutation worth 0.001% of fitness is selected against at 0.001% strength, which is below the noise of drift.  The objection assumes that visible variance in the total translates into effective removal of each component.  It does not, and the magnitude of the shortfall is calculable.

Verdict: Serious as posed, answered by the arithmetic.

14. "Selection acts on sperm and eggs too"

The objection.  Enormous numbers of gametes are produced and most fail.  That is another layer of selection.

Response.  Keightley and colleagues addressed this directly: "the fraction of sites in the genome effectively selected in germ line cell lineages is likely to be small, and most spontaneous abortions occur for nongenetic reasons or because of major genetic defects."¹²  A sperm cell expresses a small fraction of the genome.  Selection among sperm cannot police genes for eyesight.

Verdict: Weak.

About the simulations

15. "Mendel's Accountant is not a credible tool"

The objection.  It was written by creationists, published in a computer-science venue where the biology was not refereed by population geneticists, and has never been benchmarked against SLiM or any mainstream simulator.  Felsenstein: "it seems to have a number of arbitrary properties that make it differ a lot from the standard models."²³

Response.  Largely fair, and the main article does not use it.  Two qualifications.  Its central abstraction — genomes divided into linkage blocks carrying summed effects — was independently adopted by the Masel lab at Arizona for their own mainstream code, so the design is not creationist hackery.²⁴  And the one documented anomaly in its selection algorithm makes selection stronger than standard theory, which means the program understates mutation accumulation rather than exaggerating it.

Verdict: Serious against Mendel; irrelevant to the main article.

16. "Basener and Sanford's 2018 paper does not show what is claimed"

The objection.  Their central result depends on a truncation rule they impose by hand, and they say so in the paper: without the finite-population condition, "some numerical solutions went to an equilibrium."  Joe Felsenstein and Michael Lynch — the same Lynch quoted in the main article for the "very real costs" of accumulated defects — called the rule "not a reasonable way to take genetic drift into account, as all fitness classes are subject to random fluctuations."²⁵  Their model is also haploid, asexual and non-recombining, and in the declining case "the mutation rate is, in effect, 1000 times the selection coefficient against the mutant allele" — a condition under which Haldane showed selection loses, in 1927: "if selection acts against mutation, it is ineffective provided that the rate of mutation is greater than the coefficient of selection."²⁵  Hancock and Stern Cardinale published a formal reply in the same journal.²⁶

Response.  Conceded on the specific result.  The paper proves much less than the headlines claimed, and any article citing it as peer-reviewed support for genetic entropy without disclosing the truncation rule is being dishonest.

Two things should be said in its favour, because they are true.  It is a real extension of Fisher's equations to include mutation, published after review in a mathematics journal, and Basener is a mathematician writing mathematics.²⁷  And its qualitative headline — that Fisher's fundamental theorem does not guarantee fitness increase once mutation is in the model — is not a creationist eccentricity.  Warren Ewens, the field's authority on the theorem, put it far more harshly in Evolution in 2024: the theorem as Fisher gave it "was not correct and was derived from the explicitly demonstrated mathematical errors described above.  It should be banned from the literature."²⁸  Ewens mentions neither Basener, Sanford, nor creationism.

The residual point worth making is that U >> s is not obviously unrealistic for humans specifically, where U is in the units and individual s values are 10⁻⁵ to 10⁻³.  But that has to be argued, not assumed.

And the rebuttal has limits of its own, which can be checked because its authors deposited their code.  The simulation in Hancock and Stern Cardinale's reply is public.²⁹  Its parameters are not the regime genetic entropy is a claim about:

The decisive number is the last one nobody quotes.  With a gamma of mean 0.5 and shape 0.5, and N = 1,000, only about 2.5% of their deleterious mutations fall below the drift barrier — the other 97.5% are plainly visible to selection.  Their mean deleterious effect is 500 times their mean beneficial effect.  So the model demonstrates that selection removes mutations selection can see, which nobody disputes, rather than testing whether mutations selection cannot see accumulate, which is the actual claim.  Recombination a hundred times the human rate also suppresses Hill-Robertson interference, the exact mechanism that makes linked deleterious mutations hard to purge.  And their density rule, fitnessScaling = K / individualCount, scales fitness up as the population shrinks, which makes extinction close to structurally impossible — so "the population survived" carries little weight.

None of that rescues Basener and Sanford, and it should not be read as doing so.  Their paper's own defect stands on its own.  It does mean the rebuttal is a rebuttal of a specific model, not a demonstration that the underlying process does not occur — and its authors do not claim otherwise.

This article has not read the paper itself; it is paywalled with no open copy, confirmed against Unpaywall, OpenAlex, Europe PMC and Semantic Scholar.  The criticisms above are of the deposited simulation only.²⁹

Verdict: Fatal to the paper as usually cited; the narrow mathematical point stands and has better sources.

