Information theory has been developed from "communications theory" and "physical measurement theory" perspectives. These perspectives overlap in asserting that information is (1) anything transmitted from a "source" through a "channel" to a "receiver" and (2) an abstraction rather than a material part of the system. In classical communications theory, the amount of information sent from a source is calculated using a statistical entropy function. Errors in transmission resulting from poor encoding at the source or from noise in the transmission channel reduce the proportion of the transmitted information which is actually recorded by the receiver. All processes affecting transmission and reception of information thus decrease its entropy from its maximum at the source. Because physical entropies increase as a result of work done on the system, the communications view of entropy is generally considered non-physical.
Physical measurement theory provides a second formalism. Brillouin (1962) distinguished "free information", an abstraction involved in descriptive exercises, and "bound information", referring to material properties of the system (although information is not a material part of the system per se). Bound information is determined with respect to the microstates of the system. Hence, it is also calculated using a statistical entropy function but, contrary to communications theory, is expected to exist only in systems for which there is a non-arbitrary microstate/macrostate distinction. Bound information is defined as
I = Hmax - Hobs
where Hmax refers to the totally relaxed state of the system (usually estimated by a randomization of the observed components of the system). Brillouin defined I as "negentropy", which is converted into bound information by measurement(measuring devices are receivers), so negentropy = information. Information thus has a physical basis, but is not a material part of the system.
Brooks and Wiley (1988) concluded that there must be an additional conception of information for biology because biological information (based in the nucleic acids that comprise genes) has both communication functions and a physico-chemical basis. The conception must also account for the growth of information through time, manifested through phylogenetic diversification. What was needed, therefore, was an account of biological information that (1) was physically realistic (has an objective material basis), (2) was intrinsic to the system rather than to devices for measuring the system, and (3) could grow spontaneously over time. Two basic issues with respect to information and entropy are (1) whether information can be a material part of a system rather than just an abstract representation and (2) whether or not there is an objective difference between macrostates and microstates in calculations of informational entropies, the formal basis for representing information.
The new view of biological information is related to concepts about a system's causal capacity, or its ability to impose distinctions on its surroundings. This shifts the emphasis from the manner in which measuring devices are affected to how the system produces effects on measuring devices. Collier (1986, 1988, 1990; also Brooks et al., 1989; Banerjee et al., 1990) proposed that material information systems occur as arrays, or multi-dimensional messages, in which macrostate/microstate distinctions are distinguished non-arbitrarily. For this information to be related to physical concepts there must be (1) a physical (material) basis for the information, (2) an energetic cost in producing the information, and (3) a real (non-arbitrary) macrostate/microstate distinction. The discovery of the chemical structure and function of DNA provided the material basis for biological information, satisfying (1) above.
Energy dissipated within the system as a result of work done (including heat-generating transformations) is intropy, or internal entropy (Ulanowicz, 1986). Energy that is converted into structure (conservative transformations) is enformation, or intrinsic information (Collier, 1990). All conservative processes are coupled with heat-generating processes, so there is an energetic cost associated with producing and maintaining biological information (Maurer and Brooks, 1991). Intropy and enformation are inter-convertable (e.g., energy from the surroundings can be converted into structure, say glycogen, which can be converted into heat, then dissipated from the system). Intropy is converted into enformation by cohesive properties of the system. Cohesive properties, ranging from molecular affinities, to cell-cell adhesion, to genetic compatibility, mate recognition, and genealogy, also provide resistance to fluctuations from lower levels of organization, allowing macroscopic properties to emerge. The major transitions in evolution discussed by Maynard Smith and Szathmary (1995) are associated with the evolution of new forms of cohesion, which permit information to be stored and transmitted more efficiently.
Macrostate/microstate distinctions are determined objectively by part/whole associations. The number of accessible microstates is increased by production of new components, either at a given level or through the opening up of new organizational levels. Biological systems accomplish this by conservative transformations. For example, auto-catalytic processes producing monomers make "monomer space" available for molecular evolution. Some monomers have high chemical affinities for each other, and will spontaneously clump into dimers and polymers. Once polymers begin to form, "polymer space" becomes available to the evolving system. At this level, polymers are macrostates and monomer and dimer distributions are microstates. Causal interactions among polymers create new levels of organization in which polymer distributions are the microstates and new levels of organization are the macrostates, and so on. Each new functional level creates a hierarchy of increasing structural intricacy, manifested by increasing allocation of the entropy production in structure. A protein coding unit might be considered a macrostate, while all the actual coding sequences for that protein would be the microstates; a locus could be a macrostate, and all alleles corresponding to that locus the micro-states; phenotypes could be macrostates, and all genotypes corresponding to a given phenotype would be microstates (Layzer, 1978, 1980; Collier, 1986; Wiley, 1990). Encoded information is also the carrier of the cohesive properties, so production of biological information involves simultaneous production of variation and constraints, ensuring that the flow of biological information through time will be a combination of persistence and change.
