Citation: Pattee, H.H.. . "The Physics of Symbols: Bridging the Epistemic Cut". Biosystems. Vol. 60, pp. 5-21.
This paper is also available from Elsevier's web site
Abstract:Evolution requires the genotype-phenotype distinction, a primeval epistemic cut that separates energy-degenerate, rate-independent genetic symbols from the rate-dependent dynamics of construction that they control. This symbol-matter or subject-object distinction occurs at all higher levels where symbols are related to a referent by an arbitrary code. The converse of control is measurement in which a rate-dependent dynamical state is coded into quiescent symbols. Non-integrable constraints are one necessary conditions for bridging the epistemic cut by measurement, control, and coding. Additional properties of heteropolymer constraints are necessary for biological evolution.
Keywords: symbol reference, measurement and control, non-integrable constraints, subject-object
"How, therefore, we must ask, is it possible for us to distinguish the living from the lifeless if we can describe both conceptually by the motion of inorganic corpuscles?"
Karl Pearson The Grammar of Science
At the end of the nineteenth century there was little interest among scientists in dualistic and vitalistic views of life. It was because Karl Pearson (1937) believed that life was entirely a physical process that he was led to ask (in 1892) what physically distinguishes the living from the lifeless. He suggested that organic molecules have "secondary characteristics," but whether these characteristics could be derived from fundamental inorganic laws, "we have not at present the means of determining." Early in this century, largely as a result of quantum theory, there arose a feeling among many physicists that there was still some essential mystery to life that might require serious reinterpretation of physics. For example, in 1949 Max Delbrück, influenced by Bohr's (1933) earlier speculations1, wrote: "It may turn out that certain features of the living cell, including perhaps replication, stand in a mutually exclusive relationship to the strict application of quantum mechanics, and that a new conceptual language has to be developed to embrace this situation." Gunther Stent (1966) described the situation in the 1940s in the following way: "Thus it was the romantic idea that 'other laws of physics' (Schrödinger) might be discovered by studying the gene that really fascinated physicists. This search for the physical paradox, this quixotic hope that genetics would prove incomprehensible within the framework of conventional physical knowledge, remained an important element in the psychological infrastructure of the creators of molecular biology."
Then in the early 1950s began an explosive growth of molecular biology and origin of life experiments, the latter beginning in 1953 with the abiogenic synthesis of amino acids by Miller and Urey. The same year Watson and Crick announced the double helix of DNA. Some of the other major advances were the use of X-ray diffraction to derive the structure of myoglobin and hemoglobin by Perutz and Kendrew, and the isolation of a DNA polymerase by Kornberg. This was followed in the 1960s with the breaking of the genetic code by Nirenberg and Khorana, the discovery of gene regulation by Jacob and Monod, the discovery of messenger RNA by Brenner, Jacob and Meselson, the sequencing of transfer RNA by Holly, and the discovery of plasmids by Lederberg.
By 1970 there was no longer much interest in possible paradoxes or revisions of physical theories to accommodate living systems. Nothing new appeared to be needed. Kendrew (1967) summarized the molecular biologists' position in Scientific American: ". . . up to the present time conventional, normal laws of physics and chemistry have been sufficient." This is now the generally accepted view among biologists2. But this reductionist view is really only an response to dualism and vitalism. This view does not even address Pearson's question. If it were stated as an "answer" it would be a total non sequitur: Life is distinguished from the lifeless because it follows the conventional, normal laws of physics and chemistry of lifeless matter.
In contrast to this dominant reductionist view of molecular biology, there continued to be a minority of more skeptical and holistically minded thinkers who believed that physical laws are incomplete or inapplicable in their present form (e.g., Wigner, 1961; Burgers, 1965; Elsasser, 1975; Rosen, 1991)3. There have also continued to be many speculations about whether life can be adequately explained by classical models without incorporating quantum dynamics.
In the last decade there has arisen in addition to these opposing schools of physical reductionists and physical skeptics, a third school that models life and evolution disregarding elementary physical laws altogether. Some well-known examples are Langton's (1989) replicating cellular automata, Ray's (1992) Tierra program, Holland's (1995) Echo model using genetic algorithms, random Boolean nets of Kauffman (1993), Fontana's (1992) algorithmic chemistry, and many artificial life computer simulations. Von Neumann (1966) is often cited as the founder of artificial life studies because of his logical theory of self-replication, but it is important to emphasize that he did not believe that such physics-free models would answer, "the most intriguing, exciting, and important question of why the molecules . . . are the sort of things they are4.
Many other abstract descriptions of life now fall under the title of complexity theory. This field is dominated by mathematical approaches, nonlinear dynamics, ergodic theory, random manifolds, self-organized criticality, and information and game theory (e.g., Cowan, Pines, and Meltzer, 1994). Complexity theorists are looking for universal principles of complex systems that apply at all levels, from spin glasses and sandpiles to cells and societies. The relation of these models to biology, and even to physics, is often a controversial issue. The power of computers to simulate models of self-replication, development, evolution, and ecology have resulted in many interesting behaviors. Computation also allows the study of nonlinear dynamics that generate endless formal complexity. However, because of the high degree of abstraction, these simulations are often difficult to interpret, and their applicability to biology is uncertain. Direct empirical justification is hard to find for such abstract models. In any case, since these models do not directly involve any microscopic physical laws and apply to both living and lifeless systems they do not address Pearson's question. If asked Pearson's question, the physics-free modeler would answer that the essential properties of life are distinguished by abstract relations that do not depend on any particular physical realization.
Most recently there is great interest in computer-controlled robots situated in a real physical environment. The computer control is often an artificial neural network, and adaptive learning may involve genetic algorithms (e.g., Varela et al., 1991; Brooks, 1992; Brooks and Maes, 1994; Clark, 1997). These models generally favor a coherent dynamic interpretation of control rather than the symbolic, rule-based "representations" of older artificial intelligence models. Since the environment is real there is no need to model any physics. The concept of symbol is usually regarded as an artifact generated by an underlying dynamics. The physical world consists of only those aspects of the environment that the robot can actually detect.
This type of dynamical control for sensorimotor behavior appears to be a plausible model that would help to account for the speed and complexity of responses in organism with relatively complex sensorimotor behavior and small brains (although it has not yet done so). Any form of symbolic representation or rule-based computation at neuronal speeds is simply too slow and would require much larger brains. Insect behaviors, such as flying around obstacles, landing on a twig, or mating in flight in a gusty wind, do not allow any solution except by some form of coherent real-time dynamics. If asked Pearson's question, a roboticist would probably claim that with respect to sensorimotor control there is no fundamental physical distinction between living organism and adaptive dynamically controlled robots, even though they would agree that there is at present an enormous gap between the most complex robots and the simplest insects.
