Those interested in the relationship between environment structure and behavior--the topic of this special issue of Adaptive Behavior--will find much of value in Peter Godfrey-Smith's new book, Complexity and the Function of Mind in Nature (hereafter CFMN; all page citations are to CFMN unless otherwise indicated). The writing is clear and concise, aptly balancing precision and breadth, and a host of relevant issues are raised and advanced. Although my comments here will focus only on the book's fundamental conceptual framework for how organisms relate to their environments, I enthusiastically recommend the entire book.

My primary concern is the extent to which organisms construct their environments. Some definitions from CFMN will help us pursue this topic. Externalist explanations are "explanations of properties of organic systems in terms of properties of their environments", while internalist explanations explain "one set of organic properties in terms of other internal or intrinsic properties of the organic system" (p. 30). Constructivist explanations (the converse of externalist explanations) are "explanations of environmental properties in terms of properties of an organic system" (p. 30). To insure that constructivism and externalism are compatible, Godfrey-Smith adheres to a narrow conception of constructivism which requires that "an organic system changes or determines the intrinsic properties of objects external to it" (Godfrey-Smith, this issue, p. XXX; see also CFMN, pp. 145f). In other words, for an organic system to construct its environment in this narrow sense, it is not sufficient that the system's faculties determine what constitutes its environment; more than this, the organic system must actually intervene causally in the external world. This narrow conception of constructivism allows Godfrey-Smith to make a sharp contrast between, on the one hand, an organism constructing its environment and, on the other hand, an organism changing itself rather than its environment and so merely accommodating its environment (p. 147). (Hereafter I will always use "construct" and its cognates in Godfrey-Smith's preferred narrow sense.)

Classifying explanations into these categories involves some subtleties. For one thing, although the definitions might suggest that the distinction between externalist and internalist explanations is dichotomous, Godfrey-Smith is clear that the distinction defines the poles of a continuous range of positions. The explanations of most organic systems invoke both internal and external factors (p. 51), so the degree to which an explanation is externalist or internalist depends on "[t]he amount of emphasis placed on the two types of factors", and this "can vary continuously" (p. 53). More important for present purposes, the externalism/internalism distinction is also context-dependent. There are many features of systems that we might seek to explain, and we must independently assess the relative explanatory weight of external and internal factors for each such feature. Thus, the external/internal polarity with respect to a single system is multidimensional. Similarly, the fate of externalism or internalism can depend on our perspective on the system being explained. Godfrey-Smith illustrates this perspective switch with evolutionary game theory; the explanation of the behavior of any individual in the population will be externalist, but the explanation of the population as a whole will be internalist (p. 49).

Godfrey-Smith's typology of explanations often provides no simple or univocal answer about whether a given explanatory project supports externalism or internalism. I count this as a virtue, on the grounds that the typology allows us to do justice to the complexities of intellectual life. For example, it allows us to identify precisely a characteristic and distinctive feature of explanations in the new interdisciplinary field known simply as "complexity", especially that branch of it sometimes called "artificial life". These fields tend to focus on systems that have at least two levels of description and explanation: a macro level, consisting of a large population of interacting agents, and a micro level, consisting of the agents themselves. The behavior of the macro-level population is simply the aggregate of the behaviors of all the micro-level agents which constitute it. The behavior of each micro-level agent is based on selective information about its own local environment, and its behavior directly affects only its own local environment. The goal of much work in these fields is to explain how macro-level populations exhibit certain kinds of complex organization and behavior. In artificial life, for example, one central question is to explain how a population could produce open-ended evolution. Much of this work explores whether these macro-level phenomena are (mainly) due to intrinsic features of the population&endash;an internalist research program, at least at the macro level. However, this macro-level internalism results from micro-level externalist processes. In particular, the typical pattern of explanation is that externalism at the micro level have effects which cause constructionism at the micro level, and these two factors then generate internalism at the macro level.