According to the above view, organisms are material information systems. Relating this to the question of biological signals requires that we re-visit the basic formalism of information theory. Gatlin (1972) proposed that the genetic system is the source of information (sender), reproduction and ontogeny the channel through which information is transmitted, and the environment is the receiver. Natural selection became the form of the environment that determines the meaning of the information. Genetic data thus become phenotypic signals as a result of reproduction and ontogeny, and become biological information as a result of causal interactions between the phenotype and the environment. Brooks and McLennan (1990) noted that the environment cannot be a receiver in a physical sense, because it only eliminates biological information; it does not actually measure or interpret the information. Because biological systems are localized in both time and space, the receiver could be a "time" rather than a "place". In their formalism, the source is a genealogical system at time t0, and the channel is reproduction and ontogeny, and the receiver is the source genealogical system at time t1...n. If the source precedes the receiver in time, it can produce the system that acts as receiver, which can then become a source itself (Csanyi, 1989; Matsuno, 1989). A similar perspective has been used in designing self-correcting computer programs. This reinforces the biological analogy, because DNA has significant self-repair capabilities and sexual reproduction may enhance those capabilities (Bernstein et al., 1989). Can this view of the nature of biological systems be extended to explain signals produced by them?
Some internal signaling systems produce effects in the organism that are apparent to other organisms (observer, receiver) in the environment. Such external signals are by-products of internal-state signaling, and thus are not produced intentionally as signals sent to specific organisms. They exist in the absence of any external receiver because they emerge from a situation in which the signaler is the receiver. In this sense, self-signaling is entirely consistent with the view of biological information systems elaborated above. A by-product (emergent property) of these effects may be predictable responses by other organisms. Just as the effects of internal signaling in the sender may have external manifestations, the ability to receive a particular signal, and the meaning imputed to it, are properties of the receiving organism's internal signaling system, and do not originate intentionally. We know of no case in which organisms have evolved mechanisms specifically for receiving a particular signal from their surroundings prior to the origin of that signal.
Consider cases in which non-intentional manifestations of internal signaling are perceived by other organisms. There are three classes of possible outcomes. First, the effect may evoke no response by the observer. In practice it may be difficult to distinguish a case in which the organism perceives a signals and rejects the information (does not respond) from a case in which the organism does not perceive the signal (Gerhardt et al., 1994; McLennan and Ryan, submitted). For example, the sight of an antelope will evoke a hunting response in hungry lions, virtually no response in lions that have just fed, and no response from a daisy. Second, the effect may produce aggregating responses by the receiver that are positive to both sender and receiver (e.g., they represent potential mates), or positive to one but not the other (one may represent food to the other). And third, the effect may produce avoidance responses by the receiver.
In such cases the sender is not sending a message to any particular receiver-it is unintentionally broadcasting. The receivers in the environment are not intentionally receiving a particular signal. Nonetheless, the perception of the unintentionally broadcast signals produces an additional level of selection, one in which receivers become agents of selection. The fate of the organisms unintentionally producing a signal is determined by a complex balance of costs and benefits associated with the three response categories listed above. So long as the total benefits outweigh the total costs to the organism producing the effect (thesender), the signal will persist through continued reproduction, thus having a net cohesive influence on the genealogical system. Organisms are simultaneously in the environment and part of the environment (Maynard Smith, 1976), so it is only a small step to move from the emergence of mutual signaling systems -- produced unintentionally -- to a coevolutionary dynamic in which another level of cost-benefit considerations emerges, one involving enhanced cohesion and functional integration accompanied by more constraints through mutual inter-dependence.
Organisms are thus at once signals themselves, in the sense of being signal senders, and signal-bearers, in the sense that some of their functional and structural attributes may be perceived by other organisms in their surroundings. If external signals are not sent intentionally, but are by-products of internal signaling, then the effects of such 'internal conversations' may have more than one meaning to observers in the environment. The evolutionary fate of signals (and the signal bearers) will thus be an outcome of all the ways in which other organisms perceive the signal bearer. In addition, if the signals that selection can act upon directly at this level are only a subset of all the internal signaling, which selection can affect only indirectly, we must assume that various selection pressures, operating on different levels of biological organization, may themselves conflict to some extent. For example, consider an organism which changes color when it eats a particular type of high-energy food. The change in color is an unintentional by-product of an internal signal of well-being (more energy coming into the organism's metabolic system) that may also make the organism highly visible to predators. Selection will simultaneously favor enhancing efficiency in obtaining high energy food and minimizing external manifestations of that success.
