Introduction to No Free Lunch

William A. Dembski


1 The Third Mode of Explanation
1.1 Necessity, Chance, and Design
1.2 Rehabilitating Design
1.3 The Complexity-Specification Criterion
1.4 Specification
1.5 Probabilistic Resources
1.6 False Negatives and False Positives
1.7 Why the Criterion Works
1.8 The Darwinian Challenge to Design
1.9 The Constraining of Contingency
1.10 The Darwinian Extrapolation
2 Another Way to Detect Design?
2.1 Fisher's Approach to Eliminating Chance
2.2 Generalizing Fisher's Approach
2.3 Case Study: Nicholas Caputo
2.4 Case Study: The Compressibility of Bit Strings
2.5 Detachability
2.6 Sweeping the Field of Chance Hypotheses
2.7 Justifying the Generalization
2.8 The Inflation of Probabilistic Resources
2.9 Design by Comparison
2.10 Design by Elimination
3 Specified Complexity as Information
3.1 Information
3.2 Syntactic, Statistical, and Algorithmic Information
3.3 Information in Context
3.4 Conceptual and Physical Information
3.5 Complex Specified Information
3.6 Semantic Information
3.7 Biological Information
3.8 The Origin of Complex Specified Information
3.9 The Law of Conservation of Information
3.10 A Fourth Law of Thermodynamics?
4 Evolutionary Algorithms
4.2 Optimization
4.3 Statement of the Problem
4.4 Choosing the Right Fitness Function
4.5 Blind Search
4.6 The No Free Lunch Theorems
4.7 The Displacement Problem
4.8 Darwinian Evolution in Nature
4.9 Following the Information Trail
4.10 Coevolving Fitness Landscapes
5 The Emergence of Irreducibly Complex Systems
5.1 The Causal Specificity Problem
5.2 The Challenge of Irreducible Complexity
5.3 Scaffolding and Roman Arches
5.4 Co-optation, Patchwork, and Bricolage
5.5 Incremental Indispensability
5.6 Reducible Complexity
5.7 Miscellaneous Objections
5.8 The Logic of Invariants
5.9 Fine-Tuning Irreducible Complexity
5.10 Doing the Calculation
6 Design as a Scientific Research Program
6.1 Outline of a Positive Research Program
6.2 The Pattern of Evolution
6.3 The Incompleteness of Natural Laws
6.4 Does Specified Complexity Have a Mechanism?
6.5 The Nature of Nature
6.6 Must All Design in Nature Be Front-Loaded?
6.7 Embodied and Unembodied Designers
6.8 Who Designed the Designer?
6.9 Testability
6.10 Magic, Mechanism, and Design


How a designer gets from thought to thing is, at least in broad strokes, straightforward: (1) A designer conceives a purpose. (2) To accomplish that purpose, the designer forms a plan. (3) To execute the plan, the designer specifies building materials and assembly instructions. (4) Finally, the designer or some surrogate applies the assembly instructions to the building materials. What emerges is a designed object, and the designer is successful to the degree that the object fulfills the designer's purpose. In the case of human designers, this four-part design process is uncontroversial. Baking a cake, driving a car, embezzling funds, and building a supercomputer each presuppose it. Not only do we repeatedly engage in this four-part design process, but we have witnessed other people engage in it countless times. Given a sufficiently detailed causal history, we are able to track this process from start to finish.

But suppose a detailed causal history is lacking and we are not able to track the design process. Suppose instead that all we have is an object, and we must decide whether it emerged from such a design process. In that case, how do we decide whether the object is in fact designed? If the object in question is sufficiently like other objects that we know were designed, then there may be no difficulty inferring design. For instance, if we find a scrap of paper with writing on it, we infer a human author even if we know nothing about the paper's causal history. We are all familiar with humans writing on scraps of paper, and there is no reason to suppose that this scrap of paper requires a different type of causal story.