The behavior of a cellular automaton nicely illustrates this explanatory pattern; see, for example, Langton (1992). I choose this example partly because it shows the mechanisms so vividly, but also to counteract the impression given in CFMN (pp. 42f) that cellular automata research, and work in the fields of complexity and artificial life in general, tends to be anti-externalist. A cellular automaton is a lattice with a finite-state automaton or "cell" residing at each lattice site. The internal state at time t of a given cell, A, is determined by the internal states at time t-1 of the cells in a finite neighborhood of sites adjacent to A. In other words, since the environment of a given cell consists of the cells in its local neighborhood, each cell's internal state at each moment is explained by the properties of its local environment at the preceding moment. Now, consider how we are to explain some self-organizing global behavior of some cellular automaton; for example, a complex pattern of local structures that persists for extremely long periods of time, more or less regardless of the system's initial condition. If we take a "macro" stance and consider the organic system to be the whole population of cells, then the pattern of explanation will be internalist since nothing external to this system affects its behavior. At the same time, though, if we take a "micro" stance and consider each cell to be an organic system, then the system's behavior (internal state) is explained by its environment (the internal states of the neighboring cells). This is exactly the externalist pattern of explanation. Furthermore, notice that the cells in local neighborhoods interact. These neighborhoods are symmetric: cell A is in the neighborhood of cell B if and only if B is in the neighborhood of A. So, just as A's state at time t affects B's state at time t+1, B's state at time t+1 affects A's state at time t+2. Thus, when A affects the internal state of B and of the other automata in its neighborhood, this subsequently affects the internal state of A itself; that is, A changes the intrinsic properties of its own environment. This is the paradigm of a constructivist explanation. Furthermore, since the cellular automaton's self-organizing global behavior is nothing more than the aggregation of the behavior of the cells that constitute it, the global behavior arises because of the way in which the cells' internal states both depend on and affect the internal states of the cells in their local neighborhoods. Internalism at the macro level is due to externalism and constructivism at the micro level.

Much other work in the field of complexity, and especially artificial life, displays a closely analogous explanatory pattern. When Lindgren (1992), for example, presents evidence that open-ended evolution can be an intrinsic property of certain evolving populations, his explanation is internalist at the macro level. At the same time, the environment affects the behavior of the micro-level entities in the population, and the environment of a given micro-level entity is constituted by the other micro-level entities in the population. Thus, micro-level externalism and constructivism generate macro-level internalism. Analogous conclusions holds for Kauffman's (1993) explanations of evolutionary self-organzation, and Holland's (1992) and Ray's (1992) explanations of the emergence of complex ecologies, to pick just a few more examples. So, this work is neither predominantly internalist nor predominantly externalist; both kinds of principles operate in a way that connects the micro and macro levels of organization. It is true that from one perspective the ultimate concern of this work looks internalist&endash;showing how properties like self-organization and open-ended evolution are intrinsic properties of certain kinds of systems. But this signals no pervasive sympathy with internalism, because the mechanisms depend on externalist explanations at the micro level.

The foregoing explanation of macro-level internalism highlights a general principle about how behavior adapts to environmental structure: specifically, that interactions among adapting organisms are a powerful and ubiquitous mechanism by which organisms construct their own environments. A combination of factors produce this mechanism. First, organisms adapt their behavior to their environment; this is externalism. Next, the organisms interact with each other in various ways (more on this in a moment). Through these interactions the organisms both constitute and, thus, change a significant portion of each other's environments. Finally, when one organism changes the environment of neighboring organisms, those neighboring organisms will adapt their behavior to those environmental changes. Since those neighboring organisms constitute a significant part of the original organism's environment, the original organism's adaptive behavior has effects which change its own environment; this is constructivism. Complex macro-level self-organization can arise spontaneously because micro-level externalism combined with micro-level interaction generate micro-level constructivism.