The problem with this view is that the dynamics of sensorimotor control and learning is only one aspect of life. The problem of reliable self-replication, the origin of novel sense organs and motor structures, i.e., open-ended evolution, as yet has no model based solely on a temporal dynamics. Even if cognition and brain function should turn out to be described as a temporally coded dynamics with no static symbol structures, that will not adequately describe the quiescent molecular structures that form the genome and the coding constraints that have been controlling protein synthesis for billions of years.
Many biologists consider physical laws, artificial life, robotics, and even theoretical biology as largely irrelevant for their research. In the 1970s, a prominent molecular geneticist asked me, "Why do we need theory when we have all the facts?" At the time I dismissed the question as silly, as most physicists would. However, it is not as silly as the converse question, Why do we need facts when we have all the theories? These are actually interesting philosophical questions that show why trying to relate biology to physics is seldom of interest to biologists, even though it is of great interest to physicists. Questioning the importance of theory sounds eccentric to physicists for whom general theories is what physics is all about. Consequently, physicists, like the skeptics I mentioned above, are concerned when they learn facts of life that their theories do not appear capable of addressing. On the other hand, biologists, when they have the facts, need not worry about physical theories that neither address nor alter their facts. Ernst Mayr (1997) believes this difference is severe enough to separate physical and biological models: "Yes, biology is, like physics and chemistry, a science. But biology is not a science like physics and chemistry; it is rather an autonomous science on a par with the equally autonomous physical sciences."
There are fundamental reasons why physics and biology require different levels of models, the most obvious one is that physical theory is described by rate-dependent dynamical laws that have no memory, while evolution depends, at least to some degree, on control of dynamics by rate-independent memory structures. A less obvious reason is that Pearson's "corpuscles" are now described by quantum theory while biological subjects require classical description in so far as they function as observers. This fact remains a fundamental problem for interpreting quantum measurement, and as I mention below, this may still turn out to be essential in distinguishing real life from macroscopic classical simulacra. I agree with Mayr that physics and biology require different models, but I do not agree that they are autonomous models. Physical systems require many levels of models, some formally irreducible to one another, but we must still understand how the levels are related. Evolution also produces hierarchies of organization from cells to societies, each level requiring different models, but the higher levels of the hierarchy must have emerged from lower levels. Life must have emerged from the physical world. This emergence must be understood if our knowledge is not to degenerate (more than it has already) into a collection of disjoint specialized disciplines.
I first became aware of Karl Pearson's question about 1939 when I was in the 8th grade. My Headmaster and science teacher, Dr. P. L. K. Gross, had given me the 1937 Everyman edition of The Grammar of Science (the first edition was published in 1892). Most of the book was beyond my comprehension, but I thought I understood the chapter on Life, subtitled, "The Relation of Biology to Physics". To make the long story of my education short, a decade later, working on my PhD in physics, I graduated to Hermann Weyl's (1949), equally profound but more up to date, Philosophy of Mathematics and Natural Science which concludes with appendices on "Physics and Biology", and "Morphe and Evolution". This became my philosophy of science source book. At the time, physics and mathematics were inseparable in my mind. I did not clearly distinguish formal symbolic models from reality. As Weyl notes, "Perhaps the philosophically most relevant feature of modern science is the emergence of abstract symbolic structures as the hard core of objectivity behind - as Eddington puts it - the colorful tale of the subjective storyteller mind."
In the 1960s my first serious thinking about the relation of abstract symbol structures to the physical laws they represent, was revealed to me by the physicist Max Born (1969) in a paper titled "Symbol and Reality," in which he recalls as a young student his own shock when it dawned on him that all our perception and mental imagery, "everything without exception," is entirely subjective, and that only by the use of symbols can we communicate any objective components of our subjective, private experiences. Born's condition for the objective use of symbols is "decidability," a term he coined to express the function of experiment. If a symbolic expression lacks empirical decidability potentially available to all observers it has no necessary relation to any objective reality. I also was intrigued by Eugene Wigner's (1960) paper "On the Unreasonable Effectiveness of Mathematics in the Natural Sciences" which asks many fundamental questions about the nature of mathematical symbols5. Thus, following Born, my effort toward answering the question: How is physics related to biology? was augmented by the question: How are physical laws related to the mathematical symbols on which their representation depends? I now often state the same questions more generally: How are universal, inexorable, natural laws related to local, arbitrary symbols?
My first motivation for understanding the relation of symbols to living organisms arose earlier from the origin of life problem. In 1954 I had completed my doctoral research on X-ray optics used to study biological structures. Discovering the physical structure of nucleic acids and proteins was then a major problem of the new molecular biology. I realized, however, that self-replication of such complex structures was an entirely different and more difficult type of problem. I thought it was obvious that reliable self-replication would require objective communication of whatever structure is replicated. In other words, for evolution to be possible, any description of a "self" must be communicated objectively to all descendent cells no matter what particular "self" is being replicated. Here, objective simply means that the same instructions will produce the same results in all descendants.
I tried several self-organizing schemes using automata models for generating and replicating simulated copolymer sequences (Pattee, 1961, 1965), but it became clear that the evolutionary potential of all these models was very limited. I eventually recognized a fundamental problem in all such rule-based self-organizing schemes, namely, that in so far as the organizing depends on internal fixed rules, the generated structures will have limited potential complexity, and in so far as any novel organizing arises from the outside environment, the novel structures have no possibility of reliable replication without a symbolic memory that could reconstruct the novel organization. The first computer simulation I felt had some interesting evolutionary potential was developed by Michael Conrad (1969) in which genetic, cellular, population, and ecological levels were all represented. However, other than abstract conservation principles, this was a physics-free model that did not address Pearson's question or the nature of symbols in measurement and control processes. (Conrad and Pattee, 1970).
By the 1970s, I believed I had some insight on Pearson's question. These ideas, which I will summarize below, were presented in the four volumes of Waddington's (1968-72) Bellagio conferences on theoretical biology. My first question then was: How can we describe in physical language the most elementary heritable symbols? It has turned out that for even the simplest known case, the gene, an adequate description requires the two irreducibly complementary concepts of dynamical laws and non-integrable constraints that are not derivable from the laws. This primeval distinction between the individual's local symbolic constraints that first appear at the origin of life and the objective universal laws, reappears in many forms at higher levels.6 From von Neumann (1955) I learned that this same epistemic cut occurs in physics in the measurement process, i.e., the fact that dynamical laws cannot describe the measurement function of determining initial conditions.
Later I saw these as special cases of the general epistemic problem: how to bridge the separation between the observer and the observed, the controller and the controlled, the knower and the known, and even the mind and the brain. This notorious epistemic cut has motivated philosophical disputes for millennia, especially the problem of consciousness that only recently has begun to be treated as possibly an empirically decidable problem (e.g., Shear, 1997; Taylor, 1999). My second question was whether bridging the epistemic cut could even be addressed in terms of physical laws.