Nevertheless, when it comes to living things, the biological community holds that a very different type of causal story is required. To be sure, the biological community admits that biological systems appear to be designed. For instance, Richard Dawkins writes, "Biology is the study of complicated things that give the appearance of having been designed for a purpose." Likewise, Francis Crick writes, "Biologists must constantly keep in mind that what they see was not designed, but rather evolved." Or consider the title of Renato Dulbecco's biology text: The Design of Life. The term "design" is everywhere in the biological literature. Even so, its use is carefully regulated. According to the biological community, the appearance of design in biology is misleading. This is not to deny that biology is filled with marvelous contrivances. Biologists readily admit as much. Yet as far as the biological community is concerned, living things are not the result of the four-part design process described above.

But how does the biological community know that living things are only apparently and not actually designed? According to Francisco Ayala, Charles Darwin provided the answer: "The functional design of organisms and their features would therefore seem to argue for the existence of a designer. It was Darwin's greatest accomplishment to show that the directive organization of living beings can be explained as the result of a natural process, natural selection, without any need to resort to a Creator or other external agent. The origin and adaptation of organisms in their profusion and wondrous variations were thus brought into the realm of science." Is it really the case, however, that the directive organization of living beings can be explained without recourse to a designer? And would employing a designer in biological explanations necessarily take us out of the realm of science? The purpose of this book is to answer these two questions.

The title of this book, No Free Lunch, refers to a collection of mathematical theorems proved in the past five years about evolutionary algorithms. The upshot of these theorems is that evolutionary algorithms, far from being universal problem solvers, are in fact quite limited problem solvers that depend crucially on additional information not inherent in the algorithms before they are able to solve any interesting problems. This additional information needs to be carefully specified and fine-tuned, and such specification and fine-tuning is always thoroughly teleological. Consequently, evolutionary algorithms are incapable of providing a computational justification for the Darwinian mechanism of natural selection and random variation as the primary creative force in biology. The subtitle, Why Specified Complexity Cannot Be Purchased without Intelligence, refers to that form of information, known as specified complexity or complex specified information, that is increasingly coming to be regarded as a reliable empirical marker of purpose, intelligence, and design.

What is specified complexity? An object, event, or structure exhibits specified complexity if it is both complex (i.e., one of many live possibilities) and specified (i.e., displays an independently given pattern). A long sequence of randomly strewn Scrabble pieces is complex without being specified. A short sequence spelling the word "the" is specified without being complex. A sequence corresponding to a Shakespearean sonnet is both complex and specified. In The Design Inference: Eliminating Chance through Small Probabilities, I argued that specified complexity is a reliable empirical marker of intelligence. Nevertheless, critics of my argument have claimed that evolutionary algorithms, and the Darwinian mechanism in particular, can deliver specified complexity apart from intelligence. I anticipated this criticism in The Design Inference but did not address it there in detail. Filling in the details is the task of the present volume.

The Design Inference laid the groundwork. This book demonstrates the inadequacy of the Darwinian mechanism to generate specified complexity. Darwinists themselves have made possible such a refutation. By assimilating the Darwinian mechanism to evolutionary algorithms, they have invited a mathematical assessment of the power of the Darwinian mechanism to generate life's diversity. Such an assessment, begun with the No Free Lunch theorems of David Wolpert and William Macready (see section 4.6), will in this book be taken to its logical conclusion. The conclusion is that Darwinian mechanisms of any kind, whether in nature or in silico, are in principle incapable of generating specified complexity. Coupled with the growing evidence in cosmology and biology that nature is chock-full of specified complexity (cf. the fine-tuning of cosmological constants and the irreducible complexity of biochemical systems), this conclusion implies that naturalistic explanations are incomplete and that design constitutes a legitimate and fundamental mode of scientific explanation.

In arguing that naturalistic explanations are incomplete or, equivalently, that natural causes cannot account for all the features of the natural world, I am placing natural causes in contradistinction to intelligent causes. The scientific community has itself drawn this distinction in its use of these twin categories of causation. Thus, in the quote earlier by Francisco Ayala, "Darwin's greatest accomplishment [was] to show that the directive organization of living beings can be explained as the result of a natural process, natural selection, without any need to resort to a Creator or other external agent." Natural causes, as the scientific community understands them, are causes that operate according to deterministic and nondeterministic laws and that can be characterized in terms of chance, necessity, or their combination (cf. Jacques Monod's Chance and Necessity). To be sure, if one is more liberal about what one means by natural causes and includes among natural causes telic processes that are not reducible to chance and necessity (as the ancient Stoics did by endowing nature with immanent teleology), then my claim that natural causes are incomplete dissolves. But that is not how the scientific community by and large understands natural causes.

