Specified Complexity

In 1998, mathematician and philosopher William Dembski published The Design Inference, a book that attempted to put the detection of intelligence on a rigorous mathematical footing. The question he was asking was one that forensic scientists, archaeologists, and SETI researchers already assume an answer to in practice: what distinguishes an event or pattern produced by intelligence from one produced by chance or natural law? Dembski's answer was that intelligence leaves a distinctive fingerprint he called complex specified information — a combination of high improbability and independent describability that cannot be generated by undirected processes. If the living cell contains complex specified information, as Dembski argued it does, then the inference to an intelligent cause is not merely a religious intuition but a conclusion following the same logic we already use to detect design everywhere else. His argument became the mathematical core of the Intelligent Design movement — and attracted some of the most technically sophisticated criticism in the creation-science debates.

The two ingredients

Dembski's argument begins with a distinction that seems simple but has significant technical depth. Any pattern we observe can be characterized by two independent properties:

Complexity — the probability of the pattern arising by chance. A coin flip landing heads has a complexity of 1 in 2; a specific sequence of a thousand coin flips has a complexity of 1 in 2¹&sup0;&sup0;&sup0;, which is a number so small it is effectively impossible by pure chance. Complexity is essentially what information theorists call Shannon information: a measure of how surprising or improbable a pattern is.

Specification — whether the pattern matches an independently describable target. A thousand coin flips landing in the exact sequence you get is complex, but it is no more interesting than any other sequence of a thousand coin flips. If, however, the sequence of heads and tails, when converted to binary and decoded, spells out a message in English, the pattern is both complex (still 1 in 2¹&sup0;&sup0;&sup0;) and specified — it matches an independently describable pattern (the English language and the ASCII encoding) that has nothing to do with coins. Specification is what rules out the possibility that the complexity arose by luck; a pattern that matches an independently describable target is not something we reasonably attribute to chance.

Dembski calls the combination of these two properties "complex specified information," or CSI. The core claim: CSI is reliably produced by intelligence, and there is no known undirected process that generates it. When we find a skull with a bullet hole, we infer murder rather than natural weathering. When we find the Rosetta Stone, we infer inscription rather than erosion. The inference to design is not magic; it is the recognition of a pattern that chance and law do not produce.

The application to biology

Dembski's argument, and the Intelligent Design case generally, is that the DNA molecule contains exactly this kind of complex specified information in quantities that dwarf anything we encounter elsewhere. The human genome contains approximately 3 billion base pairs, organized into genes that code for functional proteins. The specifications are the functional requirements of life: a sequence of amino acids either folds into a working protein or it does not. The probability of arriving at a functional protein sequence by random mutation — even given billions of years and the earth's entire population of organisms — is, on Dembski's calculation, vanishingly small; he computes a "universal probability bound" of 1 in 10¹&sup5;&sup0; beyond which we can reasonably conclude that chance is not the explanation.

The bacterial flagellum — already discussed on the Irreducible Complexity page — is one instantiation of this argument. The DNA encoding the flagellum's 30-some proteins contains CSI: it is highly improbable and it is specified by the functional requirement of a working rotary motor. But Dembski's argument is broader than any single molecular machine. It is about the information content of the genome as a whole, and the question of whether natural selection acting on random mutations — an undirected process — can generate the kind of information we observe in living systems.

The mainstream response

Critics of Dembski's argument have attacked both the mathematical framework and its application to biology. The principal objections:

The specification criterion is either vague or question-begging. Biologists argue that Dembski's notion of "specification" smuggles in the design conclusion. The specifications for biological function — "must fold into a working protein," "must drive a flagellum" — are not independent of the biology; they are derived from observing what living things do. When we call a protein "specified," we are observing that it works in a biological system that already exists. Critics argue this is circular: we call it specified because it functions in something we already know is designed.