Of course my answers to Pearson's question were not complete, nor of much interest to biologists. I had only stated some necessary but by no means sufficient conditions for the physical description of symbolic control. This was the easy part of the question. My concept of what symbolic behavior must also entail was greatly enlarged by Ernst Cassirer's (1957) Philosophy of Symbolic Forms. Later, at the Bellagio meetings, the philosopher Marjorie Grene (1974) introduced me to Michael Polanyi's insights on the failure of all our symbolic expressions, especially formal mathematical expressions, to achieve the ideal of objectivity. Polanyi's (1964) anti-reductionist arguments show how all of our explicit symbolic descriptions must be grounded in a reservoir of ineffable structures and subsidiary knowledge. But more important to me at the time was his paper, "Life's Irreducible Structure" (Polanyi, 1968) because it made the same essential point I had made that the structural complexity we associate with life can only be described in the language of physics as special constraints or machine-like "boundary conditions" that "harness" the laws, but that are not formally derivable from physical laws7. Polanyi also recognized the irreducibility of all higher evolved functional hierarchical levels to the lower levels from which they evolved (Pattee, 1969a).
The most profound historical influence was John von Neumann. His 1966 discussion of self-reproducing automata suggested that efficient control of dynamical construction requires a non-dynamic "quiescent description", and this I interpreted as equivalent to an epistemic cut between objective dynamical laws and subjective non-dynamic symbolic constraints describing the "self". Von Neumann also asked a question that I found to be closely related to Pearson's question: Why are the basic macromolecules of organisms so much larger than the fundamental particles of physical theory?4
Equally influential was von Neumann's (1955) discussion of the necessity of an epistemic cut in any measurement process (see Sec. 9) showing that the function of measurement is necessarily irreducible to the dynamics of the measuring device. This logic is closely related to the necessary separation of symbols and dynamics for control of self-replication since measurement and control are inverse processes, i.e., measurement transforms physical states to symbols in memory, while memory-stored controls transform symbols to physical states.
"Again, if all movement is interconnected, the new arising from the old in a determinate order, if the atoms never swerve so as to originate some new movement that will snap the bonds of fate, the everlasting sequence of cause and effect what is the source of the free will possessed by living things throughout the earth?" [Lucretius, On the Nature of Things].
If life is distinguished by memory-stored controls, and if memory and control imply alternative movements, then in order to answer Pearson's question we must first answer Lucretius' question. How do we snap the bonds of the inexorable dynamical laws that do not allow new alternative movements? To understand the problem it is essential to focus on how physical laws are actually described in their mathematical form. The "unreasonable effectiveness" of this type of formal description is largely the result of the precision of its execution (which I will not duplicate here). This formulation has a developmental history that is also important. Newtonian dynamics began with point masses moving under laws of gravitational force or generally under laws derived from the potential and kinetic energy of the system. These laws are expressed as differential equations in time (equations of motion) that define an infinite family of possible orbits in the state space (phase space).
Three closely related epistemic conditions are fundamental for this type of dynamical description and need to be emphasized: (1) To begin this type of description, the world must be separated into the states and the laws that change the states. The detailed (microscopic) laws are expressed as rate-dependent differential equations that define families of orbits in a state space. This means that the paths from states to states are unambiguously deterministic and reversible. (2) Only when a particular system is located in this space by the act of measurement (determining its initial conditions, i.e., the positions and velocities of all particles at a particular time) do the equations lead to any observable consequences and allow an actual orbit to be calculated by integrating the equations of motion. (3) Finally, and most important, this form of description can claim to be objective only if the act of measurement does not influence the form of the laws, and if the laws do not influence the act of measurement.
The first extension of this model was to solid bodies that can be pictured as many point masses held together by fixed (non-dynamic) forces. The nature of these forces was a mystery to Newton. We now attribute them to electromagnetic, quantum, or fundamental particle forces that in principle may also be described in more detail by dynamical laws. Usually these internal forces do no work and therefore play no role in the dynamical laws of the solid body. They may be interpreted as fixed, thereby greatly reducing the number of variables (degrees of freedom) that enter into the equations of motion. These internal (reactive or geometric) forces are called forces of constraint. What we call more or less rigid structures, from natural molecules, crystals, and rocks, to artificial tables, buildings, and bridges are held together by forces of constraint.
However, there are also flexible forces of constraint that hold together the innumerable articulated assemblies of rigid structures we call machines, as well as labile assemblies of not-so-rigid structures like the biopolymers that execute measurement and control processes in organisms. It is the physical descriptions of these flexible and articulated constraints that need to be explained in more detail in order to begin to answer both Lucretius' and Pearson's questions.
Before defining these flexible constraints I need to emphasize the generality of the dynamics that they can control. The Newtonian or classical picture is often believed to have been replaced by relativity and quantum theory, but this is not a fair assessment of Newtonian dynamics. First, classical laws are still valid for gravitational forces and velocities small compared to the velocity of light. Second, the results of all measurements of both relativistic and quantum mechanical systems must be expressed in this classical language. It is true that the forms of these modern dynamical laws are different from Newton's, and that the concept of state is defined entirely differently, but the three fundamental epistemic conditions must still hold for all dynamical laws to make objective sense. It is still required that (1) the laws and the initial conditions be crisply separated, (2) initial conditions must be determined by measurements, and (3) measurement and laws must not influence each other8. It is for this reason that Eugene Wigner (1982) considered Newton's greatest discovery, not his laws but rather, "his sharp separation of initial conditions and laws of nature."
This highly developed intellectual distinction between initial conditions and laws is a form of epistemic cut, but I want to make the point that this cut has a primitive origin and is found in all living organisms. It is simply an extreme case of the distinction made, even by the first cells, between stimuli that cannot be correlated and stimuli that can be correlated or that follow a recognizable pattern. In terms of information storage, we say that some records of events can have a compressed description (like laws) because of intrinsic correlations, while other records (like initial conditions) have no shorter description than the records themselves. Dynamical laws express a maximal compression of all events of a particular type, namely, those on orbits determined by one set of initial conditions. Given the initial conditions, all past and future events are determined by the laws (except for singularities). Historically this behavior led to the ideal of Laplacean determinism. Today it is more modestly called state-determined behavior. Nevertheless, viewed as a formal description all dynamical laws remain syntactically deterministic, even if they are interpreted as probabilities. Furthermore, our perceptions as well as our natural languages support a deterministic, either-or logical syntax and causal semantics that conform to a classical dynamics. It is for this reason that the interpretation of quantum theory is largely ineffable.
This maximal compression of events by dynamical laws has indeed proven unreasonably effective in describing all the fundamental microscopic laws. But we know from everyday decision-making that all our experiences are not completely compressible and that our life is not state-determined. As anyone can directly observe, there exists between these extremes of universal and maximally compressible laws and local incompressible initial conditions many intermediate levels of natural and artificial local constraint structures that have measurement and control functions and that exhibit various degrees of local and partially compressible behaviors. As Gell-Mann (1994) has observed: "the effective complexity [of the universe] receives only a small contribution from the fundamental laws. The rest comes from the numerous regularities resulting from 'frozen accidents'." I would add that to be effective in evolution these regularities from frozen accidental constraints must be heritable. That means they must be reconstructible from a memory. Both machines and organisms are characterized by such constraints. The question remains: How can such special-purpose constraints be described in precise physical terms?
