Signs of Intelligence

A Primer on the Discernment of Intelligent Design

by William A. Dembski

Intelligent design examines the distinction between three modes of explanation: necessity, chance, and design. In our workaday lives we find it important to distinguish between these modes of explanation. Did she fall or was she pushed? And if she fell, was it simply bad luck or was her fall unavoidable? More generally, given an event, object, or structure, we want to know:

1. Did it have to happen?
2. Did it happen by accident?
3. Did an intelligent agent cause it to happen?

Given an event to be explained, the first thing to determine is whether it had to happen. If so, the event is necessary. By “necessary” I don’t just mean logically necessary, as in true across all possible worlds, but I also include physical necessity, as in a law-like relation between antecedent circumstances and consequent events. Not all events are necessary.

Events that happen but do not have to happen are said to be contingent. In our workaday lives we distinguish two types of contingency, one blind, the other directed. A blind contingency lacks a superintending intelligence and is usually characterized by probabilities. Blind contingency is another name for chance. A directed contingency, on the other hand, is the result of a superintending intelligence. Directed contingency is another name for design.

An Ancient Question

This characterization of necessity, chance, and design is pretheoretical and therefore inadequate for building a precise scientific theory of design. We therefore need to inquire whether there is a principled way to distinguish these modes of explanation. Philosophers and scientists have disagreed not only about how to distinguish these modes of explanation, but also about their very legitimacy. The Epicureans, for instance, gave pride of place to chance. The Stoics, on the other hand, emphasized necessity and design, but rejected chance. In the Middle Ages Moses Maimonides contended with the Islamic interpreters of Aristotle who viewed the heavens as, in Maimonides’s words, “the necessary result of natural laws.” Where the Islamic philosophers saw necessity, Maimonides saw design.

In arguing for design in his Guide for the Perplexed, Maimonides looked to the irregular distribution of stars in the heavens. For him that irregularity demonstrated contingency. But was that contingency the result of chance or design? Neither Maimonides nor the Islamic interpreters of Aristotle had any use for Epicurus and his views on chance. For them chance could never be fundamental but was at best a placeholder for ignorance. Thus for Maimonides and his Islamic colleagues the question was whether a principled distinction could be drawn between necessity and design. Maimonides, arguing from observed contingency in nature, said yes. The Islamic philosophers, intent on keeping Aristotle pure of theology, said no.

A Modern Demise

Modern science has also struggled with how to distinguish between necessity, chance, and design. Newtonian mechanics, construed as a set of deterministic physical laws, seemed only to permit necessity. Nonetheless, in the General Scholium to his Principia, Newton claimed that the stability of the planetary system depended not only on the regular action of the universal law of gravitation, but also on the precise initial positioning of the planets and comets in relation to the sun. As he explained:

---

Though these bodies may, indeed, persevere in their orbits by the mere laws of gravity, yet they could by no means have at first derived the regular position of the orbits themselves from those laws. . . . [Thus] this most beautiful system of the sun, planets, and comets, could only proceed from the counsel and dominion of an intelligent and powerful being.

Like Maimonides, Newton saw both necessity and design as legitimate explanations, but gave short shrift to chance.

Newton published his Principia in the seventeenth century. By the nineteenth century, necessity was still in, chance was still out, but design had lost much of its appeal. When asked by Napoleon where God fit into his equations of celestial mechanics, astronomer and mathematician Laplace famously replied, “Sire, I have no need of that hypothesis.” In place of a designing intelligence that precisely positioned the heavenly bodies, Laplace proposed his nebular hypothesis, which accounted for the origin of the solar system strictly as the result of natural gravitational forces.

Since Laplace’s day, science has largely dispensed with design. Certainly Darwin played a crucial role here by eliminating design from biology. Yet at the same time science was dispensing with design, it was also dispensing with Laplace’s vision of a deterministic universe (recall Laplace’s famous demon who could predict the future and retrodict the past with perfect precision provided that present positions and momenta of particles were fully known). With the rise of statistical mechanics and then quantum mechanics, the role of chance in physics came to be regarded as ineliminable. Consequently, a deterministic, necessitarian universe has given way to a stochastic universe in which chance and necessity are both regarded as fundamental modes of scientific explanation, neither being reducible to the other. To sum up, contemporary science allows a principled distinction between necessity and chance, but repudiates design.