The probability calculations ignore cumulative selection. Dembski's numbers treat the arrival of a functional protein as a single random draw from all possible sequences. But evolution does not work that way. Natural selection is not random; it preferentially preserves sequences that are more functional and eliminates those that are less. Richard Dawkins's "Weasel" demonstration and the broader theory of cumulative selection show that what appears to be an astronomically improbable target can be reached step by step, where each step is individually much more probable. Dembski is aware of this objection and has responses to it, but critics argue his responses don't fully close the problem.

The No Free Lunch theorems don't apply as Dembski claims. Dembski drew on the mathematical "No Free Lunch" (NFL) theorems of Wolpert and Macready, which show that no optimization algorithm outperforms any other when averaged over all possible problems. Dembski argued this means Darwinian evolution cannot generate CSI. Wolpert himself, along with other mathematicians, responded that the NFL theorems simply do not apply to biological evolution the way Dembski claimed: the theorems concern averaging over all possible fitness landscapes, and biological evolution operates on specific, structured fitness landscapes, not random ones.

The displacement problem response. Dembski argues that even if evolution can generate CSI, the information required to structure the evolutionary process itself had to come from somewhere — pushing the problem back one level. Critics argue this regression applies equally to any designer: where did the designer's information come from? Dembski's answer is that the designer is not a contingent being requiring explanation in the same way.

Where the argument is strongest and where it strains

The specification-complexity argument is strongest when applied to the origin of the first information-bearing molecule. Even granting everything that Darwinian evolution can do — cumulative selection, large populations, long time — it requires a self-replicating molecule to start. The origin of the first replicator is not explained by natural selection because natural selection requires a replicator to operate. The problem of getting the first information-bearing sequence, in a world without selection, is genuinely hard, and this is where Dembski's argument has the most traction (the Origin of Life page covers this in detail).

The argument strains when applied to the subsequent evolution of living things. Given a self-replicating system and natural selection, the question of how much complexity selection can generate is an open empirical question, and the mainstream answer — that it can generate a great deal — has significant experimental support. Using the absence of a detailed mechanistic account of every evolutionary step as evidence for design is the "God of the gaps" pattern this site generally avoids: the gap may yet be filled.

The honest verdict

Dembski's contribution was to attempt a mathematical formalization of what many people intuit about information and intelligence — that information has a source, and that the source of the information in DNA is a question worth asking rigorously. That attempt drew genuinely technical criticism that has not been fully resolved in Dembski's favor, and the mainstream scientific community does not accept CSI as a reliable design detector.

What remains after the technical debate is what was there before it: the argument from the information content of living systems to an intelligent origin is not unreasonable, and it rests on an observation that everyone grants — that DNA carries information, that information is ordinarily associated with minds, and that the inference from information to intelligence is not in itself irrational. What is contested is whether that inference can be formalized in Dembski's way, and whether it can be sustained against the cumulative-selection reply for biological evolution past the origin point.

The psalmist's observation precedes the mathematics by three thousand years: "For thou hast possessed my reins: thou hast covered me in my mother's womb. I will praise thee; for I am fearfully and wonderfully made: marvellous are thy works; and that my soul knoweth right well" (Psalm 139:13–14). The argument from design in living systems is not new with Dembski. What Dembski attempted was to give it a mathematical vocabulary that could meet science on its own terms. Whether that attempt fully succeeded is a technical question. That the design intuition it formalized is reasonable is harder to dismiss.

Related: Irreducible Complexity (Behe's molecular-machine argument, a specific case of the broader ID claim), DNA and the Question of Information (the information content of DNA explored directly), The Origin of Life (where Dembski's argument is strongest), The Fine-Tuning of the Universe (the cosmological design argument), Arguments Worth Retiring (on the "God of the gaps" risk). Primary sources: William Dembski, The Design Inference (Cambridge, 1998); No Free Lunch (Rowman & Littlefield, 2002). Counter-arguments: Wolpert (1996); Mark Perakh, Unintelligent Design (Prometheus, 2003); Elliott Sober, Evidence and Evolution (Cambridge, 2008).