Since we know that a heritable genetic memory is an essential condition for life, my approach to the problem of determinism began by expressing the precise requirements for a constraint that satisfies the conditions for heritability. I can do no better than to restate my early argument (Pattee, 1969b):
"A physical system is defined in terms of a number of degrees of freedom which are represented as variables in the equations of motion. Once the initial conditions are specified for a given time, the equations of motion give a deterministic procedure for finding the state of the systems at any other time. Since there is no room for alternatives in this description, there is apparently no room for hereditary processes. . . The only useful description of memory or heredity in a physical system requires introducing the possibility of alternative pathways or trajectories for the system, along with a 'genetic' mechanism for causing the system to follow one or another of these possible alternatives depending on the state of the genetic mechanism. This implies that the genetic mechanism must be capable of describing or representing all of the alternative pathways even though only one pathway is actually followed in time. In other words, there must be more degrees of freedom available for the description of the total system than for following its actual motion. . . Such constraints are called non-holonomic."
In more common terminology, this type of constraint is a structure that we say controls a dynamics. To control a dynamical systems implies that there are control variables that are separate from the dynamical system variables, yet they must be described in conjunction with the dynamical variables. These control variables must provide additional degrees of freedom or flexibility for the system dynamics. At the same time, typical control systems do not remove degrees of freedom from the dynamical system, although they alter the rates or ranges of system variables. Many artificial machines depend on such control constraints in the form of linkages, escapements, switches and governors. In living systems the enzymes and other allosteric macromolecules perform such control functions. The characteristic property of all these non-holonomic structures is that they cannot be usefully separated from the dynamical system they control. They are essentially nonlinear in the sense that neither the dynamics nor the control constraints can be treated separately.
This type of constraint, that I prefer to call non-integrable, solves two problems. First, it answers Lucretius' question. These flexible constraints literally cause "atoms to swerve and originate new movement" within the descriptive framework of an otherwise deterministic dynamics (this is still a long way from free will). They also account for the reading of a quiescent, rate-independent memory so as to control a rate-dependent dynamics, thereby bridging the epistemic cut between the controller and the controlled. Since law-based dynamics are based on energy, in addition to non-integrable memory reading, memory storage requires alternative states of the same energy (energy degeneracy). These flexible, allosteric, or configuration-changing structures are not integrable because their motions are not fully determined until they couple an explicit memory structure with rate-dependent laws (removal of degeneracy).
The crucial condition here is that the constraint acts on the dynamic trajectories without removing alternative configurations. Thus, the number of coordinates necessary to specify the configuration of the constrained system is always greater than the number of dynamic degrees of freedom, leaving some configurational alternatives available to "read" memory structures. This in turn requires that the forces of constraint are not all rigid, i.e., there must be some degeneracy to allow flexibility. Thus, the internal forces and shapes of non-integrable structures must change in time partly because of the memory structures and partly as a result of the dynamics they control. In other words, the equations of the constraint cannot be solved separately because they are on the same formal footing as the laws themselves, and the orbits of the system depend irreducibly on both (Whittaker, 1944; Sommerfeld, 1956; Goldstein, 1953; Neimark and Fufaev, 1972).
What is historically amazing is that this common type of constraint was not formally recognized by physicists until the end of the last century (Hertz, 1894). Such structures occur at many levels. They bridge all epistemic cuts between the controller and the controlled, the classifier and the classified, the observer and the observed. There are innumerable types of non-integrable constraints found in all mechanical devices in the forms of latches, and escapements, in electrical devices in the form of gates and switches, and in many biological allosteric macromolecules like enzymes, membrane channel proteins, and ciliary and muscle proteins. They function as the coding and decoding structures in all symbol manipulating systems.
There is a significant difference between the way non-integrable constraints are treated in classical and quantum mechanics. Describing non-integrable constraints in quantum theory restricts the wave function as if some measurement is being made (Eden 1951). This led me to speculate that the non-integrable constraints of the enzyme molecule execute a quantum measurement and rate control function with a specificity and speed that no classical measurement and control device could match. (Pattee, 1967).
The concept of constraint is not considered fundamental in physics because the (internal, geometric reactive) forces of constraint can, in principle, be reduced to active impressed forces governed by energy-based microscopic dynamical laws. The so-called fixed geometric forces are just stationary states of a faster, more detailed dynamics. This reducibility to microscopic dynamics is possible in principle for structures, even if it is computationally completely impractical. However, describing any bridge across an epistemic cut by a single dynamical description is not possible even in principle.
The most convincing general argument for this irreducible complementarity of dynamical laws and measurement function comes again from von Neumann (1955, p. 352). He calls the system being measured, S, and the measuring device, M, that must provide the initial conditions for the dynamic laws of S. Since the non-integrable constraint, M, is also a physical system obeying the same laws as S, we may try a unified description by considering the combined physical system (S + M). But then we will need a new measuring device, M', to provide the initial conditions for the larger system (S + M). This leads to an infinite regress; but the main point is that even though any constraint like a measuring device, M, can in principle be described by more detailed universal laws, the fact is that if you choose to do so you will lose the function of M as a measuring device. This demonstrates that laws cannot describe the pragmatic function of measurement even if they can correctly and completely describe the detailed dynamics of the measuring constraints.
This same argument holds also for control functions which includes the genetic control of protein construction. If we call the controlled system, S, and the control constraints, C, then we can also look at the combined system (S + C) in which case the control function simply disappears into the dynamics. This epistemic irreducibility does not imply any ontological dualism. It arises whenever a distinction must be made between a subject and an object, or in semiotic terms, when a distinction must be made between a symbol and its referent or between syntax and pragmatics. Without this epistemic cut any use of the concepts of measurement of initial conditions and symbolic control of construction would be gratuitous.
"That is, we must always divide the world into two parts, the one being the observed system, the other the observer. In the former, we can follow up all physical processes (in principle at least) arbitrarily precisely. In the latter, this is meaningless. The boundary between the two is arbitrary to a very large extent. . . but this does not change the fact that in each method of description the boundary must be placed somewhere, if the method is not to proceed vacuously, i.e., if a comparison with experiment is to be possible." (von Neumann, 1955, p.419)
The epistemic cut or the distinction between subject and object is normally associated with highly evolved subjects with brains and their models of the outside world as in the case of measurement. As von Neumann states, where we place the cut appears to be arbitrary to a large extent. The cut itself is an epistemic necessity, not an ontological condition. That is, we must make a sharp cut, a disjunction, just in order to speak of knowledge as being "about" something or "standing for" whatever it refers to. What is going on ontologically at the cut (or what we see if we choose to look at the most detailed physics) is a very complex process. The apparent arbitrariness of the placement of the epistemic cut arises in part because the process cannot be completely or unambiguously described by the objective dynamical laws, since in order to perform a measurement the subject must have control of the construction of the measuring device. Only the subject side of the cut can measure or control.