Bacon & Aristotle

But was science right to repudiate design? My aim in The Design Inference (Cambridge University Press, 1998) is to rehabilitate design. I argue that design is a legitimate and fundamental mode of scientific explanation on a par with chance and necessity. Since my aim is to rehabilitate design, it will help to review why design was removed from science in the first place. Design, in the form of Aristotle’s formal and final causes, had after all once occupied a perfectly legitimate role within natural philosophy, or what we now call science. With the rise of modern science, however, these causes fell into disrepute.

We can see how this happened by considering Francis Bacon. Bacon, a contemporary of Galileo and Kepler, though himself not a scientist, was a terrific propagandist for science. Bacon was concerned about the proper conduct of science and provided detailed canons for experimental observation, the recording of data, and drawing inferences from data. What interests us here, however, is what he did with Aristotle’s four causes. For Aristotle, to understand any phenomenon properly, one had to understand its four causes, namely its material, efficient, formal, and final cause.

Two points about Aristotle’s causes are relevant to this discussion. First, Aristotle gave equal weight to all four causes and would have regarded any inquiry that omitted one of his causes as fundamentally deficient. Second, Bacon adamantly opposed the inclusion of formal and final causes within science (see his Advancement of Learning). For Bacon, formal and final causes belonged to metaphysics and not to science. Science, according to Bacon, needed to limit itself to material and efficient causes, thereby freeing science from the sterility that inevitably results when science and metaphysics are conflated. This was Bacon’s line, and he argued it forcefully.

We see Bacon’s line championed in our own day. For instance, in his book Chance and Necessity, biologist and Nobel laureate Jacques Monod argued that chance and necessity alone suffice to account for every aspect of the universe. Now whatever else we might want to say about chance and necessity, they provide at best a reductive account of Aristotle’s formal causes and leave no room for Aristotle’s final causes. Indeed, Monod explicitly denies any place for purpose within science.

Now I don’t want to give the impression that I’m advocating a return to Aristotle’s theory of causation. There are problems with Aristotle’s theory, and it needed to be replaced. My concern, however, is with what replaced it. By limiting scientific inquiry to material and efficient causes, which are of course perfectly compatible with chance and necessity, Bacon championed a view of science that could only end up excluding design.

The Design Instinct

But suppose we lay aside a priori prohibitions against design. In that case, what is wrong with explaining something as designed by an intelligent agent? Certainly there are many everyday occurrences that we explain by appealing to design. Moreover, in our daily lives it is absolutely crucial to distinguish accident from design. We demand answers to such questions as, Did she fall or was she pushed? Did someone die accidentally or commit suicide? Was this song conceived independently or was it plagiarized? Did someone just get lucky on the stock market or was there insider trading?

Not only do we demand answers to such questions, but entire industries are also devoted to drawing the distinction between accident and design. Here we can include forensic science, intellectual property law, insurance claims investigation, cryptography, and random number generation—to name but a few. Science itself needs to draw this distinction to keep itself honest. As a January 1998 issue of Science made clear, plagiarism and data falsification are far more common in science than we would like to admit. What keeps these abuses in check is our ability to detect them.

If design is so readily detectable outside science, and if its detectability is one of the key factors keeping scientists honest, why should design be barred from the actual content of science? There’s a worry here. The worry is that when we leave the constricted domain of human artifacts and enter the unbounded domain of scientific inquiry, the distinction between design and non-design cannot be reliably drawn. Consider, for instance, the following remark by Darwin in the concluding chapter of his Origin of Species:

Several eminent naturalists have of late published their belief that a multitude of reputed species in each genus are not real species; but that other species are real, that is, have been independently created. . . . Nevertheless they do not pretend that they can define, or even conjecture, which are the created forms of life, and which are those produced by secondary laws. They admit variation as a vera causa in one case, they arbitrarily reject it in another, without assigning any distinction in the two cases.

It’s this worry of falsely attributing something to design (here construed as creation) only to have it overturned later that has prevented design from entering science proper.

This worry, though perhaps understandable in the past, can no longer be justified. There does in fact exist a rigorous criterion for discriminating intelligently from unintelligently caused objects. Many special sciences already use this criterion, though in a pretheoretic form (e.g., forensic science, artificial intelligence, cryptography, archeology, and the Search for Extra-Terrestrial Intelligence). In The Design Inference I identify and make precise this criterion. I call it the complexity-specification criterion. When intelligent agents act, they leave behind a characteristic trademark or signature—what I call specified complexity. The complexity-specification criterion detects design by identifying this trademark of designed objects.