This mechanism is well known, of course, but its scope and power can be underestimated, especially when the micro-level processes involved are indirect. To get a sense of how widespread the mechanism is, consider an extreme case: whether an organism changes the intrinsic properties of its environment simply by moving from one location to another. Considered apart from its effects, an organism's movement is just a spatial dislocation which does not change the physical properties of the organism's external environment. Moving one's self around in the environment contrasts with changing other (non-self) portions of one's environment. As Godfrey-Smith points out (p. 149), an organism's movement affects which part of the world constitutes its environment but this does not by itself amount to a change in the intrinsic properties of that environment. Thus, if we ignore subsequent effects, an organism's movement does not construct its environment. At the same time, however, because organisms interact, one organism's mere movement can cause other organisms to change, and this does change the intrinsic properties of (i.e., construct) the first organism's environment. Of course, when organisms construct their environments, they typically do so by moving; beavers build dams and spiders spin webs by directly manipulating their physical surroundings. What might be less obvious, though, is the ease with which organisms can still construct their environments even if they do not directly alter any of the physical features of those environments. One way to construct the environment indirectly is to send a signal that causes other animals to change the physical structure of the environment. Krebs and Dawkins (1984), for example, discuss the different kinds of coevolution involved in cooperative and non-cooperative communication, and how both are influenced by environmental constraints on signal detectability and discriminability. A much simpler and more pervasive method of indirect environment construction, though, is the micro-level mechanism we have been discussing. That is, given even very limited interactions among organisms, an organism's mere change of location can have effects which change its environment.

Precisely this mechanism is evident in an extremely simple version of the familiar agent-based models studied in artificial life. Consider a population of agents which compete for food in a finite two-dimensional world. The food is sparsely distributed throughout the world, and its location and quantity never changes (food is immediately replenished after being eaten by an agent). The agents use local sensory information and a genetically-encoded sensorimotor strategy to move in search of food. At any given moment only one agent can occupy a given site in the world. If an agent tries to move to a site that is already occupied, it moves to the nearest unoccupied site. The agents can sense only the food in the local neighborhood, not nearby agents. And their behavioral repertoire is limited to moving and eating; agents cannot directly interact. One might think that these agents cannot construct their environment. After all, they cannot change the food distribution and they cannot directly interact with other agents, and there is nothing else in this simple world. In fact, one might invent a model like this with the intention of creating a situation in which organism cannot construct their environment (if, for example, one sought to learn how adaptation depends on spatiotemporal structure in static environments). However, we should notice that an agent's environment includes the locations of nearby agents, for this determines which sites are free to be occupied. So, agents can interact indirectly through their movement, by controlling which sites other agents can occupy. Furthermore, this is a relevant part of the environment; an agent needs food to survive and the food's spatial distribution is fixed, so which sites an agent can occupy is a matter of life or death. Thus, an agent's movement changes its neighbors' environments. And this relationship is symmetrical; when the movement of agent A affects the environment of agent B, this then affects B's movement, which in turn affects A's environment. Thus, even this very narrow and almost invisible form of interaction is enough to enable an agent's mere movement to trigger a genuine change in the intrinsic properties of its own environment. Of course, artificial and natural agents often directly change the physical properties of their environment, that is, directly construct their environment. (One common form of direct environmental construction occurs when organisms change the spatial distribution of food in their environment by eating it; see, for example, Packard (1989), Bedau and Packard (1992), and Todd, Wilson, Somayaji, and Yanco (1994). Organisms can directly construct their environments in other ways, of course; see, for example, Kirsh, this issue. Bedau (1994) has proposed one general method for measuring the extent to which an adapting population changes the structure of its environment.) But even when organisms cannot directly change their environment, limited and indirect interactions among organisms can be enough to cause environmental construction indirectly. If organisms are doing nothing more than merely moving about without directly altering any of the intrinsic properties of their environment, they will still almost inevitably end up constructing their environment.