The concept of an epistemic cut must first arise at the genotype-phenotype control interface. Imagining such a subject-object distinction before life existed would be entirely gratuitous, and to limit control only to higher organisms would be arbitrary. The origin problem is still a mystery. What is the simplest epistemic event? One necessary condition is that a distinction is made by a subject that is not a distinction derivable from the object. In physical language this means a subject must create some form of distinction or classification between physical states that is not made by the laws themselves (i.e., measuring a particular initial condition, removing a degeneracy or breaking a symmetry). In the case of the cell, the sequences of the gene are not distinguished by physical laws since they are energetically degenerate. Where does a new distinction first occur? It is where this memory degeneracy is partially removed, and that does not occur until the protein folding process. Transcription, translation, and copying processes treat all sequences the same and therefore make no new distinctions, but of course they are essential for constructing the linear constraints of the protein that partially account for the way it folds. The folded protein removes symbol vehicle degeneracy, but it still has many degenerate states (many conformations) that are necessary for it to function as a non-integrable constraint.
It is important to recognize that the details of construction and folding at this primeval epistemic cut make no sense except in the context of an entire self-replicating cell. A single folded protein has no function unless it is a component of a larger unit that maintains its individuality by means of a genetic memory. We speak of the genes controlling protein synthesis, but to accomplish this they must rely on previously synthesized and organized enzymes and RNAs. This additional self-referent condition for being the subject-part of an epistemic cut I have called semantic (or semiotic) closure (Pattee, 1982, 1995). This is the molecular chicken-egg closure that makes the origin of life problem so difficult.
We can now give a direct answer to Pearson's question: It is not possible to distinguish the living from the lifeless by the most detailed "motion of inorganic corpuscles" alone. The logic of this answer is that life entails an epistemic cut that is not distinguishable by microscopic (corpuscular) laws. As von Neumann's argument shows, any distinction between subject and object requires a description of the constraints that execute measurement and control processes; and such a functional description is not reducible to the dynamics that is being measured or controlled.9
This is still far from a complete answer. There is more to evolvability than heritable variation and natural selection and control of dynamical laws by memory and non-integrable constraints. It is not at all obvious why such local control structures should persist in a real world full of uncorrelated irregular events. Controlling predictable physical laws is only part of the problem of survival. There are additional physical conditions for evolvability. These are not abstract principles but specific requirements on how efficaciously the epistemic cut is actually bridged. I will only mention three of these conditions. Evolution depends critically on (1) how easily gene sequences corresponding to functional proteins can be found, (2) how reliably these sequences can control construction of proteins, and on (3) how smoothly or gradually variations in the sequences can produce adaptation of function. In other words, evolvability depends on the many physical and statistical details of how the actual epistemic bridge from symbols to dynamics is executed
There is a real conceptual roadblock here. In our normal everyday use of languages the very concept of a "physics of symbols" is completely foreign. We have come to think of symbol systems as having no relation to physical laws. This apparent independence of symbols and physical laws is a characteristic of all highly evolved languages, whether natural or formal. They have evolved so far from the origin of life and the genetic symbol systems that the practice and study of semiotics does not appear to have any necessary relation whatsoever to physical laws. As Hoffmeyer and Emmeche (1991) emphasize, it is generally accepted that, "No natural law restricts the possibility-space of a written (or spoken) text.," or in Kull's (1998) words: "Semiotic interactions do not take place of physical necessity." Adding to this illusion of strict autonomy of symbolic expression is the modern acceptance of abstract symbols in science as the "hard core of objectivity" mentioned by Weyl. This isolation of symbols is what Rosen (1987) has called a "syntacticalization" of our models of the world, and also an example of what Emmeche (1994) has described as a cultural trend of "postmodern science" in which material forms have undergone a "derealization".
Another excellent example is our most popular artificial assembly of non-integrable constraints, the programmable computer. A memory-stored programmable computer is an extreme case of total symbolic control by explicit non-integrable hardware (reading, writing, and switching constraints) such that its computational trajectory determined by the program is unambiguous, and at the same time independent of physical laws (except laws maintaining the forces of normal structural constraints that do not enter the dynamics, a non-specific energy potential to drive the computer from one constrained state to another, and a thermal sink). For the user, the computer function can be operationally described as a physics-free machine, or alternatively as a symbolically controlled, rule-based (syntactic) machine. Its behavior is usually interpreted as manipulating meaningful symbols, but that is another issue. The computer is a prime example of how the apparently physics-free function or manipulation of memory-based discrete symbol systems can easily give the illusion of strict isolation from physical dynamics.
This illusion of isolation of symbols from matter can also arise from the apparent arbitrariness of the epistemic cut. It is the essential function of a symbol to "stand for" something - its referent - that is, by definition, on the other side of the cut. This necessary distinction that appears to isolate symbol systems from the physical laws governing matter and energy allows us to imagine geometric and mathematical structures, as well as physical structures and even life itself, as abstract relations and Platonic forms. I believe, this is the conceptual basis of Cartesian mind-matter dualism. This apparent isolation of symbolic expression from physics is born of an epistemic necessity, but ontologically it is still an illusion. In other words, making a clear distinction is not the same as isolation from all relations. We clearly separate the genotype from the phenotype, but we certainly do not think of them as isolated or independent of each other. These necessary non-integrable equations of constraint that bridge the epistemic cut and thereby allow for memory, measurement, and control are on the same formal footing as the physical equations of motion. They are called non-integrable precisely because they cannot be solved or integrated independently of the law-based dynamics. Consequently, the idea that we could usefully study life without regard to the natural physical requirements that allow effective symbolic control is to miss the essential problem of life: how symbolic structures control dynamics.
Finally, I will summarize some of the physical requirements for successfully bridging the epistemic cut. In effect we are answering von Neumann's "most intriguing, exciting, and important question of why the molecules . . . are the sort of things they are." First is the search problem. It was a problem for Darwin, and with the discovery of the DNA helix and the code that precisely maps base sequences to protein sequences the search problem appeared worse. By assuming that molecular details are significant one sees a base sequence space that is hopelessly large for any detailed search. But while this assumption is correct for the symbolic side of the cut we now know that the assumption is wrong for the function on the other side of the cut. Bridging the epistemic cuts implies executing classifications of physical details, and the quality of the classifications determine the quality of function. We know that protein sequences are functionally highly redundant and that many amino acid replacements do not significantly alter the function. We also know that many base sequence aliases can construct proteins with essentially the same shape. Also, simplified models of RNA secondary folding suggest that the search is not like looking for a specific needle in an infinite haystack, but looking for any needle in a haystack full of needles that are uniformly distributed (e.g., Schuster, 1994). There is also evidence that the search is far more efficient than classical blind variation. Artificial genetic algorithms have shown unexpected success in finding acceptable solutions for many types of search problems that appear logically or algorithmically intractable.