The Complexity-Specification Criterion

A detailed explanation and justification of the complexity-specification criterion is technical and can be found in The Design Inference. Nevertheless, the basic idea is straightforward and easily illustrated. Consider how the radio astronomers in the movie Contact detected an extraterrestrial intelligence. This movie, based on a novel by Carl Sagan, was an enjoyable piece of propaganda for the SETI research program—the Search for Extra-Terrestrial Intelligence. To make the movie interesting, the SETI researchers in Contact actually did find an extraterrestrial intelligence (the non-fictional SETI program has yet to be so lucky).

How, then, did the SETI researchers in Contact convince themselves that they had found an extraterrestrial intelligence? To increase their chances of finding an extraterrestrial intelligence, SETI researchers monitor millions of radio signals from outer space. Many natural objects in space produce radio waves (e.g., pulsars). Looking for signs of design among all these naturally produced radio signals is like looking for a needle in a haystack. To sift through the haystack, SETI researchers run the signals they monitor through computers programmed with pattern-matchers. So long as a signal doesn’t match one of the pre-set patterns, it will pass through the pattern-matching sieve (and that even if it has an intelligent source). If, on the other hand, it does match one of these patterns, then, depending on the pattern matched, the SETI researchers may have cause for celebration.

The SETI researchers in Contact did find a signal worthy of celebration, namely the following:

11011101111101111111011111111111011111111111110
1111111111111111101111111111111111111011111111
1111111111111110111111111111111111111111111110
1111111111111111111111111111110111111111111111
1111111111111111111111011111111111111111111111
1111111111111111110111111111111111111111111111
1111111111111111011111111111111111111111111111
1111111111111111110111111111111111111111111111
1111111111111111111111111101111111111111111111
1111111111111111111111111111111111111111110111
1111111111111111111111111111111111111111111111
1111111111111111110111111111111111111111111111
1111111111111111111111111111111111111111111101
1111111111111111111111111111111111111111111111
1111111111111111111111111101111111111111111111
1111111111111111111111111111111111111111111111
1111111111111101111111111111111111111111111111
1111111111111111111111111111111111111111111111
1111110111111111111111111111111111111111111111
1111111111111111111111111111111111111111111111
1111011111111111111111111111111111111111111111
1111111111111111111111111111111111111111111111
1111111111011111111111111111111111111111111111
1111111111111111111111111111111111111111111111
11111111111111111111

The SETI researchers in Contact received this signal as a sequence of 1,126 beats and pauses, where 1s correspond to beats and 0s to pauses. This sequence represents the prime numbers from 2 to 101, where a given prime number is represented by the corresponding number of beats (i.e., 1s), and the individual prime numbers are separated by pauses (i.e., 0s). The SETI researchers in Contact took this signal as decisive confirmation of an extraterrestrial intelligence.

What is it about this signal that implicates design? Whenever we infer design, we must establish three things: contingency, complexity, and specification. Contingency ensures that the object in question is not the result of an automatic and therefore unintelligent process that had no choice in its production. Complexity ensures that the object is not so simple that it can readily be explained by chance. Finally, specification ensures that the object exhibits the type of pattern characteristic of intelligence. Let us examine these three requirements more closely.

Contingency

In practice, to establish the contingency of an object, event, or structure, one must establish that it is compatible with the regularities involved in its production, but that these regularities also permit any number of alternatives to it. Typically these regularities are conceived as natural laws or algorithms. By being compatible with but not required by the regularities involved in its production, an object, event, or structure becomes irreducible to any underlying physical necessity. Michael Polanyi and Timothy Lenoir have both described this method of establishing contingency.

The method applies quite generally: the position of Scrabble pieces on a Scrabble board is irreducible to the natural laws governing the motion of Scrabble pieces; the configuration of ink on a sheet of paper is irreducible to the physics and chemistry of paper and ink; the sequencing of DNA bases is irreducible to the bonding affinities between the bases; and so on. In the case at hand, the sequence of 0s and 1s to form a sequence of prime numbers is irreducible to the laws of physics that govern the transmission of radio signals. We therefore regard the sequence as contingent.

Complexity

To see next why complexity is crucial for inferring design, consider the following sequence of bits:

110111011111

These are the first twelve bits in the previous sequence representing the prime numbers 2, 3, and 5 respectively. Now it is a sure bet that no SETI researcher, if confronted with this twelve-bit sequence, is going to contact the science editor at the New York Times, hold a press conference, and announce that an extraterrestrial intelligence has been discovered. No headline is going to read, “Aliens Master First Three Prime Numbers!”

