Origins & Design 19:2
Issue 37

Design in Nature and the Nature of Design

Dennis L. Feucht
Innovatia Laboratories
14554 Maplewood Road
Townville, Pennsylvania 16360

The science of design has been maturing for some time. It is engineering. A logical first step toward addressing the question of design of life in nature is to begin with what is known about the nature of design from engineering. This paper surveys the relevance of functional theories to design of living organisms, and their relationship to chance, Darwinism and the engineering of robotic systems.

As biologists and other scientists debate about design in nature, the field for which design is the distinctive subject-matter remains largely ignored.1 Few biologists have extensive experience designing anything intended to achieve a specified function subject to physical constraints. Consequently, arguments for or against design in nature and in particular, the design of living organisms are often made without recourse to the extensive body of insights into the nature of design found in various fields of engineering. Engineering applies constraints which are largely physical and economic to problem-solving. While engineering design is constrained by scientific principles, the achievement of a desired function in a device or process follows from a distinct knowledge of design not found in any of the physical or life sciences.

Design is avoided in science for historical reasons. Early scientists recognized Aristotle’s fallacy of attributing motives to nature. Ancient and medieval teleology was rejected largely because explanations of ultimate purpose are insufficient to account for how an object behaves or what it is. Scientific tradition has consequently avoided attribution of purpose to natural phenomena, including biological systems. Scientific research is largely limited to theories of structure and causes of behavior. These kinds of theories explain what a thing is and how it behaves but not what it is for.

Engineers, in contrast, design devices or processes to achieve a function they have in mind. But nature has no mind and therefore could have no purpose for itself. Aristotle’s direct attribution of mind to nature reflected the ancient pagan view that various regions of nature have personal attributes embodying fickle and unpredictable gods. From this perspective, it is unreasonable to expect discovery of rational order in nature, but the subsequent effort has nevertheless been remarkably successful. Because the only other place we find such adherence to rational order is in human minds, it is germane to explore the extent to which nature itself is the product of a rational mind. This premise, in contrast to paganism, is consistent with the biblical view of the early scientists: that nature was the result of an intelligent creator. The question of interest is whether objective means can be established by which to recognize design beyond artifacts of known human design.

A general approach to the unknown is to begin with the known designs and what is known about design from engineering. Once a formal basis for what constitutes design is established, extrapolation to nature can be attempted. To seek design where it is in question first requires a clear understanding of what is being sought. This understanding centers on theories of function.

Theories of Function

Engineering knowledge consists of how to design devices and processes which rearrange and transform matter, convert energy and restructure the environment to solve human problems. We are now in an era where much engineering development involves communication and transformation of information. While all branches of engineering participate in this aspect of design, electronics, computer and especially software engineering (often called “computer science”) are heavily oriented toward it.

In recent years, research into the representation of physical domains from a computational perspective 3,4 has led to three kinds of descriptions that are essential for capturing expert knowledge about physical systems:

Functional descriptions are required to understand devices: objects intended for a purpose. For example, consider electronic devices. Electronic circuit simulators can predict circuit behavior given its structure (represented as a schematic diagram or netlist). The structure is defined by the components of the device and how they are interconnected. Electrical circuit laws describe constraints on the possible behaviors of a given circuit structure. But constraints on behavior alone fail to offer the additional insight needed to know which of the many possible structures will behave as desired. Furthermore, an understanding of how a component of the system contributes to overall function is not made explicit by its behavioral constraints. An additional set of constraints related to achieving a goal are essential to understanding the functional aspects of a system.

The idea of design requires definition. Based on work in qualitative understanding of physical systems from a computational perspective, design is related to theories of function and has three essential aspects:

Solving a problem or achieving a purpose can be represented as finding a path to a goal. The path consists of a sequence of steps, where at each step a decision among alternative choices must be made. The plan is the sequence of decisions that leads from the starting state to the goal state. It is realized in the given domain, its “problem-space,” which is given by the scientific (structural and behavioral) theories that constrain the realization of the goal. The search for a plan to reach the goal applies these theories as constraints on the possible alternatives at each step. Additionally, a theory of function is applied to realize the plan. It guides the search through the usually vast possibilities of problem-space. The better the functional theory, the more optimal the search for the path to the goal and the more easily the generation of the plan. A theory of function also relates properties of plans to the kinds of goals they achieve. It reveals patterns in problem-space and ultimately makes explicit the nature of the search for a solution path.

