Towards the naturalization of agency based on an interactivist account of autonomy

This paper attempts to provide the basis for a broader naturalized account of agency. Naturalization is considered as the need for an ongoing and open-ended process of scientific inquiry driven by the continuous formulation of questions regarding a phenomenon. The naturalization of agency is focused around the interrelation of the fundamental notions of autonomy, functionality, intentionality and meaning. Certain naturalized frameworks of agency are critically considered in an attempt to bring together all the characteristic properties that constitute an autonomous agent, as well as to indicate the shaping of these notions/properties during the development and the evolution of its agential capacity. Autonomy and interaction are proved to be key concepts in this endeavor. 2009 Elsevier Ltd. All rights reserved. 1. What does it mean to naturalize? There are many different kinds of naturalism, but almost all of its adherents and especially those who advocate epistemological naturalism (viz. Quine, 1969, ‘‘Epistemology Naturalized’’, see e.g. Feldman, 2006), have as a common point the fact that they provide different answers from those of traditional epistemologists, to crucial epistemological questions such as the source of particular beliefs, when particular beliefs are valuable, and how to form such beliefs regarding a certain inquiry. As a result, explanation within naturalized epistemology is understood in a different sense than in traditional epistemology, on which inferences are justified by a priori (and often universal and ontologically valid) beliefs, and by observed data on the behavior of the system. [email protected] (T. Spyrou), [email protected] (J. Darzentas). d. All rights reserved. s, A. et al., Towards the naturalization of agency based on an s in Psychology (2009), doi:10.1016/j.newideapsych.2009.09.005 A. Arnellos et al. / New Ideas in Psychology xxx (2009) 1–16 2 ARTICLE IN PRESS Naturalization requires the justification of an explanation based on natural relations or interactions. Such an explanation is not just an observer’s adaptive strategy for interpreting the behavior of other systems, in terms of the observer’s beliefs and desires, as Kampis (1999) suggests, it is also an attempt to look inside the system and try to understand and explain how it works. This seems to be a valid strategy for naturalism, as in such cases, the respective explanations can be objectively verified. For this reason, as Etxeberria and Moreno (2001) argue, there is a beneficial influence from the ‘natural’ sciences to the ‘human’ sciences, where the benefit is advancement towards a better explanation of the phenomenon under investigation. One should also keep inmind that the history of science is not just amere accumulation of empirical data regarding the observed phenomenon that fits to the curves of an analytical model. More importantly, it also involves the radical transformation of the observer’s understanding regarding something that he has already observed. Faith (2000) describes these transformations as further naturalizations made possible by the discovery of new and different mechanisms underlying the respective phenomena. Since science is inherently progressive, the resulting explanatory principles and rules regarding those phenomena should not be bounded. On the contrary, they should bemapped onto this progressive nature and hence, naturalization has no end or a specific and discrete final state, but it is an ongoing and open-ended process of scientific inquiry. In other words, naturalization is the continuous formulation of questions regarding a phenomenon considering the quantitative but also the qualitative progress of science regarding notions and beliefs pertaining to this phenomenon, and aiming towards a better understanding and modeling of this phenomenon. Thus described, naturalization can be considered as the cornerstone of interdisciplinary research, a wider paradigm inwhich contemporary researchers and scholars should try, in general, to analyze and define in an open-ended manner the notion of agency and in particular, to understand and explain the complex cognitive phenomena inwhich an agent is involved. In such anaturalistic domain of explanation, agency may acquire many different scientific descriptions and explanations, as long as the theoretical notions used in these descriptions and explanations are naturalistically valid (with reference to contemporary scientific findings); as long they are not based on metaphysical assumptions and a priori judgments. For example, the well-known problem of intentionality and of the intentional behavior of an agent cannot be taken to be answeredwith a simple appeal to God or to somemysterious dark forceswith a metaphysical grounding. At present, this would not be classified as a naturalistic explanation of intentionality. On the other hand, agents act so as to reach out towards theworld and as Kampis (1999) argues, the acts and thoughts of an agent always have a target, an object, a referent, or in general, a state of affairs. This provides agency with a certain directedness, and the attempt to explain this directedness by postulating mental states appears to serve the purpose of naturalization well, if not well enough. An explanation based on the existence of neural mechanisms that somehow manage to be in a state that underwrites