The problem is that this sequence is much too short (and thus too simple) to establish that an extraterrestrial intelligence with knowledge of prime numbers produced it. A randomly beating radio source might by chance just happen to produce this sequence. A sequence of 1,126 bits representing the prime numbers from 2 to 101, however, is a different story. Here the sequence is sufficiently long (and therefore sufficiently complex) to allow that an extraterrestrial intelligence could have produced it.

Complexity as I am describing it here is a form of probability. (Later in this essay I will require a more general conception of complexity to unpack the logic of design inferences. But for now complexity as a form of probability is all we need.) To see the connection between complexity and probability, consider a combination lock. The more possible combinations of the lock, the more complex the mechanism and correspondingly the more improbable that the mechanism can be opened by chance. Complexity and probability therefore vary inversely: the greater the complexity, the smaller the probability. Thus to determine whether something is sufficiently complex to warrant a design inference is to determine whether it has sufficiently small probability.

Even so, complexity (or improbability) isn’t enough to eliminate chance and establish design. If I flip a coin 1,000 times, I’ll participate in a highly complex (i.e., highly improbable) event. Indeed, the sequence I end up flipping will be one in a trillion trillion trillion . . . , where the ellipsis needs 22 more “trillions.” This sequence of coin tosses won’t, however, trigger a design inference. Though complex, this sequence won’t exhibit a suitable pattern. Contrast this with the previous sequence representing the prime numbers from 2 to 101. Not only is this sequence complex, but it also embodies a suitable pattern. The SETI researcher who in the movie Contact discovered this sequence put it this way: “This isn’t noise, this has structure.”

Specification

What is a suitable pattern for inferring design? Not just any pattern will do. Some patterns can legitimately be employed to infer design whereas others cannot. The intuition underlying the distinction between patterns that alternately succeed or fail to implicate design is, however, easily motivated. Consider the case of an archer. Suppose an archer stands 50 meters from a large wall with bow and arrow in hand. The wall is sufficiently large that the archer cannot help but hit it. Now suppose each time the archer shoots an arrow at the wall, the archer paints a target around the arrow so that the arrow sits squarely in the bull’s-eye. What can be concluded from this scenario? Absolutely nothing about the archer’s ability as an archer. Yes, a pattern is being matched; but it is a pattern fixed only after the arrow has been shot. The pattern is thus purely ad hoc.

But suppose instead the archer paints a fixed target on the wall and then shoots at it. Suppose the archer shoots a hundred arrows, and each time hits a perfect bull’s-eye. What can be concluded from this second scenario? Confronted with this second scenario we are obligated to infer that here is a world-class archer, one whose shots cannot legitimately be referred to luck, but rather must be referred to the archer’s skill and mastery. Skill and mastery are of course instances of design.

The archer example introduces three elements that are essential for inferring design:

1. A reference class of possible events (here the arrow hitting the wall at some unspecified place);

2. A pattern that restricts the reference class of possible events (here a target on the wall); and

3. The precise event that has occurred (here the arrow hitting the wall at some precise location).

In a design inference, the reference class, the pattern, and the event are linked, with the pattern mediating between event and reference class, and helping to decide whether the event is due to chance or design. Note that in determining whether an event is sufficiently improbable or complex to implicate design, the relevant improbability is not that of the precise event that occurred, but that of the target/pattern. Indeed, the bigger the target, the easier it is to hit it by chance and thus apart from design.

The type of pattern in which an archer fixes a target first and then shoots at it is common to statistics, where it is known as setting a rejection region prior to an experiment. In statistics, if the outcome of an experiment falls within a rejection region, the chance hypothesis supposedly responsible for the outcome is rejected. The reason for setting a rejection region prior to an experiment is to forestall what statisticians call “data snooping” or “cherry picking.” Just about any data set will contain strange and improbable patterns if we look hard enough. By forcing experimenters to set their rejection regions prior to an experiment, the statistician protects the experiment from spurious patterns that could just as well result from chance.

Now a little reflection makes clear that a pattern need not be given prior to an event to eliminate chance and implicate design. Consider the following cipher text:

nfuijolt ju jt mjlf b xfbtfm

Initially this looks like a random sequence of letters and spaces—you lack any pattern for rejecting chance and inferring design.

But suppose that someone comes along and tells you to treat this sequence as a Caesar cipher, moving each letter one notch down the alphabet. Now the sequence reads,

methinks it is like a weasel

Even though the pattern (in this case, the decrypted text) is given after the fact, it still is the right sort of pattern for eliminating chance and inferring design. In contrast to statistics, which always identifies its patterns before an experiment is performed, cryptanalysis must discover its patterns after the fact. In both instances, however, the patterns are suitable for inferring design.