The distinction between natural and intelligent causes now raises an interesting question when it comes to embodied intelligences like ourselves, who are at once physical systems and intelligent agents: Are embodied intelligences natural causes? Even if the actions of an embodied intelligence proceed solely by natural causes, being determined entirely by the constitution and dynamics of the physical system that embodies it, that does not mean the origin of that system can be explained by reference solely to natural causes. Such systems could exhibit derived intentionality in which the underlying source of intentionality is irreducible to natural causes (cf. a digital computer). I will argue that intelligent agency, even when conditioned by a physical system that embodies it, cannot be reduced to natural causes without remainder. Moreover, I will argue that specified complexity is precisely the remainder that remains unaccounted for. Indeed, I will argue that the defining feature of intelligent causes is their ability to create novel information and, in particular, specified complexity.

Design has had a turbulent intellectual history. The chief difficulty with design to date has consisted in discovering a conceptually powerful formulation of it that will fruitfully advance science. While I fully grant that the history of design arguments warrants misgivings, they do not apply to the present project. The theory of design I envision is not an atavistic return to the design arguments of William Paley and the Bridgewater Treatises. William Paley was in no position to formulate the conceptual framework for design that I will be developing in this book. This new framework depends on advances in probability theory, computer science, the concept of information, molecular biology, and the philosophy of science-to name but a few. Within this framework design promises to become an effective conceptual tool for investigating and understanding the world.

Increased philosophical and scientific sophistication, however, is not alone in separating my approach to design from Paley's. Paley's approach was closely linked to his prior religious and metaphysical commitments. Mine is not. Paley's designer was nothing short of the triune God of Christianity, a transcendent, personal, moral being with all the perfections commonly attributed to this God. On the other hand, the designer that emerges from a theory of intelligent design is an intelligence capable of originating the complexity and specificity that we find throughout the cosmos and especially in biological systems. Persons with theological commitments can co-opt this designer and identify this designer with the object of their worship. But this move is strictly optional as far as the actual science of intelligent design is concerned.

The crucial question for science is whether design helps us understand the world, and especially the biological world, better than we do now when we systematically eschew teleological notions from our scientific theorizing. Thus, a scientist may view design and its appeal to a designer as simply a fruitful device for understanding the world, not attaching any significance to questions such as whether a theory of design is in some ultimate sense true or whether the designer actually exists. Philosophers of science would call this a constructive empiricist approach to design. Scientists in the business of manufacturing theoretical entities like quarks, strings, and cold dark matter could therefore view the designer as just one more theoretical entity to be added to the list. I follow here Ludwig Wittgenstein, who wrote, "What a Copernicus or a Darwin really achieved was not the discovery of a true theory but of a fertile new point of view." If design cannot be made into a fertile new point of view that inspires exciting new areas of scientific investigation, then it deserves to wither and die. Yet before that happens, it deserves a fair chance to succeed.

One of my main motivations in writing this book is to free science from arbitrary constraints that, in my view, stifle inquiry, undermine education, turn scientists into a secular priesthood, and in the end prevent intelligent design from receiving a fair hearing. The subtitle of Richard Dawkins's The Blind Watchmaker reads Why the Evidence of Evolution Reveals a Universe without Design. Dawkins may be right that design is absent from the universe. But science needs to address not only the evidence that reveals the universe to be without design but also the evidence that reveals the universe to be with design. Evidence is a two-edged sword: claims capable of being refuted by evidence are also capable of being supported by evidence. Even if design ends up being rejected as an unfruitful explanatory tool for science, such a negative outcome for design needs to result from the evidence for and against design being fairly considered. Darwin himself would have agreed: "A fair result can be obtained only by fully stating and balancing the facts and arguments on both sides of each question." Consequently, any rejection of design must not result from imposing arbitrary constraints on science that rule out design prior to any consideration of evidence.