intentional behavior is a step forward towards amore naturalized explanation. But here, too, the process of naturalization remains incomplete, as this ‘magic’ state might well be immanent and independent of the environment of the agent, or simply emergent under certain contextual conditions. As we have claimed, the naturalization of agency requires the explanation of how an agent does something and hence, the naturalization of intentional behavior is not just a matter of what an agent does, but also, how it does it. Since agents seem to evolve within a dynamically changing environment, while they are always engaging in intentional interaction with it, more valid inquiry within the quest for naturalizing agency would likely be to analyze how the ongoing existence of an agent justifies their intentional behavior. Of course, at this point one may even ask if intentionality is the right property for a naturalized explanation of agency to begin with. Intentionality is one of the fundamental properties of an agent, but we will show, it is not the only one. In the next section, some familiar definitions of agency will be canvassed in an attempt to locate the proper starting point for a naturalized explanation. 2. Defining agency with a view towards naturalization Franklin and Graesser (1996) provide an interesting review of many characteristic definitions of the term ‘agent’. Some of them (those that seem closest to a stronger, living-system-centered notion of agency) are quoted below. Please cite this article in press as: Arnellos, A. et al., Towards the naturalization of agency based on an interactivist account of autonomy, New Ideas in Psychology (2009), doi:10.1016/j.newideapsych.2009.09.005 A. Arnellos et al. / New Ideas in Psychology xxx (2009) 1–16 3 ARTICLE IN PRESS Russell and Norvig (1995, p. 33) state that: ‘‘An agent is anything that can be viewed as perceiving its environment through sensors and acting upon that environment through effectors.’’ Maes (1995, p. 108) states that: ‘‘Autonomous agents are computational systems that inhabit some complex dynamic environment, sense and act autonomously in this environment, and by doing so realize a set of goals or tasks for which they are designed.’’ Hayes-Roth (1995) states that: ‘‘Intelligent agents continuously perform three functions: perception of dynamic conditions in the environment; action to affect conditions in the environment; and reasoning to interpret perceptions, solve problems, draw inferences, and determine actions.’’ B Wooldridge and Jennings (1995, p. 2) ‘‘Perhaps the most general way in which the term agent is used is to denote a hardware or (more usually) software-based computer system that enjoys the following properties: autonomy: agents operate without the direct intervention of humans or others, and have some kind of control over their actions and internal state. B Social ability: agents interact with other agents (and possibly humans) via some kind of agentcommunication language. B Reactivity: agents perceive their environment, (which may be the physical world, a user via a graphical user interface, a collection of other agents, the INTERNET, or perhaps all of these combined), and respond in a timely fashion to changes that occur in it. B Pro-activeness: agents do not simply act in response to their environment, they are able to exhibit goal-directed behavior by taking the initiative.’’ There are some interesting points in all these definitions. In the first place, all authors conceive and describe an agent in anthropomorphic terms, independently of the agent’s area of application (i.e. information system, robotic systems, web-interactive systems, etc.). They all speak of an agent as perceiving its environment, reasoning in order to interpret its perceptions and to draw inferences, acting in the environment, solving problems, communicating with other agents and thus, socializing, etc. It also seems that they all, either explicitly or implicitly, accept a kind of pro-activeness in an agent. And last, but not least, all these definitions ascribe self-rule and independence to the agent. There is nothing wrong with these features (although Maes seems to be unjustifiably willing to ascribe autonomy to a computational agent), but one should not forget that our present aim is to naturalize agency, so we must determine whether these are the most fundamental, or the most easily naturalized, of an agent’s properties. So far, most of the attempts to build artificial agents have been based on the view that concepts such as autonomy and properties such as perception and action are discrete modules of a bigger system and that if one manages to build and integrate these modules, agency will result. But concepts such as autonomy and pro-activity, and even ‘simple’ properties such as perception and inference, are not all-or-nothing propositions. A cat, a parrot and a man can all be fully described by the definition of Wooldridge and Jennings, although the degree of autonomy, pro-activity, and socialization through perception and action is not the same in these three agents. With some imagination, even a rock might be made to fit this description. Agency does not come in an all-or-nothing package and there are various levels of agency in the biological realm. Agents are not static things, but complex systems interacting with dynamic and complex environments, and, therefore, exhibiting a dynamic nature in