This conclusion is striking. If limited and indirect organism interactions are enough to give all of an organism's mere movements the power to change the environment, then constructivist explanations should be quite prevalent. This conclusion is amplified when we consider the myriad interactions that connect organisms, such as mating, various forms of competition and cooperation, including predator-prey and host-parasite relations, animal communication. In fact, the conclusion generalizes to those organic changes that are mere accommodations of the environment. A virtue of Godfrey-Smith's narrow conception of an organism constructing its environment is is its sharp contrast with the notion of an organism merely accommodating its environment by changing only itself. But given a network of interactions among organisms, an organism's accommodation to its environment can typically be expected to have effects which change the intrinsic properties of the external objects that constitute the organism's environment. Readers of CFMN might miss this conclusion, for the book's discussion of kinds of construction (ch. 5) tends to ignore the interactions among organisms and their resulting co-constitution of environments. In addition, the specific models discussed in CFMN (Part II) play down these interactions. For example, the model of phenotypic plasticity (pp. 209-216) treats an organism's environment as fixed for all time and assumes, in effect, that each organism confronts this environment in isolation from all other organisms. Clearly, then, this model leaves no room for interactions among organisms and the dynamic construction of environments that this implies. The natural way to study the effects of these interactions is with the sort of agent-based models familiar from artificial life. Because the systems of interest involve a population of entities "undergoing a kaleidoscopic array of simultaneous nonlinear interactions", as Holland puts it (1992, p. 184), analytically solvable mathematical models reveal little about the effects of the complex web of interactions binding organisms. Godfrey-Smith's sharp distinction between accommodation and construction of environment is valuable in part because it helps us discern how easily mere accommodation can cause construction. This, in turn, helps us to appreciate the significant extent to which organisms construct their environments.

Acknowledgements. Thanks to Peter Todd for useful comments on the substance and form of this note.

Bedau, M. A. (1994). The evolution of sensorimotor functionality. In P. Gaussier, & J. -D. Nicoud (Eds.), From Perception to Action (pp. 134-145). Los Alamitos, CA: IEEE Computer Society Press.

Bedau, M. A., & Packard, N. H. (1992). Measurement of evolutionary activity, teleology, and life. In C. Langton, C. Taylor, J. D. Farmer, & S. Rasmussen (Eds)., Artificial Life II (pp. 431-461). Reading, MA: Addison Wesley.

Holland, J. H. (1992). Adaptation and control in natural and artificial systems: An introductory analysis with applications to biology, control, and artificial intelligence (Rev. ed.). Cambridge, MA: MIT Press/Bradford Books.

Kauffman, S. A. (1993). The origins of order: Self-organization and selection in evolution. New York: Oxford University Press.

Krebs, J. R., Dawkins, R. (1984). Animal signals: Mind-reading and manipulation. In J. R. Krebs and N. B. Davies (Eds.), Behavioral Ecology: An Evolutionary Approach (2nd ed.). Oxford, England: Blackwell Scientific Publications.

Langton, C. G. (1992). Life at the edge of chaos. In C. Langton, C. Taylor, J. D. Farmer, & S. Rasmussen (Eds.), Artificial Life II (pp. 41-89). Reading, MA: Addison Wesley.

Lindgren, K. (1992). Evolutionary phenomena in simple dynamics. In C. Langton, C. Taylor, J. D. Farmer, & S. Rasmussen (Eds.), Artificial Life II (pp. 295-312). Reading, MA: Addison Wesley.

Packard, N. H. (1989). Intrinsic adaptation in a simple model for evolution. In C. Langton (Ed.), Artificial Life (pp. 141-155). Reading, MA: Addison Wesley.

Ray, T. (1992). An approach to the synthesis of life. In C. Langton (Ed.), Artificial Life (pp. 371-408). Reading, MA: Addison Wesley.

Todd, P. M., Wilson, S. W., Somayaji, A. B., & Yanco, H. A. (1994), The blind breeding the blind: Adaptive behavior without looking. In D. Cliff, P. Husbands, J. -A. Meyer, & S. W. Wilson (Eds.), From Animals to Animats 3: Proceedings of the Second International Conference on Simulation of Adaptive Behavior (pp. 228-237). Cambridge, MA: MIT Press/Bradford Books.