Two main constraints have historically been used to keep design outside the natural sciences: methodological naturalism and dysteleology. According to methodological naturalism, in explaining any natural phenomenon, the natural sciences are properly permitted to invoke only natural causes to the exclusion of intelligent causes. On the other hand, dysteleology refers to inferior design-typically design that is either evil or incompetent. Dysteleology rules out design from the natural sciences on account of the inferior design that nature is said to exhibit. In this book, I will address methodological naturalism. Methodological naturalism is a regulative principle that purports to keep science on the straight and narrow by limiting science to natural causes. I intend to show that it does nothing of the sort but instead constitutes a straitjacket that actively impedes the progress of science.

On the other hand, I will not have anything to say about dysteleology. Dysteleology might present a problem if all design in nature were wicked or incompetent and continually flouted our moral and aesthetic yardsticks. But that is not the case. To be sure, there are microbes that seem designed to do a number on the mammalian nervous system and biological structures that look cobbled together by a long trial-and-error evolutionary process. But there are also biological examples of nano-engineering that surpass anything human engineers have concocted or entertain hopes of concocting. Dysteleology is primarily a theological problem. To exclude design from biology simply because not all examples of biological design live up to our expectations of what a designer should or should not have done is an evasion. The problem of design in biology is real and pervasive, and needs to be addressed head on and not sidestepped because our presuppositions about design happen to rule out imperfect design. Nature is a mixed bag. It is not William Paley's happy world of everything in delicate harmony and balance. It is not the widely caricatured Darwinian world of nature red in tooth and claw. Nature contains evil design, jerry-built design, and exquisite design. Science needs to come to terms with design as such and not dismiss it in the name of dysteleology.

A possible terminological confusion over the phrase "intelligent design" needs to be cleared up. The confusion centers on what the adjective "intelligent" is doing in the phrase "intelligent design." "Intelligent" can mean nothing more than being the result of an intelligent agent, even one who acts stupidly. On the other hand, it can mean that an intelligent agent acted with consummate skill and mastery. Critics of intelligent design often understand the "intelligent" in intelligent design in the latter sense and thus presume that intelligent design must entail optimal design. The intelligent design community, on the other hand, understands the "intelligent" in intelligent design simply to refer to intelligent agency (irrespective of skill, mastery, or cleverness) and thus separates intelligent design from optimality of design. But why then place the adjective intelligent in front of the noun design? Does not design already include the idea of intelligent agency, so that juxtaposing the two becomes redundant? Redundancy is avoided because intelligent design needs also to be distinguished from apparent design. Because design in biology so often connotes apparent design, putting intelligent in front of design ensures that the design we are talking about is not merely apparent but also actual. Whether that intelligence acts cleverly or stupidly, wisely or unwisely, optimally or suboptimally are separate questions.

Who will want to read No Free Lunch? The audience includes anyone interested in seriously exploring the scope and validity of Darwinism as well as in learning how the emerging theory of intelligent design promises to supersede it. Napoleon III remarked that one never destroys a thing until one has replaced it. Similarly, Thomas Kuhn, in the language of paradigms and paradigm shifts, claimed that for a paradigm to shift, there has to be a new paradigm in place ready to be shifted into. Throughout my work, I have not been content merely to critique existing theory but have instead striven to provide a positive more-encompassing framework within which to reconceptualize phenomena inadequately explained by existing theory. Much of No Free Lunch will be accessible to an educated lay audience. Many of the ideas have been presented in published articles and public lectures. I have seen how the ideas in this book have played themselves out under fire. The chapters are therefore tailored to questions people are actually asking. The virtue of this book is filling in the details. And the devil is in the details.

I have tried to keep technical discussions to a minimum. I am no fan of notation-heavy prose and avoid it whenever possible. A book of this sort, however, poses a peculiar challenge. Forms of thinking that turn biological complexity into a free lunch pervade science and are deeply entrenched. It does no good therefore to speak in generalities or point to certain obvious tensions (e.g., how can intelligence arise out of an inherently unintelligent Darwinian process? or how can we have any confidence in the reliability of our cognitive faculties if we are the result of a brute natural process for which survival and reproduction is everything and truth-seeking is incidental?). Make the book too obvious, and no one will pay it any mind. Make it too technical, and no one will read it. My strategy in writing this book, therefore, has been to include just enough technical discussion so that experts can fill in the details as well as sufficient elaboration of the technical discussion so that nonexperts feel the force of the design inference. Whether I have been successful is for others to judge.

