Natural Selection of Information: Darwin Responds to Shannon

The Engineer and the Naturalist

I have spent decades observing—barnacles clinging to rocks, finches varying across islands, pigeons bred into endless forms by fanciers. My method has always been patient accumulation: gather specimens, note variations, trace patterns through geological time. From thousands of particular observations, general principles slowly emerge.

Reading Mr. Shannon’s work, I recognize a kindred spirit working in an entirely different domain. Where I spent years measuring beak depths and wing lengths, he spent years measuring bits, channels, and entropy. Where I studied the transmission of traits from parent to offspring, he studied the transmission of signals from sender to receiver. We are both, in our different ways, students of transmission.

Yet our methods could not be more different. Shannon builds theorems with mathematical precision—his channel capacity formula, $C = B \log_2(1 + S/N)$, establishes absolute limits with the certainty of algebra. I, by contrast, can only gesture toward principles through accumulated evidence. I cannot prove natural selection with an equation. I can only demonstrate its action through countless examples, comparing variations, tracing descent, observing the struggle for existence across continents and eons.

What strikes me most forcefully, however, is that despite these different approaches—his engineering, my naturalism—we arrive at the same fundamental insights. Redundancy protects against error. Noise corrupts transmission. Reliable communication is never guaranteed. These principles, which Shannon derives from first mathematics, nature discovered through the trial of billions of deaths.

Heredity as Noisy Channel

Shannon writes that “the fundamental problem of communication is reproducing at one point a message selected at another.” Between sender and receiver lies a channel, and every channel introduces noise.

Consider this framework applied to heredity. The parent organism is the sender. The offspring is the receiver. The message is the totality of instructions encoded in what we now call DNA—the blueprints for building a functioning organism. Between parent and offspring lies the channel of reproduction.

And what a noisy channel it is! Copying errors occur with every generation. Environmental insults—radiation, chemical damage, heat—corrupt the genetic text. The message that offspring receive is never perfectly identical to the message parents sent. Variation, which I identified as the raw material of natural selection, is precisely the noise Shannon describes: unavoidable corruption introduced by imperfect transmission.

But here is what I find most remarkable about Shannon’s framework: he demonstrates that redundancy fights entropy. By adding controlled redundancy—extra information encoding relationships between message elements—receivers can detect and correct errors. The overhead reduces raw transmission rate but increases reliable throughput.

Now consider DNA’s codon degeneracy—Shannon’s own example. Sixty-four possible triplet combinations encode only twenty amino acids. Multiple codons specify the same protein building block. This is not inefficiency but insurance. When mutation corrupts a single nucleotide, the degenerate code often produces the same amino acid regardless. The organism is buffered against transcription noise.

I spent twenty years after returning from the Beagle before publishing On the Origin of Species, precisely because I understood that such claims required overwhelming evidence. Now Shannon gives me the mathematical framework to understand why this redundancy exists. Organisms without sufficient error correction—those whose genetic codes permitted every mutation to catastrophically alter proteins—simply did not survive. Natural selection, acting over billions of years, discovered optimal coding schemes by eliminating the alternatives.

Shannon’s engineers design redundancy. Evolution discovers it through extinction.

Population as Redundancy

Shannon’s analysis of neural population coding illuminates another evolutionary strategy I have long pondered. Why do populations exist at all? From a purely mechanistic view, would not a single perfect individual suffice?

The brain, Shannon observes, faces unreliable channels at every synapse. Its solution: population coding. Rather than relying on single neurons, information is distributed across many parallel channels. Individual cellular failures are tolerated because redundant encoding preserves the message.

I see the same principle operating at the level of species. A population is not merely a collection of individuals—it is a redundancy system. If one organism fails through predation, disease, or environmental catastrophe, others survive to transmit their genetic information forward. The population-level message—the accumulated adaptations refined over countless generations—persists even when individual carriers are lost.

Genetic diversity within populations extends this principle further. A population of identical individuals might be exquisitely adapted to current conditions but catastrophically vulnerable to environmental change. Variation, the very noise that Shannon identifies as the enemy of reliable transmission, becomes at the population level an insurance policy against unknown future conditions.

The thermophilic archaea Shannon describes have pushed this strategy to its logical extreme. Living in environments where heat continuously damages DNA, these organisms cannot prevent corruption. Instead, they accept ongoing errors and repair through communal DNA exchange. Each cell serves as a backup for its neighbors. Recombination reconstructs damaged sequences by borrowing intact copies from the population. This is Shannon’s distributed error correction implemented through biological mechanism, discovered not through mathematical analysis but through billions of years of evolutionary pressure.

I cannot help but feel humility before such elegant solutions. Shannon derives these principles through brilliant theoretical work. Life discovered them through the blind process of variation and selection, without any understanding of the mathematics involved. The thermophile has no concept of information theory—yet its survival strategy instantiates Shannon’s theorems more robustly than any engineered system.

Natural Selection as Information Filter

Shannon defines information as the resolution of uncertainty. Before a message arrives, multiple possibilities exist. The message eliminates possibilities, narrowing outcomes. Information is measured by this reduction.

Natural selection operates as precisely such an information filter. Before selection acts, a population contains variation—many possible phenotypes, many possible genetic configurations. The environment then presents challenges: predators, climate, competition for resources. Selection eliminates variants poorly suited to these challenges. The population that survives carries information about its environment, encoded in the very adaptations that permitted survival.

Over generations, this filtering accumulates. Each round of selection resolves uncertainty about what configurations permit survival. The signal—adaptive traits—is preserved and refined. The noise—maladaptive variants—is eliminated. The signal-to-noise ratio of the population, if I may borrow Shannon’s terminology, improves with each selective event.

But here lies a crucial tension between Shannon’s framework and evolutionary reality. Shannon’s engineers design for known channels—fixed noise characteristics, stable transmission conditions. Evolution cannot predict future environments. The channel itself changes unpredictably across geological time. Ice ages arrive. Continents drift. Competitors invade. The noise statistics of the hereditary channel are never stable.

Evolution’s response to this uncertainty is, again, diversity. Rather than optimizing for a single fixed channel, populations maintain variation—exploring multiple possible solutions simultaneously. When the channel changes, some variants that were previously neutral or maladaptive may suddenly become advantageous. The population hedges its information-theoretic bets through maintained diversity.

This is not optimal engineering. It is robust engineering—design for unknown unknowns through distributed exploration.

The Mathematics of What Nature Knew

Shannon gave us the language to describe what life has been doing for four billion years. His theorems formalize principles that natural selection discovered empirically through the elimination of alternatives. Every organism alive today is proof that error correction works—their ancestors successfully transmitted genetic information through the noisy channel of reproduction. Every extinction event is proof that channel capacity can be exceeded—that populations sometimes fail to maintain sufficient redundancy or adaptability.

What I find most profound in Shannon’s work is his observation that information is medium-independent. The mathematics of entropy and redundancy apply whether transmitted through copper wire, genetic molecule, or neural spike. Similarly, I have observed that adaptive principles follow the same laws regardless of organism. The struggle for existence shapes finch and barnacle, orchid and human, through identical mechanisms. Universal principles transcend specific instances.

The struggle for existence, I now see, is fundamentally a struggle against entropy. Organisms must transmit their accumulated adaptations—information refined over billions of years—through noisy hereditary channels. Natural selection is nature’s error-correction algorithm, preserving signal and eliminating noise across generations.

Shannon and I, engineer and naturalist, arrive at the same destination through different paths. He derives from theory what evolution discovered through practice. Together, our work reveals something profound: that the deepest principles of communication and survival are one and the same.