Patterns thus divide into two types: those that in the presence of complexity warrant a design inference and those that despite the presence of complexity do not warrant a design inference. The first type of pattern I call a specification, the second a fabrication. Specifications are the non- ad hoc patterns that can legitimately be used to eliminate chance and warrant a design inference. In contrast, fabrications are the ad hoc patterns that cannot legitimately be used to warrant a design inference.

To sum up, the complexity-specification criterion detects design by establishing three things: contingency, complexity, and specification. When called to explain an event, object, or structure, we have to decide: are we going to attribute it to necessity, chance, or design? According to the complexity-specification criterion, to answer this question is to answer three simpler questions: Is it contingent? Is it complex? Is it specified? Consequently, the complexity-specification criterion can be represented as a flowchart with three decision nodes. I call this flowchart the Explanatory Filter.

Independent Patterns Are Detachable

For a pattern to count as a specification, the important thing is not when it was identified, but whether in a certain well-defined sense it is independent of the event it describes. Drawing a target around an arrow already embedded in a wall is not independent of the arrow’s trajectory. Consequently, such a target/pattern cannot be used to attribute the arrow’s trajectory to design. Patterns that are specifications cannot simply be read off the events whose design is in question. Rather, to count as specifications, patterns must be suitably independent of events. I refer to this relation of independence as detachability, and say that a pattern is detachable only if it satisfies that relation.

Detachability can be understood as asking this question: Given an event (whose design is in question) and a pattern describing it, would we be able to construct that pattern if we had no knowledge of which event occurred? Assume an event has occurred. A pattern describing the event is given. The event is one from a range of possible events. If all we knew was the range of possible events without any specifics about which event actually occurred, could we still construct the pattern describing the event? If so, the pattern is detachable from the event.

A Trick with Coins

To see what’s at stake, consider the following example. (It was this example that finally clarified for me what transforms a pattern simpliciter into a pattern qua specification.) The following event E to all appearances was obtained by flipping a fair coin 100 times:

THTTTHHTHHTTTTTHTHTTHHHTTHTHHHTH
HTTTTTTTHTTHTTTHHTHTTTHTHTHHTTHH
HTTTHTTHHTHTHTHHHHTTHHTHHHHTHHH
HTT

E

Is E the product of chance or not? A standard trick of statistics professors with an introductory statistics class is to divide the class in two and have students in one half of the class each flip a coin 100 times and write down the sequence of heads and tails on a slip of paper; students in the other half each generate with their minds a “random looking” string that mimics the tossing of a coin 100 times and also write down the sequence of heads and tails on a slip of paper. When the students then hand in their slips of paper, it is the professor’s job to sort the papers into two piles, those generated by flipping a fair coin, and those concocted in the students’ heads. To the amazement of the students, the statistics professor is typically able to sort the papers with 100 percent accuracy.

There’s no mystery here. The statistics professor simply looks for a repetition of six or seven heads or tails in a row to distinguish the truly random from the pseudo-random sequences. In a hundred coin flips, one is quite likely to see such a repetition. On the other hand, people concocting pseudo-random sequences with their minds tend to alternate between heads and tails too frequently. Whereas with a truly random sequence of coin tosses there is a 50 percent chance that one toss will differ from the next, as a matter of human psychology people expect that one toss will differ from the next around 70 percent of the time.

How, then, will our statistics professor fare when confronted with E? Will she attribute E to chance or to the musings of someone trying to mimic chance? According to the professor’s crude randomness checker, E would be assigned to the pile of sequences presumed to be truly random, for E contains a repetition of seven tails in a row. Everything that at first blush would lead us to regard E as truly random checks out. There are exactly 50 alternations between heads and tails (as opposed to the 70 that would be expected from human beings trying to mimic chance). What’s more, the relative frequencies of heads and tails check out: there were 49 heads and 51 tails. Thus it’s not as though the coin supposedly responsible for generating E was heavily biased in favor of one side versus the other.

But Is It Really Chance?

Suppose, however, that our statistics professor suspects she is not up against a neophyte statistics student, but instead a fellow statistician who is trying to put one over on her. To help organize her problem, study it more carefully, and enter it into a computer, she will find it convenient to let strings of 0s and 1s represent the outcomes of coin flips, with 1 corresponding to heads and 0 to tails. In that case the following pattern D will correspond to the event E:

0100011011000001010011100101110111000000010010
0011010001010110011110001001101010111100110111
10111100

D

Now, the mere fact that the event E conforms to the pattern D is no reason to think that E did not occur by chance. As things stand, the pattern D has simply been read off the event E.