No Free Lunch has the following logical structure. Chapter 1 presents a nontechnical summary of my work on inferring design and makes the connection between my previous work and Darwinism explicit. Chapter 2 rebuts critics who argue that specified complexity is not a well-defined concept and cannot form the basis for a compelling design inference. In particular, I offer there a simplified account of specification. Chapter 3 translates the design-inferential framework of chapters 1 and 2 into a more powerful information-theoretic framework. Chapter 4 shows how this information-theoretic approach to design withstands and then overturns the challenge of evolutionary algorithms. In particular, I show that evolutionary algorithms cannot generate specified complexity. Chapter 5 then shows how the theoretical apparatus developed in the previous chapters can be applied to actual biological systems. Finally, chapter 6 examines what intelligent design means for science.

What follows is a chapter by chapter summary of the book.

Chapter 1: The Third Mode of Explanation.

How is design empirically detectable and thus distinguishable from the two generally accepted modes of scientific explanation, chance and necessity? To detect design, two features must be present: complexity and specification. Complexity guarantees that the object in question is not so simple that it can readily be attributed to chance. Specification guarantees that the object exhibits the right sort of pattern associated with intelligent causes. Specified complexity thus becomes a criterion for detecting design empirically. Having proposed a theoretical apparatus for detecting design, I next consider the challenge that Darwin posed to design historically and indicate why his challenge is viewed among many scientists as counting decisively against design. Essentially, Darwin opposed to design the joint action of chance and necessity and therewith promised to explain the complex ordered structures in biology that prior to him were attributed to design.

Chapter 2: Another Way to Detect Design?

Many in the scientific and philosophical community have staked their hopes on explaining specified complexity by means of evolutionary algorithms. Yet even without evolutionary algorithms to explain specified complexity, few are prepared to embrace design. One approach, now increasingly championed by the philosopher of science Elliott Sober, is to attack specified complexity head-on and claim that it is a spurious concept, incapable of rendering design testable in the case of natural objects, and that a precise probabilistic and complexity-theoretic analysis of specified complexity vitiates the concept entirely. In critiquing my approach to detecting design, Sober has tied himself to a likelihood framework for probability that is itself highly problematic. This chapter demonstrates that specified complexity is a well-defined concept and that it readily withstands the criticisms raised by Sober and his colleagues.

Chapter 3: Specified Complexity as Information.

Intelligent design can be formulated as a theory of information. Within such a theory, specified complexity becomes a form of information that reliably signals design. As a form of information specified complexity also becomes a proper object for scientific investigation. This chapter takes the ideas of chapters 1 and 2 and translates them into an information-theoretic framework. This reframing of intelligent design within information theory powerfully extends the design-inferential framework developed in chapter 1 and makes it possible accurately to assess the power (or lack thereof) of the Darwinian mechanism. The upshot of this chapter is a conservation law governing the origin and flow of information. From this law it follows that specified complexity is not reducible to natural causes and that the origin of specified complexity is best sought in intelligent causes. Intelligent design thereby becomes a theory for detecting and measuring information, explaining its origin, and tracing its flow.

Chapter 4: Evolutionary Algorithms.

This chapter is the climax of the book. Here I examine evolutionary algorithms, which constitute the mathematical underpinnings of Darwinism. I show that evolutionary algorithms are in principle incapable of generating specified complexity. Whereas this result follows immediately from the conservation of information law in chapter 3, this law involves a high level of abstraction, so that simply applying the law does not make clear just how limited evolutionary algorithms really are. In this chapter I therefore examine the nuts and bolts of the evolutionary algorithms: phase spaces, fitness landscapes, and optimization algorithms. An elementary combinatorial analysis shows that evolutionary algorithms can no more generate specified complexity than can five letters fill ten mailboxes.

Chapter 5: The Emergence of Irreducibly Complex Systems.