The second requirement is for reliable self-replication. This is a complex adaptive balancing act between conflicting requirements at many levels. On the one hand, complete reliability would not allow any search, variation, or evolution at all. On the other hand, too little reliability will produce extinction by an error catastrophe. At the folding level where the degeneracy of base sequences is partially removed, there must be a balance between a stable energy landscape to allow rapid folding and permanence, and the complex conformational degeneracies necessary for flexible specific binding and rapid catalysis. The folding process is uniquely complex in many ways. It is a transformation across all three spatial dimensions, over temporal scales covering many orders of magnitude, and involving strong bonds and many weaker forces in coherent highly nonlinear interactions. The complexity of any detailed quantum mechanical description of such non-integrable constraints means that such folding problems can only be treated statistically. Even formulating a microscopic description appears intractable. It is not even obvious that a linear sequence of several hundred amino acids, or any such heteropolymer, should fold reliably into a specific globular shape. That such flexible globules should be able to perform high-speed, highly specific catalysis is even less obvious. Yet we know this is the case, and we usually take these incredible functions for granted (e.g., Frauenfelder and Wolynes, 1994).
The last requirement I mentioned was how smoothly variations in the genetic sequences can produce adaptation in functions. Here again there must be a balance between conflicting requirements. Rapid folding and stability of a protein requires steep energy landscapes, while optimization of function requires fine tuning of the folded shape of the protein by small changes in genetic sequences. This requires a relatively smooth energy landscape. Balancing these requirements is eased by large enough molecules so that major folding conditions are buffered from local fine-tuning changes in sequences (e.g., Conrad, 1990). The degree to which these and other requirements are met by natural selection on the one hand and by non-selective ordering principles on the other will only be decided by empirical study of the molecular details.
Pearson's question is about understanding the nature of life. Understanding will depend in part on objective criteria like finding agreement with facts. But there are also subjective criteria, such as the level of abstraction one finds interesting, and the degree of generality one can tolerate in analogies and metaphors. For example, Newton's laws abstract the earth to a mere mass point, an abstraction much too extreme to interest most biologists. Nevertheless, life could not evade these laws, and the course of evolution is profoundly affected by them, as Haldane (1927) pointed out in his essay "On Being the Right Size." At the other extreme, quantum theory is much too detailed to interest most biologists, but life could not evade these laws either, and may in fact depend on them in essential ways we do not yet appreciate. Evolution by heritable variations and selection has resulted in many strange structures and behaviors, but all of them, without exception, obey gravitational, electrodynamic, and quantum and statistical dynamical laws. Artificial life models, complexity theory, and studies of complex adaptive systems are motivated by the hope of discovering more of these abstract and general laws that life must follow.
But there is another type of subjective feeling about understanding life that motivated Pearson's question, the same, I think, that motivated Lucretius' and von Neumann's questions. It is a feeling of paradox, the same feeling that motivated Bohr, Wigner, Polanyi, the skeptics, and somewhat ironically, the founders of what is now reductionist molecular biology, like Delbrück. They all believed that life follows laws, but from their concept of law, they could not understand why life was so strikingly different from non-life. So I find another way of asking this type of question: What exactly does our view of universal dynamical laws abstract away from life, so that the striking distinctions between the living and the lifeless become obscure and apparently paradoxical?
My first answer is that dynamical language abstracts away the subject side of the epistemic cut. The necessary separation of laws and initial conditions is an explicit principle in physics and has become the basis (and bias) of objectivity in all the sciences. The ideal of physics is to eliminate the subjective observer completely. It turned out that at the quantum level this is a fundamental impossibility, but that has not changed the ideal. Physics largely ignores the exceptional effects of individual (subjective) constraints and boundary conditions and focusses on the general dynamics of laws. This is because constraints are assumed to be reducible to laws (although we know they are not reducible across epistemic cuts) and also because the mathematics of complex constraints is often unmanageable. Philosophers have presented innumerable undecidable metaphysical models about the mind-brain cut, and physicists have presented more precise but still undecidable mathematical models about quantum measurement. But at the primeval level, where it all began, the genotype-phenotype cut is now taken for granted as ordinary chemistry.
My second answer is that if you abstract away the details of how subject and object interact, the "very peculiar range" of sizes and behaviors of the allosteric polymers that connect subject and object, the memory controlled construction of polypeptides, the folding into highly specific enzymes and other functional macromolecules, the many-to-many map of sequences to structures, the self-assembly, and the many conformation dependent controls - in other words, if you ignore the actual physics involved in these molecules that bridge the epistemic cut, then it seems unlikely that you will ever be able to distinguish living organisms by the dynamic laws of "inorganic corpuscles" or from any number of coarse-grained artificial simulations and simulacra of life. Is it not plausible that life was first distinguished from non-living matter, not by some modification of physics, some intricate nonlinear dynamics, or some universal laws of complexity, but by local and unique heteropolymer constraints that exhibit detailed behavior unlike the behavior of any other known forms of matter in the universe?
1. Bohr's views are difficult to abbreviate without misrepresentation: In Light and Life Bohr (1933) begins: "On the one hand, the wonderful features which are constantly revealed in physiological investigations and which differ so markedly from what is known of inorganic matter have lead biologists to the belief that no proper understanding of the essential aspects of life is possible in purely physical terms. On the other hand, the view known as vitalism can hardly be given an unambiguous expression by the assumption that a peculiar vital force, unknown to physics, governs all organic life. Indeed, I think we all agree with Newton that the ultimate basis of science is the expectation that nature will exhibit the same effects under the same conditions. If, therefore, we were able to push the analysis of the mechanism of living organisms as far as that of atomic phenomena, we should not expect to find any features foreign to inorganic matter." He then goes on: " . . .the idea suggests itself that the minimum freedom we must allow the organism will be just large enough to permit it, so to say, to hide its secrets from us. On this view, the very existence of life must in biology be . . . [like the quantum of action] taken as a basic fact that cannot be derived from ordinary mechanical physics."
2. The consensus among biologists in the 1970s can be found in Biology and the Future of Man, edited by Philip Handler, a report of a committee of the National Academy of Sciences charged with presenting "a complete overview of the highlights of current understanding of the Life Sciences." Handler's preface states: "The theme of this presentation is that life can be understood in terms of the laws that govern and the phenomena that characterize the inanimate, physical universe and , indeed, that at its essence life can be understood only in the language of chemistry."
3. One must consult their extensive writings for an adequate perspective of these ideas. Eugene Wigner (1961) could not accommodate self-replication using the linear laws of quantum theory. J. M. Burgers (1965) following the philosopher A. N. Whitehead, believed: "There are essential features of life which do not have an unambiguous relationship to states of matter or fields that can be objectively characterized [by normal physics and mathematics]. Their explanation requires reference to subjective features." Burgers viewed these features as a form of memory inherent in the prebiotic universe but emerging and strengthening gradually in the course of evolution. Walter Elsasser (1975) accepted quantum theory as correctly applying to life in principle, but believed that there were also "biotonic" laws governing the "unfathomable complexity" that made life irreducibly nonlinear. Elsasser and Burgers both felt that life stores some forms of information in other than "mechanical means". Robert Rosen (1991) argues that it is "this very segregation into independent categories of causation" that prevents the Newtonian picture from describing the"entailments" and "linkage" relations characteristic of life. He focused on the class of formal structures that model these relations.