A male stickleback fish is ready to mate, and signals his readiness. Did the desire (an intention to mate) emerge and somehow cause the signal to appear? Or did signal and intention emerge as a result of the same non-intentional changes in photo-period and water temperature that produced physiological and hormonal changes in the fish, resulting in the emergence of the intention to mate and the signaling of those intentions? Is intention intentional or is it a by-product (emergent property) of other things? As with the case of mimicry, is not enough to ask how efficient or how unambiguous a signal is in a particular context. Death is surely unintentional in most cases, and yet it is a highly unambiguous signal to myriads of organisms whose lives depend on decomposing dead organic matter.
If we accept the proposition that signals emerged non-intentionally, then a sender cannot control the way in which a signal is interpreted. The male stickleback's signal to conspecific females that he is ready to mate is also a signal to himself of his having ingested a substantial quantity of high energy food, and is also a signal to predators in the environment that he is edible. He has eaten high energy food, and that fact is manifested in a change in his body color. Metaphorically, he intends to mate, and his intention is signaled in the same change in body color; he does not intend to be eaten, but he signals to some receivers that he is edible.
In a world in which receivers evolve independently of the desires and intentions of senders, intention may not be significant. Intentionality is casually important only when the desire precedes, and is causally responsible for the construction of, the signal of that desire. Such intentionality is teleological, and thus in a Darwinian rather than Lamarckian world restricted to only some forms of human signaling activity. This may be one reason human experience gives lesser credence to intentions than to actions ('The road to Hell is paved with good intentions').
Consider different observers, each with a different construction and constitution, including ways in which their surroundings are monitored. They may perceive the signaler in different ways (i.e. they may perceive different attributes-no organism is capable of perceiving all the signals in its surroundings), they may perceive the same feature of the signaler in different ways (e.g. different photo-receptors have different spectral sensitivities), and they may respond to the same signal in different ways (e.g. a female stickleback will try to mate, a heron will try to eat). These responses are not dictated directly by the sender, so the question of meaning is a question of the properties of the observers ("beauty is in the eye of the beholder"). It is in the physico-chemical nature of organisms to assess themselves ('How am I? I'm hungry.') and their surroundings ('There is a stickleback, that means food to me. I'll eat him.') through elaborate self-signaling. The hunter did not 'intend' to seek a red signal in order to eat. The stickleback did not intend to send a signal meaning'I'm edible, come eat me'. But the temporal and spatial conjunction of two different autonomous systems ("sender" and "receiver"), each self-absorbed and self-preoccupied, led to a predictable outcome.
The possible outcomes of all interactions create an array of possible meanings for any signal projected into the environment. All meaning in this sense is contextual and thus contingent. The range of meanings, and the richness of meanings, will be proportional to the complexity of the surroundings and the diversity and sophistication of the sensory systems.
It is not a signal unless it is transmitted from one organism to another:
This perspective rules out cases of internal signaling, depriving the explanatory construct of any way to explain the origin of signals sent into the environment without invoking intentionality. It also creates a paradox because there is no way for organisms to receive intentionally-produced signals without internal signaling systems providing the organisms with information about their surroundings.
It is not a signal unless it has been designed by natural selection for a particular purpose:
This perspective mistakenly bestows natural selection with creative powers. As in the first case, there is no way for external signaling to emerge (e.g. from internal signaling) in a non-intentional manner. In this case, however, intentionality is transferred from the sender to an abstract causal agent, known only by inference and in retrospect (arguing from effects to cause). It requires that both the sender and the receiver evolve simultaneously, and for the specific purpose of exchanging a particular signal between them; otherwise selection could not act on the function of the signal, which emerges from the interaction between sender and receiver. If the signal has been designed for a particular purpose, how can we explain the evocation of different responses in different receivers, and different responses in the same receiver? There are only two possible answers to this question: (1) natural selection is extraordinarily ineffective or (2) signals emerge in a complex world of multiple costs and benefits, where selection would favor the persistence of the attribute if the benefits outweigh the costs.
It is not a signal unless it conveys true information:
If the sender is responsible for the truth of information being sent, truth must be an objective and unambiguous quantity, something on which there is little human consensus. In addition, theremust be intentionality behind the broadcasting of signals. If truth is something assessed by the receiver, rather than imparted by the sender, then it need not be objective and unambiguous, since the same signal may convince one observer that an organism is truly a leaf mimic while convincing another organism that it is truly a leaf. When it comes to biological signals truth, like beauty, is in the mind of the beholder. Thus, any attribute of an organism that evokes a response in itself or another organism is, trivially, true for the receiving organism, according to the meaning the receiving organism places on the signal. Truth is associated with meaning more than with information.