Specified complexity as a reliable empirical marker of intelligence is all fine and well, but if there are no complex specified systems in nature, what then? The previous chapters establish that specified complexity reliably signals design, not that specified complexity is actualized in any concrete physical system. This chapter examines how we determine whether a physical system exhibits specified complexity. The key to this determination, at least in biology, is Michael Behe's notion of irreducible complexity. Irreducibly complex biological systems exhibit specified complexity. Irreducible complexity is therefore a special case of specified complexity. Because specified complexity is a probabilistic notion, determining whether a physical system exhibits specified complexity requires being able to calculate probabilities. One of the objections against intelligent design becoming a viable scientific research program is that one cannot calculate the probabilities needed to confirm specified complexity for actual systems in nature. This chapter shows that even though precise calculations may not always be possible, setting bounds for the relevant probabilities is possible, and that this is adequate for establishing specified complexity in practice.

Chapter 6: Design as a Scientific Research Program.

Having shown that specified complexity is a reliable empirical marker of intelligence and having overturned the main scientific objections raised against it, I conclude this book by examining what science will look like once design is readmitted to full scientific status. The worry is that attributing design to natural systems will stultify science in the sense that once a scientist concedes that some natural system is designed, all the scientist's work is over. But this is not the case. Design raises a host of novel and interesting research questions that it does not make sense to ask within a strictly Darwinian or naturalistic framework. One such question is teasing apart the effects of natural and intelligent causation. For instance, a rusted old Cadillac is clearly designed but also shows the effects of natural causes (i.e., weathering). Intelligent design is capable of accommodating the legitimate insights of Darwinian theory. In particular, intelligent design admits a place for the Darwinian mechanism of natural selection and random variation. But as a framework for doing science, intelligent design offers additional tools for investigating nature that render it conceptually more powerful than Darwinism.

Ideally, this book should be read from start to finish. Nevertheless, because this is not always possible, let me offer the following suggestions for reading the book. Chapter 1 is the most accessible chapter in the book and is prerequisite for everything that follows. This material needs to be under the reader's belt. Sections 1.1 to 1.7 present a nontechnical summary of my previous work on inferring design, and readers familiar with it can skip these sections without loss. On the other hand, sections 1.8 to 1.10 are new and make explicit the connection between my previous work and Darwinism. Readers definitely need to read these sections. Chapter 2 is primarily directed at critics. This is the most technical chapter, and readers persuaded by my previous work may want to skip it on their initial reading. Chapters 3 and 4 translate the design-theoretic framework of chapters 1 and 2 into an information-theoretic framework. Chapter 3 presents the general theory whereas chapter 4 looks specifically at evolutionary algorithms. For nontechnical readers, I recommend a light perusal of chapter 3 and then a careful examination of chapter 4. Chapter 5 brings theory in contact with biological reality. This is where most of the current controversy lies, and readers will not want to miss this chapter. Chapter 6, on the other hand, looks at the broader implications of intelligent design for science and can be read at leisure.

There are nontechnical readers who can comfortably wade past technical mathematical discussions without being intimidated; and then there are math phobics whose eyes glaze over and brains shut down at the sight of technical mathematical discussions. This book can also be read with profit by math phobics. I suggest reading sections 1.1-1.10, 5.1-5.7, 5.9, and 6.1-6.10 in order. The only thing one needs to know about mathematics to read these sections is that powers of ten count the number of zeroes following a one. Thus 103 is 1,000 (a thousand has three zeroes after the initial one), 106 is 1,000,000 (a million has six zeroes after the initial one), etc. Reading these sections will provide a good overview of the current debate regarding intelligent design, particularly as it relates to Michael Behe's work on irreducibly complex molecular machines. Math phobics who then want to see why evolutionary algorithms cannot do the design work that Darwinists regularly attribute to these algorithms can read sections 4.1-4.2 and 4.7-4.9.

One final caution: Even though much in this book will look familiar to readers acquainted with my previous work, this familiarity can be deceiving. I have already noted that sections 1.1 to 1.7 present a nontechnical summary of my work on inferring design and that readers familiar with it can skip these sections without loss. But other sections, though apparently covering old ground, in fact differ markedly from previous work. For instance, two of my running examples in The Design Inference were the Caputo case (an instance of apparent ballot-line fraud) and algorithmic information theory. The case studies in sections 2.3 and 2.4 re-examine these examples in light of criticisms brought against them. Except for chapter 1, arguments and topics revisited are in almost every instance reworked or beefed up.

William A. Dembski
Baylor University
Waco, Texas