But D need not have been read off of E. Indeed, D could have been constructed without recourse to E. To see this, let us rewrite D as follows:

0
1
00
01
10
11
000
001
010
011
100
101
110
111
0000
0001
0010
0011
0100
0101
0110
0111
1000
1001
1010
1011
1100
1101
1110
1111
00

D

By viewing D this way, anyone with the least exposure to binary arithmetic immediately recognizes that D was constructed simply by writing binary numbers in ascending order, starting with the one-digit binary numbers (i.e., 0 and 1), proceeding then to the two-digit binary numbers (i.e., 00, 01, 10, and 11), and continuing on until 100 digits were recorded. It’s therefore intuitively clear that D does not describe a truly random event (i.e., an event gotten by tossing a fair coin), but rather a pseudo-random event, concocted by doing a little binary arithmetic.

Side Information Does the Trick

Although it’s now intuitively clear why chance cannot properly explain E, we need to consider more closely why this mode of explanation fails here. We started with a putative chance event E, supposedly gotten by flipping a fair coin 100 times. Since heads and tails each have probability 1/2, and since this probability gets multiplied for each flip of the coin, it follows that the probability of E is 2^–100, or approximately 10^–30.

In addition, we constructed a pattern D to which E conforms. Initially D proved insufficient to eliminate chance as the explanation of E since in its construction D was simply read off E. Rather, to eliminate chance we also had to recognize that D could have been constructed quite easily by performing some simple arithmetic operations with binary numbers. Thus to eliminate chance we needed to employ additional side information, which in this case consisted of our knowledge of binary arithmetic. This side information detached the pattern D from the event E and thereby rendered D a specification.

For side information to detach a pattern from an event, it must satisfy two conditions, conditional independence and tractability. First, the side information must be conditionally independent of the event E. Conditional independence, a well-defined notion from probability theory, means that the probability of E doesn’t change once the side information is taken into account. Conditional independence is the standard probabilistic way of unpacking epistemic independence. Two things are epistemically independent if knowledge about one thing (in this case the side information) does not affect knowledge about the other (in this case the occurrence of E). This is certainly the case here since our knowledge of binary arithmetic does not affect the probabilities we assign to coin tosses.

The second condition, the tractability condition, requires that the side information enable us to construct the pattern D to which E conforms. This is evidently the case here as well since our knowledge of binary arithmetic enables us to arrange binary numbers in ascending order, and thereby construct the pattern D.

But what exactly is this ability to construct a pattern on the basis of side information? Perhaps the most slippery words in philosophy are “can,” “able,” and “enable.” Fortunately, just as there is a precise theory for characterizing the epistemic independence between an event and side information—namely, probability theory—so too there is a precise theory for characterizing the ability to construct a pattern on the basis of side information—namely, complexity theory.

Complexity Theory

Complexity theory, conceived now quite generally and not merely as a form of probability, assesses the difficulty of tasks given the resources available for accomplishing those tasks. If I may generalize computational complexity theory, it ranks tasks according to difficulty, and then determines which tasks are sufficiently manageable to be doable or tractable. For instance, given current technology we find sending a person to the moon tractable, but sending a person to the nearest galaxy intractable.

In the tractability condition, the task to be accomplished is the construction of a pattern and the resources for accomplishing that task are side information. Thus, for the tractability condition to be satisfied, side information must provide the resources necessary for constructing the pattern in question. All of this admits a precise complexity-theoretic formulation and makes definite what I called “the ability to construct a pattern on the basis of side information.”

Taken jointly, the tractability and conditional independence conditions mean that side information enables us to construct the pattern to which an event conforms, yet without recourse to the actual event. This is the crucial insight. Because the side information is conditionally and therefore epistemically independent of the event, any pattern constructed from this side information is obtained without recourse to the event. In this way any pattern that is constructed from such side information avoids the charge of being ad hoc. These, then, are the detachable patterns. These are the specifications.

Why the Criterion Works

The complexity-specification criterion is exactly the right instrument for detecting design. To see why, we need to understand what makes intelligent agents detectable in the first place. The principal characteristic of intelligent agency is choice. Even the etymology of the word “intelligent” makes this clear. “Intelligent” derives from two Latin words, the preposition inter, meaning between, and the verb lego, meaning to choose or select. Thus, according to its etymology, intelligence consists in choosing between. For an intelligent agent to act is therefore to choose from a range of competing possibilities.