Functional Versus High-Level Behavioral Descriptions

In some fields, such as thermodynamics, a detailed behavioral (causal) explanation may be unobtainable, even while the essential insights are revealed in a more general or higher-level (systemic) theory. Statistics are used because the behavior is too complicated; the interactions of components of the system (gas molecules) are not able to be decomposed into a hierarchy of levels of explanation with well-defined and manageable interactions between them. (In other words, they are not modular.) The underlying mechanisms are well identified: structural (atomic) and causal (kinetic) theories provide low-level explanations, yet an abstraction hierarchy of increasingly general levels of explanation cannot be built because the interaction between levels is not known. This limitation is one of being unable to manage the complexity of the behavioral theory. This problem is distinct from the need for a functional theory and the two can easily be confused.

High-level behavioral descriptions are not functional descriptions. For example, a functional description of a rocket engine would take into account additional constraints necessary in achieving its goal: to provide thrust for the rocket. The detailed functional description is the specification of the device. The design of the engine is the plan to achieve the goal, and it will include necessary limitations not found in the thermodynamics, such as chamber temperature limits (to keep it from melting and failing to achieve the goal due to structural failure). Fault conditions for the engine are nowhere found in its behavioral (thermodynamic) description, yet for a successful design, the observed behavior would be found to be constrained by additional limitations imposed on the engine according to its specification (functional definition). These additional constraints for achieving the goal are not found at any conceptual level of the behavioral description, but are necessary for the functional description.

Design consists of ...

Design Subsumes Chance

Functional descriptions are not in the same category as descriptions based on chance, for chance is neither a causal nor functional description of anything. It is an epistemological claim, a claim of ignorance of mechanisms, and as such does not tell how nature is constrained by structure or behavior.

Collective or systemic properties can be expressed through statistics but fail to explain them. Causality operates in physical theories as a kind of necessity 5 and constrains the possible behaviors of a given structure; chance does not. For a random event, its possible outcomes are equally likely; a priori information that would favor one or more outcomes over the others is lacking. Though the outcome is unknown, it is assumed that the mechanism producing it is causal. For example, thermodynamics employs statistical quantities like pressure, and though something about causes is expressed in the kinetic theory of gases, an understanding of how N gas molecules interact eludes scientific explanation at this time. Causal behavior underlies the observed data and makes the statistical results possible, though its theory is as yet unknown. Statistics can reveal patterns in nature but cannot produce the logical relationships required of a rational theory. Correlations are not causes.

Chance offers an invitation to provide a superior rational theory that accounts for life’s development, and from which causal explanations can be given and predictions made. Functional theories displace chance, but obviate accident. Consequently, because Darwinian theory is based on chance, it is an invitation to discover its underlying rational basis, if it is there at all. In spite of the common accidentalist interpretation of Darwinian theory, Darwinian biologists use functional descriptions extensively, usually in the form of plan fragments, to give plausibility to incremental structural development. If successful, a more comprehensive functional account of life would not necessarily invalidate Darwinism, but would merely deprive it of intellectual value, as heliocentric theory did to the Ptolemaic view. The difficulty in offering complete plans for large structures is paralleled by the difficulty in robotics of complete scene recognition or computer-based diagnosis; the domain requires development of functional theories that explain more.

Figure 2
Figure 2. Structural hierarchy of an automobile--except for the expansion under Battery. These are not elements of structure, but are properties associated with the battery useful for automobile repair. They are more closely related to function than structure.

Unlike chance, accident is an alternative to design. From a wider perspective, accidentalist interpretations of Darwinism are the equivalent of God-of-the-gaps explanations for divine creation in that both fill gaps in scientific understanding with extrapolation based on worldview assumptions, whether theistic or materialistic. It is often not realized that when creative and even personal attributes are attributed to “chance,” this wider meaning relates more closely to the pagan gods of nature than the Law of Large Numbers or Bayes’ Rule. When Chance is invoked to declare that there is no ultimate purpose to nature, is the purpose of Chance in making such a declaration included? The pagan gods were annoyingly irrational, but Chance is destructively self-circular.

Search for Design in Nature

Functional descriptions are expressed in terms of behavior (just as behavior is described in terms of structure) but they impose an additional set of constraints about plans and goals. Rocket engine behavior can be explained in terms of thermodynamics, materials and fluid mechanics, yet its function is not made explicit by such explanations because they only account for what happens and reveal no hint of a priori constraints selecting such behavior.

The given definition of design provides a starting-point for hypothesizing designs in nature, and in particular the design of living organisms. Biological systems show systemic properties similar to artificially designed systems. The first item of the definition is already provided by the existing life sciences. Items 2 and 3 require functional theories. Of the possibilities allowed under 1, actual organisms conform to additional (functional) principles which entail 2 and 3.

Development of functional theories is an act of construction, of finding ways of expressing behavior and structure so that a purposive intent can be recognized. For example, in electronics, algebraic equations describing properties of circuits can be cast in multiple forms. Some of these forms (usually one) can be readily interpreted according to some higher-level theory that reveals the function of the circuit. Without such interpretive structure, the equations appear to have no additional meaning. For biology, casting the expression of behavior in a form that reveals functional properties is the essential challenge. Consequently, this might first require that the hierarchy of behavioral descriptions be reworked into novel forms.