17. "Your own simulation ignores linkage, epistasis and changing environments"

The objection.  Real genomes are inherited in chunks, mutations interact, and environments change.

Response.  Linkage is omitted deliberately, because including it makes the result worse rather than better.  The Full kernel in Drift Lab measures the difference: at a human genetic map length the cost of linkage is small, and with no recombination at all the damage roughly triples.  Epistasis and changing environments are genuinely absent.  Changing environments could cut either way — a mutation that is harmful now may be useful later, but selection that fixes a currently-useful variant may also be fixing a future liability.

Verdict: Weak on linkage, fair on the rest.

About the evidence

18. "Mutation accumulation lines recover, so decline is reversible"

The objection.  Estes and Lynch took degraded C. elegans lines and restored full fitness in under 80 generations, at a rate "at least three times that of mutational degradation."³⁰  Katju and colleagues found no significant decline at all at population sizes of 10 or 100 over 409 generations.³¹

Response.  This is the strongest empirical evidence against the strong form of the claim and it should be stated without softening.  Two limits.  Recovery in MA lines is dominated by removal of the strongly harmful mutations that accumulate fastest under a single-individual bottleneck — the class selection is best at removing, and not the class at issue.  And a 409-generation experiment at N = 100 cannot detect a process operating on timescales of 4N generations with effects of order 1/(2N).

Verdict: Serious.

19. "Fruit flies would already be extinct"

The objection.  Scott Buchanan's arithmetic: at the 0.05% per generation decline measured in MA lines, over 72,000 generations of Drosophila evolution, fitness would fall to 2.3 × 10⁻¹⁷ of ancestral.  They are still here.

Response.  The calculation uses the wrong rate.  0.05% per generation is what MA lines lose under relaxed selection with a single-pair bottleneck.  The wild rate is the fixation-driven rate, and Drosophila has an effective population size around a million.  Running the same calculation at that population size gives a fitness loss over 72,000 generations of about 0.1%, not seventeen orders of magnitude.

That is a real answer, and it also concedes something important: at Drosophila's population size genetic entropy is negligible.  The argument is specifically about large-genome, small-population, low-fecundity species.  Humans are the paradigm case; flies are not.

Verdict: Confused as posed, but it correctly limits the scope of the claim.

20. "Bacteria and viruses have far higher mutation rates and are fine"

The objection.  RNA viruses mutate a million times faster per site than humans.  They persist.

Response.  Per site, yes; per genome per generation in functional DNA, no.  E. coli's genomic deleterious rate is on the order of 10^-4 per generation, and the authors who dismantle the load argument for humans concede it holds for microbes:  "For many microbes, then, Haldane load theory may be appropriate."¹⁴  Humans are at 2 to 10 — four to five orders of magnitude higher.  And a microbe with a burst size in the hundreds produces, every generation, offspring with no new mutations at all, which selection can then pick.  A mammal with four offspring cannot.

Verdict: Confused, and it argues for the thesis rather than against it.

21. "Most beneficial mutations work by breaking things, so they do not count as construction"

The objection.  Stated here as an objection to the mainstream position rather than to this article, because it is the argument creationists most often make and it is usually made too strongly.  The claim: a measured beneficial fraction of 1-7% cannot mean evolution is building anything, because nearly all of those mutations are beneficial by destruction — deleting an unneeded gene, breaking a regulator, shedding a cost.

Response.  The direction is right, the mechanism is documented, and the usual quantification is invented.

What is established: Hottes and colleagues found at least one beneficial null mutation in 139 of 144 environmental conditions.³²  A 2024 meta-analysis of 65 evolution experiments concludes loss-of-function mutations "play a basic role in rapid adaptation."³³  Behe's survey of laboratory evolution reached the same conclusion, and Jerry Coyne granted the observation while disputing its extrapolation.³⁴

What is not established is any percentage.  The same meta-analysis found nonsense substitution significantly exceeding synonymous substitution in 17 of 65 experiments, not 64 of 65.³³  Anyone claiming "99% of beneficial mutations are destructive" is quoting a number that does not exist in the literature.

Three further caveats that a careful critic will raise, all fair.  Regulatory retuning is not obviously destruction.  Duplication-and-divergence genuinely builds, and the long-term E. coli experiment's citrate innovation arose that way.  And gene loss can increase evolvability rather than end it, which Helsen and colleagues showed directly in yeast.³⁵

The defensible version of the objection is not a percentage at all.  It is that fitness assays cannot see mechanism, so the beneficial rate every balance calculation depends on is silent on the question those calculations are being used to answer.  That is a gap in the evidence, not a demonstration of the conclusion.