It is not a signal until a receiver converts it into information:
This view invokes yet another form of intentionality. A signal cannot evolve without the sender intending to evoke a particular response in the receiver, implying that the receiver exists prior to the signal. This perspective does not equate information with truth, but it does equate information with meaning. It also treats information as an abstraction rather than as material. If this is the case, signals and the information they transmit have no causal basis and are irrelevant to evolutionary theory. If information has a material basis, however, it can evolve and be acted upon by evolutionary processes. In that case, information should be equated with signal, with meaning emerging from the interaction between the signal and the receiver.
Bernstein, H., F.A. Hopf, and R.E. Michod. 1988. Is meiotic recombination anadaptation for repairing DNA, producing genetic variation, or both? In The Evolution of Sex, ed. Michod, R.E. and B.R. Levin. p. 139-160. Sunderland Mass.: Sinauer Assoc.
Brillouin, L. 1962. Science and Information Theory. 2nd ed. New York: Academic Press.
Brooks, D.R., J. Collier, B.A. Maurer, J.D.H. Smith, and E.O. Wiley. 1989. Entropyand information in evolving biological systems. Biol. Philos. 4: 407-432.
Brooks, D.R. and D.A. McLennan. 1990. Searching for a general theory of biological evolution. J. Ideas 1: 35-46.
Brooks, D.R. and E.O. Wiley. 1988. Evolution as Entropy: Toward a Unified Theory of Biology. 2nd ed. Univ. Chicago Press, Chicago. 2nd. ed.
Collier, J. 1986. Entropy in evolution. Biol. Philos. 1: 5-24.
---. 1988. The dynamics of biological order. in Entropy, Information and Evolution: New Perspectives on Physical and Biological Evolution, p. 227-242, ed. B. Weber, D.J. Depew, and J. D. Smith. MIT Press, Cambridge.
Collier, J. 1990. Two faces of Maxwell's demon reveal the nature of irreversibility. Stud. Hist. Philos. Sci.
Csanyi, V. 1989. Evolutionary Systems and Society: A General Theory. Durham, N.C.: Duke Univ. Press.
Darwin, C. 1859. Origin of Species. London: J. Murray.
Gatlin, L.L. 1972. Information Theory and the Living System. New York: Columbia Univ. Press.
Gerhardt, H. C., M. L. Dyson, S. D. Tanner, and C. G. Murphy. 1994. Female treefrogs do not avoid heterospecific calls as they approach conspecific calls: implications for mechanisms of mate choice. Anim. Behav. 47: 1323-1332.
Krebs, J. R. and R. Dawkins. 1984. Mind-reading and manipulation. In Behavioural Ecology: An Evolutionary Approach. 2nd ed. p. 380-402. Sunderland, Mass.: Sinauer Mass.
Layzer, D. 1978. A macroscopic approach to population genetics. J. Theor. Biol. 73: 769-788.
Layzer, D. 1980. Genetic variation and progressive evolution. Amer. Nat. 115: 809-826.
Matsuno, K. 1989. Protobiology: Physical Basis of Biology. Boca Raton, Florida: CRC Press.
Maurer, B.A. and D.R. Brooks. 1991. Energy flow and entropy production in biological systems. J. Ideas. 2: 48-53.
Maynard Smith, J. 1976. What determines the rate of evolution? Amer. Nat. 110: 331-338.
Maynard Smith, J. and E. Szathmary. 1995. The Major Transitions in Evolution. Oxford: W.H.Freeman Spektrum.
McLennan, D.A. and M.J. Ryan. submitted. The effect of phylogenetic distance on mate recognition based on olfactory cues in swordtails. Behav. Ecol.
Ulanowicz, R.E. 1986. Growth and Development: Ecosystems Phenomenology. New York: Springer-Verlag.
Wiley, E.O. 1990. in Evolutionary Innovations. M.H. Nitecki, ed. Chicago: University of Chicago Press.
Deborah A McLennan is assistant professor, Department of Zoology, University of Toronto. Her speciality is the evolution of animal communication systems. She has done groundbreaking studies using stickleback fish and has demonstrated that the male nuptial signal was a complex mosaic of three colours, rather than a single colour, and that female sticklebacks exhibit nuptial colouration, making the breeding cycle of these fish a dialogue between partners rather than a male monologue, as had been assumed before. Her post-doctoral work demonstrated an olfactory system redundant with the visual system in the mating behaviour of swordtail fish. She is a pioneer in the area of comparative evolutionary studies using phylogenetic information to establish experimental protocols. In addition to more than forty scientific articles, Dr McLennan is co-author of Phylogeny, Ecology and Behavior: A Research Program in Comparative Biology (University of Chicago Press, 1991) and Parascript: Parasites and the Language of Evolution (Smithsonian Institution Press, 1993).