Figure 2
Figure 3. Physical interpretation of algebraic expression for electronic circuit requires that it be written in a physically meaningful form.

Functional descriptions of biological systems are specifications that select the existing behavior and structure of actual organisms.6 If the functional theory producing such specifications imposes necessary constraints upon behavior and these behavioral predictions match what is observed, then such a functional theory predicts what is observed according to logical necessity, as is required of scientific theories.

Design techniques in engineering are usually only sufficient to produce a device and cannot be shown to be necessary; that is, to be the only ways to produce it (though they may be the only known ways). Physical causes are necessary for observed effects and the rational structure of causal theories manifest a corresponding logical necessity. Yet causal theories cannot be a sufficient explanation in that unknown causes not anticipated by the theory might also produce the given effect. The more comprehensive the theory the greater the fraction of all possible causes it accounts for. (Relatively comprehensive theories are usually raised to the status of scientific “law.”) Similarly, a comprehensive functional theory accounts for all possible plans and their characteristics. By enumerating them, it predicts the possible ways the goal can be achieved.

Some plans are more optimal than others. If a functional theory is sufficiently developed for some domain of living systems, it can be used to assess the optimality of the plans for achieving life goals. However, even if actual living systems are found to have developed optimally, this fact would not infer the necessity of an intelligent modifier of nature because the behavioral and structural constraints may have forced an optimal or near-optimal plan anyway.

The crucial question is whether sufficient evidence can be found in nature to reveal the intentions of a creative agent within the created result itself. Given a comprehensive theory of function for life based on known physical theory, three possible outcomes are:

Multiple plans by which complex life could have arisen (optimally or not). This case would be critical only in assessing the possibility of the Darwinian notion of development based on a continuity of steps using only known causal operators (known physical laws) on known kinds of natural structures (matter and energy of early Earth).

No possible plan. Then an intelligent agent acting on nature as we know it or undiscovered mechanisms in nature are possibilities.

A unique plan. If life could have developed in only one way, then the causal and structural constraints are determinative of life. Either there is no intelligent cause of nature or the intelligence is revealed in the unique set of physical constraints necessary for life’s development.

Physicist Eugene Wigner noted the unreasonableness of how human rationality, as expressed in mathematics, mirrors what is found in nature. A Creator of the universe can be sought at the level of rationality in nature generally (as do theistic evolutionists) or in a more specific revelation of intelligent acts affecting regions of nature, such as the origin of the biosphere (as do intelligent design theorists).

The complexity of life makes development of significant functional theories challenging. However, precedent has been set in engineering, particularly among roboticists who, in a sense, are working on the same problem but from the other end; from a design point of view. Instead of analyzing existing life, they are attempting to produce embodied intelligence (broadly defined), beginning with what is presently known from science, engineering and mathematics, philosophy and theology.

Darwinism and Engineering

The Darwinian “mechanism” (whatever the causes) of natural selection by an environment presented with alternative organisms through genetic mutation is claimed to have produced the present complexity from the simplest forms of life. If so, from an engineering viewpoint, this at first appears to be a highly empirical way to design more complex organisms, much like Thomas Edison randomly trying various materials for light-bulb filaments with little selection criteria. The search mechanism is to randomly select a given operator at each branch in the search tree of problem space. However, because the actual mechanisms of the Darwinian “mechanism” are unidentified, it is not possible to conclude that the search is entirely unguided. Intelligent-design theorists postulate that it is guided in a top-down, model-driven or goal-driven way while theistic evolutionists postulate bottom-up, data-driven or environment-driven guidance. Highly non-optimal search suggests to others a lack of any guidance.

Life development issues are similar in kind to major issues of robotics and control theory. Models of robots can be viewed as dual abstraction hierarchies 6, the sensory one running from the environment upward, to successively higher levels of abstraction from the acquired data at the lowest level. The action hierarchy runs top-down, starting with the highest goal of the robot, at the level of greatest cognitive activity and model of the environment. The action hierarchy decomposes high-level goals into successively smaller tasks until, at the lowest level, actuators interact with the environment to effect a change. Cross-coupling between corresponding levels of the dual hierarchies allows for feedback and decision-making at each level, thus guiding the search for the goal. Some roboticists emphasize the importance of the top-down, or model-driven, approach while others opt for data-driven architectures. Intelligent design corresponds roughly to the top-down approach and theistic evolution to the bottom-up approach. Both are guided, but in different ways corresponding to the different approaches to building robots.