Verdict: Right in direction, unquantified, and routinely overstated by the side that likes it.

22. "Phage T7 got fitter under heavy mutagenesis"

The objection.  Springman, Keller, Molineux and Bull raised T7's mutation rate a hundredfold, giving 4.3 non-lethal mutations per genome per generation.  Theory predicted fitness would fall from 18.3 to 8.6 doublings per hour.  It rose to 21.9.  Their abstract: "Failure of the theory challenges the quantitative basis of lethal mutagenesis and highlights the potential for adaptive evolution at high mutation rates."³⁶

Response.  The best experimental result against genetic entropy, and it should be reported prominently.  The most likely explanation is the burst-size arithmetic above: at 2.6 deleterious mutations per generation, any organism producing more than about 13 offspring will on average produce one with no new mutations.  Note also what the same experiment found alongside the growth-rate gain: average burst size fell 80% and 245 mutations per genome accumulated.  The authors say the two results "appear to be contradictory."

Verdict: Serious for viruses, weak for mammals.

23. "Ancient DNA does not show the predicted decline"

The objection.  Studies of ancient genomes find that genomic health improved from the deep past toward the recent past — the opposite of a simple decay model.³⁷

Response.  True, and reported in the main article.  The same study also finds that people from a few thousand years ago score healthier than people today, which the authors call perplexing.  Neither side gets a clean result.  This is the sharpest available empirical test of the claim and it has not been run properly yet.

Verdict: Serious.

24. "The recent origin of deleterious variants is demography, not decay"

The objection.  Fu and colleagues found 86% of deleterious variants arose in the last 5,000 years, and creationists quote it constantly.  The authors' explanation is explosive recent population growth: a rapidly growing population accumulates rare variants faster than selection purges them.³⁸

Response.  The authors are right and the creationist use of this paper is careless.  The finding is consistent with genetic entropy but it is not evidence for it over the standard demographic explanation, which fits the data quantitatively.

Verdict: Fatal to that particular use of the paper.

25. "Human health and height and IQ have been improving"

The objection.  People are taller, healthier and score higher on IQ tests than a century ago.

Response.  Almost all of that is environmental.  Height gains track sanitation, reduced childhood infection and nutrition rather than genetics.³⁹  The Flynn effect runs at +3.0 IQ points per decade while the measured genetic trend in Iceland runs at −0.30 points per decade — the environment is outrunning the genetics by a factor of ten.⁴⁰  Crow said the same thing in 1997: mutation accumulation "is more than compensated for by much more rapid environmental improvements."

Verdict: Weak, and it concedes the underlying genetic trend.

26. "H1N1 influenza is not going extinct from mutation"

The objection.  Carter and Sanford's influenza paper counts reassortment events as mutations, its y-axis is mislabelled, and the 2009 pandemic lineage is not descended from the 1918 human lineage at all.  Strain turnover is antigenic drift under herd immunity, which is well supported and quantitative.  Codon bias as a fitness proxy is unreliable in viruses.

Response.  Conceded.  The main article does not lean on the influenza argument, and says so.

Verdict: Fatal to the influenza argument.

About the conclusions

27. "Crow, Muller and Lynch are being cited against their own meaning"

The objection.  Sanford cites all three as if they supported general genetic entropy.  They do not.  Lynch's 2016 abstract identifies the problem as "the recent detachment from the challenges of the natural environment," and restricts his ~1% per generation figure to "societies where medical care is widely applied."⁴¹  Crow's 1–2% is explicitly about the last few centuries in wealthy nations.  All three are describing relaxed selection under modern conditions, not an inherent failure of selection.

Response.  Conceded fully, and documented in the main article with both versions of the Crow quotation.  What these authors do support is the narrower claim that mutation is overwhelmingly harmful and that the human deleterious rate is high enough to be a genuine problem.  That is worth something, but it is not what they are usually quoted as saying.

Verdict: Fatal to that use of those sources.

28. "Sanford's book misquotes Eyre-Walker and Keightley"

The objection.  Page 170 of Genetic Entropy quotes the 1999 Nature paper with an ellipsis that removes the conditional clause "if mutational effects on fitness were to combine in a multiplicative way" and the following sentence naming the authors' own resolution.

Response.  Conceded, documented, and displayed side by side in the main article.

Verdict: Fatal to the quotation.  The underlying claim about U in that paper is accurately reported elsewhere.

29. "This is a young-earth argument working backwards"

The objection.  The conclusion was fixed before the analysis.  Sanford became a young-earth creationist and then wrote a book arguing the genome cannot be old.

Response.  Motive is not an argument, but the charge has enough force that the main article does not draw the young-earth conclusion.  What the numbers support is that a real, permanent process of degradation runs at a calculable rate, and that the mainstream answer to it depends on a compensating rate nobody has measured.  What the numbers do not support is a specific date.  Anyone claiming otherwise is doing arithmetic with parameters selected to reach an answer.