This is true not just of humans, but of animals as well as of extraterrestrial intelligences. A rat navigating a maze must choose whether to go right or left at various points in the maze. When SETI researchers attempt to discover intelligence in the extraterrestrial radio transmissions they are monitoring, they assume an extraterrestrial intelligence could have chosen any number of possible radio transmissions, and then attempt to match the transmissions they observe with certain patterns as opposed to others. Whenever a human being utters meaningful speech, a choice is made from a range of possible sound combinations that might have been uttered. Intelligent agency always entails discrimination, choosing certain things, ruling out others.

Recognizing Intelligence

Given this characterization of intelligent agency, the crucial question is how to recognize it. Intelligent agents act by making a choice. How, then, do we recognize that an intelligent agent has made a choice? A bottle of ink spills accidentally onto a sheet of paper; someone takes a fountain pen and writes a message on a sheet of paper. In both instances ink is applied to paper. In both instances one among an almost infinite set of possibilities is realized. In both instances a contingency is actualized and others are ruled out. Yet in one instance we ascribe agency, in the other chance.

What is the relevant difference? Not only do we need to observe that a contingency was actualized, but we need also to be able to give the specifications of that contingency. The contingency must conform to an independently given pattern, and we must be able independently to construct that pattern. A random inkblot is unspecified; a message written with ink on paper is specified. To be sure, the exact message recorded may not be specified. But orthographic, syntactic, and semantic constraints will nonetheless specify it.

Actualizing one among several competing possibilities, ruling out the rest, and specifying the one that was actualized encapsulates how we recognize intelligent agency, or equivalently, how we detect design. Experimental psychologists who study animal learning and behavior have known this all along. To learn a task an animal must acquire the ability to actualize behaviors suitable for the task as well as the ability to rule out behaviors unsuitable for the task. Moreover, for a psychologist to recognize that an animal has learned a task, it is necessary not only to observe the animal making the appropriate discrimination, but also to specify the discrimination.

Rats & Mazes

Thus, to recognize whether a rat has successfully learned how to traverse a maze, a psychologist must first specify which sequence of right and left turns conducts the rat out of the maze. No doubt, a rat randomly wandering a maze also discriminates a sequence of right and left turns. But by randomly wandering the maze, the rat gives no indication that it can discriminate the appropriate sequence of right and left turns for exiting the maze. Consequently, the psychologist studying the rat will have no reason to think the rat has learned how to traverse the maze.

Only if the rat executes the sequence of right and left turns specified by the psychologist will the psychologist recognize that the rat has learned how to traverse the maze. Now it is precisely the learned behaviors we regard as intelligent in animals. Hence it is no surprise that the same scheme for recognizing animal learning recurs for recognizing intelligent agency generally, to wit: actualizing one among several competing possibilities, ruling out the others, and specifying the one actualized.

Note that complexity is implicit here as well. To see this, consider again a rat traversing a maze, but now take a very simple maze in which two right turns conduct the rat out of the maze. How will a psychologist studying the rat determine whether it has learned to exit the maze? Just putting the rat in the maze will not be enough. Because the maze is so simple, the rat could by chance just happen to take two right turns, and thereby exit the maze. The psychologist will therefore be uncertain whether the rat actually learned to exit this maze or just got lucky.

But contrast this with a complicated maze in which a rat must take just the right sequence of left and right turns to exit the maze. Suppose the rat must take one hundred appropriate right and left turns, and that any mistake will prevent the rat from exiting the maze. A psychologist who sees the rat take no erroneous turns and quickly exit the maze will be convinced that the rat has indeed learned how to exit the maze, and that it was not dumb luck.

This general scheme for recognizing intelligent agency is but a thinly disguised form of the complexity-specification criterion. In general, to recognize intelligent agency we must observe an actualization of one among several competing possibilities, note which possibilities were ruled out, and then be able to specify the possibility that was actualized. What’s more, the competing possibilities that were ruled out must be live possibilities, and sufficiently numerous so that specifying the possibility that was actualized cannot be attributed to chance. In terms of complexity, this is just another way of saying that the range of possibilities is complex. In terms of probability, this is just another way of saying that the possibility that was actualized has small probability.

All the elements in this general scheme for recognizing intelligent agency (i.e., actualizing, ruling out, and specifying) find their counterpart in the complexity-specification criterion. It follows that this criterion formalizes what we have been doing right along when we recognize intelligent agency. The complexity-specification criterion pinpoints how we detect design.

Design, Metaphysics & Beyond

Where is this work on design heading? Specified complexity, that key trademark of design, is, as it turns out, a form of information (though one considerably richer than Claude Shannon’s purely statistical form of it). Although called by different names and developed with different degrees of rigor, specified complexity is starting to have an effect on the special sciences.