Figure 2
Figure 4. Dual goal-driven and data-driven hierarchies, cross-coupled at each level of abstraction. Sensory processing is given a goal context at each level; behavior generation is guided by sensed data, a context for decision-making in generating action. Goal decomposition occurs at successive (lower) levels while data abstraction occurs at successive (higher) levels in the sensory-processing hierarchy.

The data-driven emphasis in robotics research (led by Rod Brooks at the MIT AI Lab) 7 emphasizes the importance of interaction of the robot with the environment. It differs from efforts which emphasize plan cognition (extensive reasoning about what to do) from less sensory data, opting for intensive interaction with the environment and relatively simple decision-making. With either emphasis, the research goal is to achieve a superior theory of function.

Progress is usually the result of factors unanticipated in the hypothesis under test. Rationally elegant ideas often go nowhere in a real environment and fail to provide significant explanatory power. As roboticists, Flynn and Brooks have stated:

Many of the preconceived notions entertained before we started building our robots turned out to be misguided. Some issues we thought would be hard have worked successfully from day one and subsystems we imagined to be trivial have become tremendous time sinks. 8

Life is far more complex than robots containing a few microprocessors and dedicated state-machines, operating in the relatively simple environment of research lab corridors. Yet grand hypotheses about how life developed from simple to complex forms seem more readily accepted among biologists than even the more specific and concretely testable ideas of robotics among roboticists.


For biology, functional theories might not so much falsify Darwinism as provide the insights it fails to provide. Discovery of detailed mechanisms for the major advances in the development of life might only be possible given a sufficient functional theory by which to interpret information-based communication and control in organisms. Functional understanding of a broken engine, for instance, can guide the search for identification of failure symptoms (behavior) and finally, the failed device (structure). Similarly, trustworthy functional theories could guide the search for order in neurophysiology or genetics by offering a global explanation of what the components (neurons, genes) do relative to the overall function of the system of which they are components.

Biologists looking for new conceptual tools for handling complexity are finding them mainly in fields that require functional theories, such as robotics, knowledge representation (computer science) and adaptive and nonlinear systems and control theory in what are largely engineering disciplines. The concepts of these fields are not merely about how to manage structural and behavioral complexity (with abstraction hierarchies and modularity) but are also about achieving goals and developing theories about plans.

While organisms are not nearly as well understood as electronic circuits or rocket engines, the functional approach offers a productive path to significant progress in a field dominated by a search for insight at the levels of structure and behavior. Better functional descriptions of major systems in an organism could cast light on how some of the biochemically-described behaviors achieve hypothesized functions. And in reverse, biochemical and genetic constraints guide development of functional theories.

The final question is whether a living organism could be declared to be designed on the basis of an accepted functional theory of life. The functional approach attempts to determine whether natural systems can be explained more powerfully and simply by use of functional descriptions, built on existing theories of structure and behavior. Such functional theories entail goals and plans, showing how natural systems are consistent with a theory revealing their design. Yet design is an interpretation of nature that can result from successful functional theories. If such theories provide superior insight into nature, they offer the equivalent philosophical stepping-stone to a Creator as chance or contingency in Darwinian theory does for naturalism or accidentalism. For either worldview, taking the further step is a move from science to more basic beliefs.


1. Beakley, George C., Leach, H.W. (1967). Engineering: An Introduction to a Creative Profession, ch. 15 (“The Engineering Method of Problem Solving”), ch. 17 (“The Engineer-A Creative Person”), ch. 19 (“The Engineering Design Process”). London: Macmillan. return to text

2. Bobrow, Daniel G., ed (1965). Qualitative Reasoning about Physical Systems. Cambridge, MA: MIT Press.

3. Freiling, Michael Joseph (August 1977). “The Use of a Hierarchical Representation in the Understanding of Mechanical Systems.” Cambridge, MA: Ph.D. thesis, MIT. return to text

4. Papazian, Pegor (June 1991). “Principles, Opportunism and Seeing in Design: A Computational Approach.” MIT AI Lab memo no. 1309. return to text

5. Bohm, David. Causality and Chance in Modern Physics. return to text

6. Albus, James S. (1981). Brains, Behavior and Robotics, ch. 5 (“Hierarchical Goal-Directed Behavior”). New York: BYTE Books, McGraw-Hill. return to text

7. Brooks, Rodney A. (September 1987). “Planning Is Just a Way of Avoiding Figuring Out What to Do Next.” Working Paper 303, MIT AI Lab. return to text

8. Flynn, Anita M. and Brooks, Rodney A. (October 1989). “Battling Reality.” AI Memo 1148, MIT AI Lab. return to text

Copyright © 1999 Dennis L. Feucht. All rights reserved. International copyright secured.
File Date: 6.1.99