themselves. There are some dynamic and incremental conceptual and material ingredients that are complexly integrated to constitute an agent to varying degrees and at various evolutionary. Therefore, different agents exhibit different degrees of agency at a specific point in time, and the same agent may also exhibit different degrees of agency at two different points in time. Keeping naturalization in mind, one may conclude that a complete definition of the term ‘agent’ is out of the question, and any efforts in this direction should express agency as a capacity with an incremental and an evolutionary character. In order to pursue such definition, we will try to modify Kampis’s (1999) evolutionary definition of agency, which comes as a list of somewhat ad hoc properties of an agent, in such a way that the suggested definition is more susceptible to a naturalized analysis. Therefore, since the ability to act upon an environment in order to achieve goaloriented effects is proper to a cognitive agent, we suggest that a strong notion of agency calls for: Please cite this article in press as: Arnellos, A. et al., Towards the naturalization of agency based on an interactivist account of autonomy, New Ideas in Psychology (2009), doi:10.1016/j.newideapsych.2009.09.005 A. Arnellos et al. / New Ideas in Psychology xxx (2009) 1–16 4 ARTICLE IN PRESS interactivity, that is, the ability of an agent/cognitive system to perceive and act upon its environment by taking the initiative; intentionality, the ability of an agent to engage in a goal-oriented interaction driven by beliefs and desires; and autonomy, the ability of an agent to function/operate intentionally and interactively based on its own resources. This definition does not try to statewhat an agent is or what an agent should do, insteadmentioning three nested capacities that an agent should exhibit over the course of its evolutionary development. According to this definition agency requires interactivity, which in turn implies action upon the environment. This action is not accidental but intentional, as it is a goal-directed and driven by content such as beliefs and desires. Additionally, such an agent exhibits the property of autonomy as it interacts with the environment in an intentional manner based on its own resources. These three properties seem to be quite interdependent, and there is no possibility that any one of them may be increased qualitatively in isolation from the others. However, notions such as autonomy, intentionality, belief, goal-orientation, cognition, etc. are philosophically loaded and controversial. This theoretical piggy-backing may prove a terminological and conceptual burden, especially considering that such an analysis is meant to serve to inspire and to guide the creation of artificial agency. With a critical perspective in the domain of complex systems research, Collier (1999) suggests that there is a very interesting interdependence between these three properties. Specifically, Collier suggests that there is no function without autonomy, no intentionality without function and no meaning without intentionality. The interdependence is completed by consideringmeaning as a prerequisite for the maintenance of system’s autonomy during its purposeful interaction with the environment. This conceptual interdependence places some interesting constraints on the capacities that contribute to agency and sets some requirements in terms of the properties that an agent should exhibit independently of its degree of agency or cognitive capacity. These properties and their interdependence are characteristic of the strong notion of agency (i.e. the one exhibited by living systems), which is considered as emergent in the functional organization of the living/cognitive system – the agent. The term ‘functional’ is used here to denote the processes of the network of components that contribute to the autonomy of the agent and particularly, to the maintenance of the cognitive system as awhole (see e.g. Ruiz-Mirazo &Moreno, 2004). Meaning, if it is not to be considered as an ascription by an observer, should be linked with the functional structures of the agent. It should guide the constructive and interactive processes of the functional components of the cognitive system in such a way that these processes maintain and enhance the system’s autonomy. In this perspective, the enhancement of autonomy imposes certain goals on the agent itself and hence, the intentionality of the agent is guiding its behavior through meaning. It shouldbenoted that in suchanautonomous systemintentionality is not reducible to theprocessing of meanings, nor are the combinations of meanings bringing forth any ‘aboutness’. On the contrary, meaning and its functional substratum are the defining properties of an autonomous agent thatmayact intentionally. In otherwords, an autonomous systemmayact intentionally if its actions aremediated by meaning.Hence, it appears that for a systemtoexhibit agency, it needs toexhibit thedegree of autonomy that will provide for the functionality that is needed in order to support its intentional and purposeful interaction with the environment, the result of which will create new meanings that will further enhance its autonomy.Moreover, agency has an interactive and a goal-oriented character, which results from the interactivity and the intentionality of the respective cognitive system. In the rest of this paper, an attempt is made to ground these properties in naturalistic frameworks of cognitive agency, and to describe their emergence during the development and the evolution of a cognitive agent. Autonomy will be the key concept in this endeavor. 