Genetic Information
Claims that mutations cannot add information depend on a definition of information that is never given precisely.
6 min readUpdated
A common creationist argument holds that mutations can only degrade genetic information, never create it, so evolution cannot produce genuinely new features however long it runs.
The argument's strength depends entirely on what "information" means. Where the term is given a formal definition, the claim is testable and fails. Where it is not, the claim cannot be evaluated.
The claim and its sources
| Proponent | Formulation |
|---|---|
| Werner Gitt, In the Beginning Was Information (1997) | "Universal information" requires a sender with intent; matter cannot produce it |
| Answers in Genesis | Mutations shuffle or lose information; none has been observed to add any |
| Lee Spetner, Not by Chance (1997) | Observed beneficial mutations reduce specificity, so they lose information |
| John Sanford, Genetic Entropy (2005) | Deleterious mutations accumulate faster than selection removes them, so genomes degrade |
Definitions of information
| Definition | What it measures | Can mutation increase it? |
|---|---|---|
| Shannon information | Uncertainty reduced by a message; a function of sequence length and symbol probabilities | Yes. Any duplication increases it; random change to a sequence generally increases entropy |
| Kolmogorov complexity | Length of the shortest program producing the string | Yes. Random sequences have maximal Kolmogorov complexity |
| Functional information | Number of sequences achieving a specified function at a specified level | Yes, and it has been measured |
| Gitt's "universal information" | Requires a sender, intent, and meaning by definition | No, by construction |
The last row is the difficulty. Gitt's definition includes an intelligent sender as part of what information is, so concluding that information requires an intelligent sender restates the definition rather than discovering anything.
Measured increases
Gene duplication and divergence. A copying error duplicates a gene; the copy is then free to change while the original continues its function.
| Case | Detail |
|---|---|
| Human colour vision | The red and green opsin genes arose by duplication of an ancestral gene and sit adjacent on the X chromosome, about 98% identical |
| Antifreeze glycoprotein in Antarctic notothenioid fish | Evolved from a trypsinogen digestive enzyme gene, with the transitional sequence still identifiable |
| Human amylase | Populations with starch-rich diets carry more copies of AMY1, up to about 15, with correspondingly more enzyme |
| RNASE1B in leaf-eating monkeys | A duplicate of a digestive enzyme adapted to a different pH, with the original retained |
New functions from scratch. Lenski's E. coli experiment produced aerobic citrate metabolism after about 31,500 generations, a capacity used to define the species as lacking it. The frozen archive showed it required a prior potentiating mutation, then a duplication placing an existing citrate transporter under a promoter active in oxygen.
Nylonase. Bacteria able to digest nylon oligomers appeared within decades of the material's invention in 1935. Nylon did not exist before then, so the enzyme's substrate is a genuinely novel compound.
Measured functional information. Hazen and colleagues (2007) defined functional information formally and measured it increasing under selection in RNA binding experiments — a direct laboratory demonstration.
Common objections
"Duplication only copies existing information — it adds nothing new"
The copy alone contains nothing new.
The sequence of events matters. Duplication first supplies redundancy, which removes the selective constraint on one copy; that copy then accumulates changes that would have been fatal in a single-copy gene, and may acquire a new function while the original continues.
The end state is two genes doing different things where there was one. Whether that counts as "new information" depends on the definition, but the organism has a capacity it did not have.
The red and green opsins are a concrete case: trichromatic colour vision in Old World primates is a function absent in the ancestral two-gene state, and the sequence record of the duplication is still visible.
"Beneficial mutations still lose specificity — they break something"
Lee Spetner's argument, and it is correct for many well-known examples. Antibiotic resistance often results from a loss — a broken porin channel that no longer admits the drug, or a disabled regulatory gene.
Behe's Darwin Devolves (2019) generalises this into the claim that adaptation typically proceeds by breaking things.
The generalisation does not hold across the cases above. Nylonase is an enzymatic activity for a compound that did not previously exist. The Lenski citrate result added a regulatory context enabling transport under new conditions. Antifreeze glycoprotein is a functioning protein derived from a digestive enzyme.
It is true that loss-of-function mutations are more common than gain-of-function ones, because there are more ways to break a system than to improve it. That is a statement about relative frequency, not about possibility.
"Genetic entropy means genomes are degrading, not improving"
John Sanford's argument, and he is a genuine geneticist — he co-developed the gene gun. The claim is that most mutations are slightly deleterious, too small for selection to detect individually, so they accumulate and fitness declines inexorably.
The concept behind it, Muller's ratchet, is real and does operate in asexual populations without recombination.
Sexual reproduction is the standard answer: recombination allows offspring to be produced with fewer deleterious mutations than either parent, which is a principal explanation for why sex is maintained despite its costs.
The empirical problem is that the predicted decline is not observed. Sanford's model implies human fitness should have collapsed over recorded history. Lenski's populations, at 75,000 generations with no recombination and a high mutation rate, show fitness continuing to increase rather than decline.
"Shannon information isn't the relevant kind — meaning is what matters"
A fair criticism of a purely Shannon-based reply. Shannon information deliberately excludes meaning; it measures uncertainty reduction, so random noise scores high.
This is why functional information exists as a measure. Hazen's formulation counts sequences achieving a specified function at a specified level, which captures the relevant sense and remains measurable.
The measurement has been made and it increases under selection. Any definition that cannot be measured cannot support a claim about what mutations can or cannot do.
"You can't get a new gene from random mutation — the odds are impossible"
Calculations of the probability of assembling a specific protein at random produce very small numbers, and they are arithmetically correct.
They calculate the wrong thing. Evolution does not assemble proteins at random in one step; it modifies existing sequences incrementally, retaining improvements. The probability of a specific 300-residue protein arising in a single event is irrelevant to a process that does not work that way.
The calculations also usually assume one target sequence. Where the size of the functional space has been measured — as in Keefe and Szostak's 2001 work on ATP-binding proteins from random libraries — functional sequences turn out to be far more common than the single-target assumption implies.
What the evidence shows
Mutation and selection have been observed producing new genes, new enzymatic functions, and increases in formally measured functional information.
Gene duplication followed by divergence is the principal mechanism, and its products are identifiable in sequence: the duplicates remain recognisably related to their originals.
The claim that information cannot increase holds only under a definition that builds an intelligent sender into the meaning of the word, which makes it a definition rather than a finding.
The related design argument is in Irreducible Complexity.