4. "By axiomatizing automata in this manner one has thrown half the problem out the window, and it may be the more important half. One has resigned oneself not to explain how these parts are made up of real things, specifically, how these parts are made up of actual elementary particles, or even of higher chemical molecules. One does not ask the most intriguing, exciting, and important question of why the molecules or aggregates which in nature really occur in these parts are the sort of things they are, why they are essentially very large molecules in some cases but large aggregates in other cases, why they always lie in a range beginning at a few microns and ending at a few decimeter. This is a very peculiar range for an elementary object, since it is, even on a linear scale, at least five powers of ten away from the sizes of really elementary entities." (von Neumann, 1966, p. 77)
5. Wigner's (1960) set the tone of his paper with a quote from C. S. Peirce: " . . . and it is probable that there is some secret here which remains to be discovered." His point is that the most effective symbolic formalism, such as matrices, complex numbers and infinite dimensional Hilbert space, cannot be derived from physical observations or physical laws, or even common sense. Yet they appear to be perfectly suited, if not essential, for quantum theory. Wigner concluded with a speculation about life that there may be a conflict between laws of heredity and quantum theory (see also Wigner, 1961).
6. I define a symbol in terms of its structure and function. First, a symbol can only exist in the context of a living organism or its artifacts. Life originated with symbolic memory, and symbols originated with life. I find it gratuitous to use the concept of symbol, even metaphorically, in physical systems where no function exists. Symbols do not exist in isolation but are part of a semiotic or linguistic system (Pattee, 1969a). Semiotic systems consist of (1) a discrete set of symbol structures (symbol vehicles) that can exist in a quiescent, rate-independent (non-dynamic) states, as in most memory storage, (2) a set of interpreting structures (non-integrable constraints, codes), and (3) an organism or system in which the symbols have a function (Pattee, 1986). There are innumerable symbol functions at many hierarchical levels, but control of construction came first.
7. Polanyi (1968) could be included with other non-reductionists (note 3). However, I believe the logic of his argument against reductionism can be made from within normal physical theory and does not require postulating some rejection, modification, or extension of physical language or concepts. He claims only an emergent hierarchy of physical "boundary condition" structures obeying all the normal laws, but acquiring complexity that needs new observables and new models. Emergence, a concept once shunned by philosophers and physicists, is now accepted in artificial life, complexity theory, neuroscience, philosophy, and physics (e.g., Ray, 1992; Cariani, 1992; Anderson, 1972, 1997).
8. The explicit recognition that these conditions are an essential requirement for objectivity arose only gradually, culminating in this century with what are now called invariance or symmetry principles (e.g., Wigner, 1964). All candidate theories must first conform to these principles. Ideally, objectivity is the belief that exactly the same events would occur whether or not they were actually observed. Except for the very rare "quantum non-demolition" case, this ideal cannot be reached since measurement in quantum theory alters the state of the system (but not the laws). Postmodern philosophers argue that this ideal of objectivity is also unattainable because of cultural influences. Even the objectivity of fundamental particles has been criticized on culture-based grounds (e.g., Pickering, 1984). Of course many aspects of our physics theories are conventional social constructs, but other aspects appear inexorably objective and have withstood far more rigorous and severe challenges from within the skeptical and competitive physics community than have been offered by the social constructivists.
9. Some dynamic modelers believe that natural selection can occur as a purely dynamic process without the need of symbolic memory constraints. For example, Goodwin (1994) states: "What this makes clear is that there is nothing particularly biological about natural selection: it is simply a term used by biologists to describe the way in which one form replaces another as a result of their different dynamic properties . . . We could, if we wished, simply replace the term natural selection with dynamic stabilization, the emergence of the stable states in a dynamical system." Similarly, Kelso (1995) states: "Thus selection and self-organization go together like bread and butter. Indeed, the language of selection is in precisely the same terms as the underlying pattern dynamics." For the many reasons given above I do not agree that dynamics alone is a rich enough language to describe life. Natural selection in particular implies much more than a population distribution with a statistical dynamics. The units forming the population must replicate with control of their own individual variations.
Anderson, P. W., 1972, "More is different". Science 177, 393-396.
Anderson, P. W., 1997, "Is measurement itself an emergent property?" Complexity 3(1), 14-16.
Bohr, Niels, 1933, "Light and life". Nature 131, 421. Reprinted in Bohr, N., Atomic Physics and Human Knowledge (John Wiley, NY) p. 3.
Born, Max, 1969, "Symbol and reality", in: Physics in My Generation (Springer-Verlag, NY) pp. 132- 146.
Brooks, R. A., 1992, "Artificial life and real robots", in: Toward a Practice of Autonomous Systems, F. J. Varela and P. Bourgine (eds.) (MIT Press, Cambridge, MA) pp. 3-10.
Brooks, R. A. and Maes, P. (eds.) 1994, Artificial Life 4 (MIT Press, Cambridge, MA).
Burgers, J. M., 1965, Experience and Conceptual Activity (MIT Press, Cambridge, MA).
Cariani, P., 1992, "Emergence and artificial life", in: Artificial Life II, C. Langton, C. Taylor, J. Farmer, S. Rasmussen, (eds.) (Addison-Wesley, Redwood City, CA) pp. 775-797.
Cassirer, E., 1957, The Philosophy of Symbolic Forms, Vol. 3, The Phenomenology of Knowledge (Yale University Press, New Haven, CT).
Clark, A., 1997, Being There (MIT Press, Cambridge, MA).
Conrad, M., 1969, Computer experiments of the evolution of co-adaptation in a primitive ecosystem, PhD Dissertation, Biophysics Program, Stanford University, Stanford, CA.
Conrad, M. and Pattee, H. H., "Evolution experiments with an artificial ecosystem". J. Theoretical Biology 28, 393-409.
Conrad, M., 1990, "The geometry of evolution". BioSystems 24, 61-81.
Cowan, G. A., Pines, D., and Meltzer, D. (eds.), 1994, Complexity, Metaphors, Models, and Reality (Addison-Wesley, Reading, MA).
Delbrück, Max, 1949, The Connecticut Academy of Arts and Sciences 38, 173.
Eden, R. J., 1951, "The quantum mechanics of non-holonomic systems". Proceedings of the Royal Society (London) 205A, 583-595.
Elsasser, W. M., 1975, The Chief Abstraction of Biology (North-Holland, Amsterdam, The Netherlands).
Emmeche, C., 1994, The Garden in the Machine (Princeton University Press, Princeton, NJ) p. 158.