Verdict: Fair warning, and the reason this article states its limits.

30. "Nobody in the field takes this seriously"

The objection.  There is no peer-reviewed literature supporting genetic entropy, and Wikipedia, TalkOrigins and RationalWiki barely mention it.

Response.  On the label, true.  On the mechanism, false.  The process is called Ohta's ratchet in the mainstream literature and its fixation counts appear in a 2025 paper from the University of Arizona.⁴²  Kondrashov's paper asking why we have not died a hundred times over is in the Journal of Theoretical Biology.  Lynch's warnings are in PNAS and Genetics.  The disagreement is about magnitude and about what follows, not about whether the process exists.

Verdict: Confused about the mechanism, right about the label.

Definitional objections

31. "Name the disease. What does a genome dying of this actually look like?"

The objection.  If humans have been degrading for thousands of generations, there should be a clinical picture — a syndrome, a measurable decline, something a physician could point at.  There is not one.  An unfalsifiable process is not science.

Response.  The question is fair and the answer is uncomfortable for both sides.

The predicted end state is stated in this article: every weakly-constrained functional site sitting in the wrong state about half the time.  There is no single disease, because the damage is spread across every system at once and no locus carries much of it.  The clinical picture would be a slow rise in the background rate of everything — a bit more cancer, a bit more metabolic disease, a bit more psychiatric illness, all of it swamped by medicine, nutrition and environment on any timescale a person can observe.

That is a real prediction and it is also, by construction, below the resolution of any assay currently available.  A per-generation fitness decline of the size the theory predicts cannot be distinguished from zero against environmental improvement running in the other direction, which is exactly what Crow said when he estimated the Drosophila rate: it "is more than compensated for by much more rapid environmental improvements."¹⁷

So the honest position is that the objection is correct that the claim is not currently testable in humans, and incorrect that this makes it unfalsifiable in principle.  It is falsifiable in fast, small-genome, low-fecundity systems, and the closest thing to a test — mutation-accumulation lines and mutator strains — has produced mixed results this article reports in full.

Verdict: Fair, and the strongest reason to hold the conclusion loosely.

32. "If it is harmful, selection removes it — otherwise it is not harmful"

The objection.  "Deleterious" means "selected against."  A mutation selection cannot see is by definition not deleterious.

Response.  This defines the problem out of existence with vocabulary.  A broken gene is broken whether or not selection is strong enough to notice.  The article's currency is not a value judgement but a count: fixed substitutions in sequence that was previously constrained.

Verdict: Confused.

33. "Fitness depends on the environment, so decline is meaningless"

The objection.  A mutation that is harmful in one environment is neutral or useful in another.  There is no absolute standard.

Response.  Most harmful mutations are harmful in any environment a species actually lives in.  And where the objection is right — sickle cell, lactase persistence, loss of function that happens to pay off — it usually describes the destruction of something that used to work.  Burning a bridge can stop an invading army.  It does not explain where bridges come from.

Verdict: Partial, and mostly beside the point.

34. "Information is not well defined, so the accounting argument is empty"

The objection.  Creationists use "information" without a definition, and the definitions that exist do not do what they need.  Shannon entropy increases with randomness.

Response.  Correct about Shannon entropy, which is why this article does not use it.  The measure used is Hazen and Szostak's functional information in its simplest case: a site that requires one specific letter out of four carries 2 bits, because that is how much a random sequence would have to be narrowed down to hit it.⁴³  That is computable, and Drift Lab's Genome kernel computes it.  What is being counted is the number of nucleotide positions still carrying the letter their function requires.

Verdict: Fair warning, answered by using a defined measure.

What is left standing

Of the objections above, six are Fatal, nine are Serious, and the rest are partial or weaker.  The two strongest that are not fatal are the polygenic stabilizing-selection argument (#10) and the resolution limit on the distribution of fitness effects (#4); neither is resolved, and both are conceded in the main article rather than answered.  Every Fatal verdict falls on a use of the argument rather than the mechanism: the load formula, the Basener-Sanford paper, the influenza paper, the Crow and Lynch citations, the Eyre-Walker quotation, and the demographic reading of Fu 2013.  Strip all six away and what remains is:

1. Harmful mutations below 1/(2N_e) fix at close to the neutral rate. Not disputed.
2. At mainstream parameters that is around one per generation in humans. Agreed to within 12–26% by a mainstream 2025 simulation, depending on how its effect sizes are read.
3. Whether this matters depends on a beneficial mutation rate that has not been measured, and on whether compensation counts as restoration.

That is a smaller claim than Genetic Entropy the book makes.  It is also one that survives its critics.

Sources