3. Second-order cybernetics: agency as self-organization In general, within the framework of second-order cybernetics (von Foerster, 1981, 2003), agency is considered as the continuous process of modification of the intentional behavior of the system Please cite this article in press as: Arnellos, A. et al., Towards the naturalization of agency based on an interactivist account of autonomy, New Ideas in Psychology (2009), doi:10.1016/j.newideapsych.2009.09.005 A. Arnellos et al. / New Ideas in Psychology xxx (2009) 1–16 5 ARTICLE IN PRESS through the constant alteration of its organization. This conception is radically different from the traditional cognitivistic frameworks (see e.g. Newell, 1980), on which agency is the gathering and assembly of well-defined representations about the environment. Consequently, the focus on agency is shifted from a serial and static mapping between the internal and the external of a cognitive system to the coupling of constant and parallel structural processes of the system with its environment. It should be noted that the broader framework of second-order cybernetics and self-organization opens the black box of the cognitive system, provided that it offers a ‘mechanism’ that supports the behavior of a cognitive agent. At the same time, it neither presupposes nor predetermines the ‘vocabulary’ with which each cognitive system chooses to shape the various states of affairs in the environment and to act about and towards them. In other words, each cognitive system constructs its own meaning based on its actions in the environment. As such, the approach of second-order cybernetics and of self-organization in general is an interesting candidate for a non-circular way of describing and explaining the reference of the thoughts of a cognitive agent. The non-explicit or in a non-instructive participation of the environment in the shaping of the cognitive process and of themeanings generated in the self-organized system endows the systemwith a high degree of autonomy. As has already been mentioned, autonomy has the primary role in the establishment of a naturalistic framework for the analysis, explanation and modeling of the emergence and further development of meaning in a cognitive systemdthe emergence and development of agency. In the following section, we provide a descriptive analysis of the framework of second-order cybernetics, indicating the main characteristics of the self-organizing approach to agency. 3.1. Organizational closure and self-reference for self-organization In the second-order cybernetic epistemology, a cognitive system is able to carry out the fundamental actions of distinction and observation. It observes its boundaries and it is thus differentiated from its environment. As the cognitive system is able to observe the distinctions it makes, it is able to refer the result of its actions back to itself. This makes it a self-referential system, providing it with the ability to create new distinctions (actions) based on previous ones, to judge its distinctions, and to increase its complexity by creating new meanings in order to interact (Luhmann, 1995). The self-referential loop can only exist in relation to an environment. In contrast with classical systemenvironment models, the external control of a cognitive system’s adaptation to its environment is replaced by systemic and operational/organizational closure (von Foerster, 1960, 1981). Due to that closure, the self-reference of an observation creates meaning inside the cognitive system, which is used as a model for further observations in order to compensate for external complexity. The system which operates on meaning activates only internal functions and structures, which von Foerster (1981) calls eigenvalues, postulating some stable structures, which are maintained in the functions of the cognitive system’s organizational dynamics (Rocha, 1996) and which serve as points of departure for further operations during its interaction with the environment. Indeed, this closure is functional in so far as the effects produced by the cognitive system are the causes for the maintenance of its systemic equilibrium by forming new and more complex organizations. Given such organizational closure, environmental complexity is based solely on system observations, thus, system reality is observation-based. As von Foerster (1976) has argued, the results of an observation do not refer directly to objects in the real world, but instead, they are the results of recurrent cognitive functions in the structural coupling between the cognitive system and the environment. In particular, von Foerster states that ‘‘Ontologically, Eigenvalues and objects, and likewise, ontogenetically, stable behavior and the manifestation of a subject’s ‘grasp’ of an object cannot be distinguished’’ (von Foerster, 1976, p. 266). Thus, each new function based on observations is a construction, it is an increase in the organizational and cognitive complexity of the agent. This process of emergent incrementation of order through the internal construction of functional organizations and simultaneous classification of the environment is a process of self-organization (von Foerster, 1960,