Fontana, W., 1992, "Algorithmic chemistry", in: Artificial Life II, C. Langton, C. Taylor, J. Farmer, S. Rasmussen, (eds.) (Addison-Wesley, Redwood City, CA) pp. 159-209.
Frauenfelder H. and Wolynes, P. G., 1994, "Biomolecules: where the physics of complexity and simplicity meet". Physics Today, February, 1994, pp. 58-64.
Gell-Mann, M., 1994, The Quark and the Jaguar (W. H. Freeman, NY) p. 134.
Goldstein, H., 1953, Classical Mechanics (Addison-Wesley, Cambridge, MA).
Goodwin, B., 1994, How the Leopard Changed Its Spots (Simon & Schuster, NY) p. 53.
Grene, M., 1974, The Knower and the Known (University of California Press, Berkeley, CA).
Haldane, J. B. S., 1927, "On being the right size", in: Possible Worlds and Other Essays (Chatto & Windus, London). Reprinted in: On Being the Right Size, J. M. Smith (ed.) (Oxford University Press, 1985)
Hertz, H., 1894, Die Principien der Mechanik in neuem Zusammenhange dargestellt (Leipzig). English translation: The Principles of Mechanics (Dover, NY, 1956).
Hoffmeyer, J. and Emmeche, C., 1991, "Code duality and the semiotics of nature," in: On Semiotic Modeling, M. Anderson and F. Merrell (eds.) Berlin: Mouton de Gruyter, pp.117-166.
Holland, J., 1995, Hidden Order (Addison-Wesley, Reading, MA).
Kauffman, S., 1993, The Origins of Order (Oxford University Press) pp. 25, 36.
Kendrew, J. C., 1967, review of Phage and the Origins of Molecular Biology. Scientific American 216, 142.
Kelso, J. A. S., 1995, Dynamic Patterns (MIT Press, Cambridge, MA) p. 183.
Kull, K., 1998, "On semiosis, Umwelt, and semiosphere." Semiotica 120, 3-4, 299-310.
Langton, C., 1989, "Artificial life", in: Artificial Life, C. Langton (ed.) (Addison-Wesley, Redwood City, CA) p. 15.
Neimark, I. I. and Fufaev, N. A., 1972, The Dynamics of Non-holonomic Systems (Translations of mathematical monographs, v. 33, American Mathematical Society).
Mayr, E., 1997, This Is Biology (Harvard University Press, Cambridge, MA) p. 32.
Pattee, H. H., 1961, "On the origin of macromolecular sequences". Biophysical Journal 1,683-710.
Pattee, H. H., 1965, "Recognition of heritable order in primitive chemical systems", in: The Origins of Prebiological Systems, S. Fox (ed.) (Academic Press, NY).
Pattee, H. H., 1967, "Quantum mechanics and the origin of life". J. Theoretical Biology 17, 410-420.
Pattee, 1969a, "How does a molecule become a message?" Developmental Biology Supplement 3, 1- 16.
Pattee, H. H., 1969b, "Physical problems of heredity and evolution", in Towards a Theoretical Biology 2, C. H. Waddington (ed.) (Edinburgh University Press) p. 274.
Pattee, H. H., 1982, "Cell psychology: An evolutionary approach to the symbol-matter problem". Cognition and Brain Theory 5(4), 325-341.
Pattee, H. H., 1986, "Universal principles of measurement and language functions in evolving systems", in: Complexity, Language, and Life, J. L. Casti and A. Karlqvist (eds.) (Springer- Verlag, Berlin) pp. 579-581.
Pattee, H. H., 1995, "Evolving self-reference: matter symbols, and semantic closure". Communication and Cognition Artificial Intelligence 12(1-2), 9-27.
Pearson, K., 1937, The Grammar of Science (J. M. Dent and Sons, London) p. 287. [First editions published 1892.]
Pickering, A., 1984, Constructing Quarks: A Sociological History of Particle Physics (Edinburgh University Press).
Polanyi, M., 1964, Personal Knowledge (Harper, New York).
Polanyi, M., 1968, "Life's irreducible structure". Science 160:1308-1312.
Ray, T., 1992, "An approach to the synthesis of life", in: Artificial Life II, C. Langton, C. Taylor, J. Farmer, S. Rasmussen (eds.) (Addison-Wesley, Redwood City, CA) pp. 371-408.
Rosen, R., 1991, Life Itself (Columbia University Press, NY).
Rosen, R., 1987, "On the scope of syntactics in mathematics and science: the machine metaphor", in: Real Brains, Artificial Minds, J. L. Casti and A. Karlqvist (eds.) (North-Holland, NY), pp. 1- 23.
Schuster, P., 1994, "How do DNA molecules and viruses explore their worlds?", in: Cowan, G. A., Pines, D., and Meltzer, D. (eds.), 1994, Complexity, Metaphors, Models, and Reality (Addison-Wesley, Reading, MA) pp. 383-414.
Shear, J., 1997, Explaining Consciousness: The Hard Problem (MIT Press, Cambridge, MA).
Sommerfeld, A., 1956, Mechanics. (Academic Press, NY) p. 80.
Stent, G. S., 1966, in: Phage and the Origin of Molecular Biology, J. Cairns, G. S. Stent, and J. D. Thompson (eds.) (Cold Springs Harbor Laboratory on Quantitative Biology, 1966) p. 4.
Taylor, J. G., 1999, The Race for Consciousness (MIT Press, Cambridge, MA).
Varela, F. J., Thompson, E., and Rosch, E., 1991, The Embodied Mind (MIT Press, Cambridge, MA).
von Neumann, J., Theory of Self-reproducing Automata, A. W. Burks (ed.) (University of Illinois Press, 1966).
von Neumann, J., 1955, The Mathematical Foundations of Quantum Mechanics (Princeton University Press, Princeton, NJ)
Waddington, C. H. (ed.), 1968-72, Towards a Theoretical Biology, vol. 1, Prolegomena, 1968; vol. 2, Sketches, 1969; vol. 3, Drafts, 1970; vol. 4, Essays, 1972, Edinburgh: Edinburgh University Press., 1970; vol. 4, Essays, 1972, Edinburgh: Edinburgh University Press.
Weyl, H., 1949, Philosophy of Mathematics and Natural Science (Princeton University Press) p. 237.
Wigner, E. P., 1960, "The unreasonable effectiveness of mathematics in the natural sciences". Communications on Pure and Applied Sciences 13, 1-14.
Wigner, E. P., 1961, "On the impossibility of self-replication", in: The Logic of Personal Knowledge (Kegan Paul, London) p. 231.
Wigner, E. P., 1964, "Events, laws, and invariance principles". Science 145, 995-999.
Wigner, E. P., 1982, "The limitations of the validity of present-day physics", in: Mind in Nature. Nobel Conference XVII (Harper & Row, San Francisco, CA) p. 119.
Whittaker, E. T., 1944, A Treatise on the Analytical Dynamics of Particles and Rigid Bodies, 4th ed. (Dover, NY) p. 33 and Chapter VIII.