For instance, specified complexity is what Michael Behe has uncovered with his irreducibly complex biochemical machines, what Manfred Eigen regards as the great mystery of life’s origin, what for cosmologists underlies the fine-tuning of the universe, what David Chalmers hopes will ground a comprehensive theory of human consciousness, what enables Maxwell’s demon to outsmart a thermodynamic system tending toward thermal equilibrium, and what within the Kolmogorov-Chaitin theory of algorithmic information identifies the highly compressible, non-random strings of digits. How complex specified information gets from an organism’s environment into an organism’s genome will be one of the key questions at an upcoming Santa Fe Institute symposium, “Complexity, Information & Design: A Critical Appraisal” (October 1999).

Shannon’s purely statistical theory of information is giving way to a richer theory of complex specified information whose possibilities are only now coming to light. A natural sequel to The Design Inference is therefore to develop a general theory of complex specified information.

Yet despite its far-reaching implications for science, I regard the ultimate significance of this work on design to lie in metaphysics. In my view, design died not at the hands of nineteenth-century evolutionary biology, but at the hands of the mechanical philosophy two centuries earlier—and that despite the popularity of British natural theology at the time. Though the originators of the mechanical philosophy were typically theists, the design they retained was at best an uneasy rider on top of a mechanistic view of nature. Design is neither use nor ornament within a strictly mechanistic world of particles or other mindless entities organized by equally mindless principles of association, even if these be natural laws ordained by God.

The primary challenge, once the broader implications of design for science have been worked out, is therefore to develop a relational ontology in which the problem of being resolves thus: to be is to be in communion, and to be in communion is to transmit and receive information. Such an ontology will not only safeguard science and leave adequate breathing space for design, but will also make sense of the world as sacrament.

The world is a mirror representing the divine life. The mechanical philosophy was ever blind to this fact. Intelligent design, on the other hand, readily embraces the sacramental nature of physical reality. Indeed, intelligent design is just the Logos theology of John’s Gospel restated in the idiom of information theory.  

William A. Dembski, Ph.D. (Mathematics, University of Chicago, and Philosophy, University of Illinois at Chicago), also holds an M.Div. from Princeton Theological Seminary. He is a fellow of the Discovery Institute’s Center for the Renewal of Science and Culture, the author of The Design Inference (Cambridge University Press), and editor of Mere Creation (InterVarsity). A new book, Intelligent Design: The Bridge Between Science and Theology, is due out shortly from InterVarsity.

A Journal of Mere Christianity—Delivered to Your Door

• Essays on theology, culture, and the Church
• Contributors from across the Christian traditions

Subscribe (Print + Online)

Six print issues (one year) of Touchstone, plus full online access and PDF downloads for only $39.95.

Subscribe (Online Only)

Get a one-year full-access subscription to the Touchstone online archives for only $19.95.

bulk subscriptions

Order Touchstone subscriptions in bulk and save $10 per sub! Each subscription includes 6 issues of Touchstone plus full online access to touchstonemag.com—including archives, videos, and pdf downloads of recent issues for only $29.95 each! Great for churches or study groups.

Transactions will be processed on a secure server.

more on evolution from the online archives

22.6—July/August 2009

Samurai Bioethics

on a Noble Defense Doomed by Darwinian Materialism by John G. West

32.2—March/April 2019

The Aura of Science

Flight from the Absolute: Cynical Observations on the Postmodern West, Volume II by Paul Gosselin by Louis Markos

22.8—November/December 2009

The Origin of Aesthetics

Looking for Beauty & Justice 150 Years After Darwin’s Classic by Charles Taliaferro

more from the online archives

30.2—March/April 2017

Rescuing Cervantes

on Reading Don Quixote in Its Original Christian Context by Luis Cortest

36.4—Jul/Aug 2023

Nothingness Rules

Our Political Void & the Disintegration of Truth by Michael Hanby

31.5—September/October 2018

Liberalism Occupied

The Rise of the Gnostic Liberal State After Christianity by Andrew Latham

calling all readers

Please Donate

"There are magazines worth reading but few worth saving . . . Touchstone is just such a magazine."
—Alice von Hildebrand

"Here we do not concede one square millimeter of territory to falsehood, folly, contemporary sentimentality, or fashion. We speak the truth, and let God be our judge. . . . Touchstone is the one committedly Christian conservative journal."
—Anthony Esolen, Touchstone senior editor

Support Touchstone

00