The Challenge of Normativity: An Examination of Three Minimal Models

Abstract

For a number of perspectives in cognitive science and philosophy, a naturalistic account of normativity has become essential to understanding the capacity for adaptive behavior characteristic of living things. In support of this goal, in this article we work toward a definition of the minimal conditions for normatively guided behavior. We begin with the proposal that normative guidance requires the capacity to discriminate and respond to local differences of favorability so that behavior effectively follows a “value gradient.” The challenge of normativity, therefore, is to understand how this capacity could emerge in a primitive lifeform without consciousness or representational machinery. To clarify this challenge, we examine three minimal models that purportedly describe or exhibit a basic form of normative behavior: (1) the autopoietic model of enactive theory; (2) the thermodynamic model of ecological psychology; and (3) the autogen model of Terrence Deacon. A comparison of the relative strengths and weaknesses of these models leads to the suggestion that a minimal form of norm-following can be achieved by an entity that appropriates preexisting gradients as “proxies” for value, provided that this appropriation is genuinely self-directed or autonomous.

Introduction

Over the last several decades, a naturalistic account of normativity has become a key desideratum for enactive theory and other perspectives within the extended 4-E family (e.g., Barandiaran & Egbert, 2014; Barham, 2012; Barrett, 2017; Bickhard, 2015; Christensen, 2012; Di Paolo, 2005; García & Barandiaran, 2025; Mojica, 2021; Moss, 2024; Weber & Varela, 2002). This development is spurred by the thesis that normativity is essential to the capacity for adaptive behavior exhibited by all forms of life (e.g., Barrett, 2020; Di Paolo, 2018; Lyon, 2006; Thompson, 2007). In short, insofar as all living things exhibit a spontaneous tendency to approach favorable states and avoid unfavorable states (cf. Dennett, 1991, pp. 174–175), they seem to be guided by basic norms of self-preservation.

The simplicity of this thesis is deceptive, however. Even among those who agree on its fundamentality, the question of how to define the normativity that underlies self-preserving behavior remains unsettled. There is no consensus about the essential features of normatively guided behavior or the conditions for its emergence. Indeed, even in cases of human behavior which no one doubts are normative, there is no widely accepted definition of normativity. Nevertheless, a naturalistic account of normativity has become the focus of a steadily growing literature, and as its desirability is increasingly taken for granted, it has become easy to lose sight of the motivations for this line of inquiry, as well as its main objectives and criteria for success. What is normativity, why is it essential to cognitive agency, and what obstacles stand in the way of a naturalistic account?

The goal of this article is to bring the challenge of normativity into sharper focus, specifically as it relates to the emergence of adaptive agency in primitive forms of life. As the starting point of inquiry, we define normative behavior in connection to value, which in turn we define to include any and all ways of ordering states as more or less favorable. We then highlight the question of whether and how values in this sense must be recognized by an organism for its behavior to be considered normative. We propose that for the behavior of an organism to be normative in the relevant sense, it must be guided in the present by a capacity to discriminate and respond to differences of value. But how is this capacity realized? If normative guidance is a condition for the adaptive agency of life, it would seem that it cannot require the conscious discrimination of value. On the other hand, if a “representation-hungry” account is to be avoided, it seems that a conscious-like sensitivity to value is required. The challenge of normativity is, therefore, to understand how a basic version of this capacity could emerge without consciousness or elaborate representational machinery. ¹

To clarify this challenge, in this article we examine three “minimal models” of normative behavior: (1) the autopoietic model of enactive theory (Barandiaran & Egbert, 2014; García & Barandiaran, 2025); (2) the thermodynamic model suggested by experimental studies in ecological psychology of dissipative structures (De Bari et al., 2024); and (3) the autogen model of Terrence Deacon (Deacon & García-Valdecasas, 2023). Although these models belong to different theoretical perspectives, they are compatible enough to be brought into conversation and, perhaps, integrated into a common approach. Most importantly, all three aim to describe the conditions for a basic form of normatively guided behavior that is both nonrepresentational and nonconscious. In what follows, we will highlight key insights, contributions, and shortcomings of each model with the intention of showing how each might benefit from the others. The purpose of this comparison is not to argue for the superiority of one approach over another or to arrive at a complete theory, but rather to clarify shared objectives and criteria. Also, drawing from these perspectives, we suggest that a minimal form of norm-following could be achieved by the self-directed appropriation of a preexisting gradient.

The Challenge of Normativity

Let us begin by taking note of the diversity of normativity, especially as it applies to human life. Broadly speaking, the realm of normativity encompasses anything to which it is possible to apply the terms ought or should (Star, 2018). Ought and should evidently apply to human life in countless different ways—from a game of cards to the raising of a child—each of which may correspond to a different kind of normativity. Even so, it is widely believed that different kinds of normativity can be unified as species of a common genus, although there is less agreement about how to do this (Wedgwood, 2018). As a result, in contemporary philosophy, the term normativity is frequently used to denote an undefined “common character” shared by terms like value, good, ought, and reason and yet “only imperfectly and incompletely picked out by any of them” (Finlay, 2010, p. 331).

This diversity is not a problem as long as it is recognized and we take care not to conflate different kinds of normativity or assume the existence of just one kind. In this section, we move toward a more focused picture of the kind of normativity that is essential to life and cognition. As will soon become apparent, the kind of normativity we are seeking remains undefined in certain key respects, but at the same time it excludes various kinds of normativity (e.g., social normativity) that figure prominently in human life.

Here, we are primarily interested in the normative guidance of behavior. Generally speaking, behaviors can be considered normative—better or worse—to the degree that they directly realize or contribute to the realization of some value. Value is itself a highly vexed topic, and its naturalization presents no less of a challenge than normativity. Fortunately, for present purposes, we can adopt a very loose definition of value, one that includes any and all reasons that could provide normative criteria for behavior so as to constitute a continuously graded scale of better and worse. To say that such reasons exist is not trivial, but it does not assume the capacity to recognize and act upon these reasons, much less think about them. It only means that for some entity (which may need further definition) states can be ranked as more or less favorable. As we will see, the determination of a continuous ranking of possible states and behaviors as better or worse—a “value gradient”—is a critical feature of the kind of normativity to be investigated here. ²

Now, with this simple definition in hand, we can begin to refine our view of the challenge of normativity.

First, returning to our initial point, it is crucial to keep in mind the many different ways in which behavior can be normative. For example, in the case of human behavior, a basic kind of normativity is entailed by the intrinsic satisfaction of movement: to the degree that it feels good, a movement is normative in that respect, regardless of what else it achieves. Also, behaviors can realize more than one value and so be normative in more than one way. So, to use the same example, a behavior that feels good might also be normative in respect of physical health, economic benefits, and social propriety. Of course, these norms do not always coincide, and as a result human behavior is often pulled in many directions at once. This point about normative diversity may seem rather distant from our current focus, but its importance cannot be overstated. Even if human normativity is unusually complex, we should not assume that simpler forms of life involve only one kind of norm.

Next, we suggest an additional condition for genuinely normative behavior. For behavior to be normative in the relevant sense—for behavior to be normatively guided—it is not enough for it to realize value by accident, or because of a past history of learning or selection ³ : the present discrimination of a relevant value gradient must play a guiding role in the performance of behavior. In other words, normative behavior must be directly responsive to differences of better and worse. This, then, is the challenge of normativity: How are normative behaviors guided by the values they serve to realize?

This question can be sharpened by considering James Barham’s helpful distinction between the “scope problem” and the “ground problem” (2012). In the most general sense, the scope problem has to do with the kinds of entities for which value-constituting reasons exist, such that normative criteria apply to their behavior. As we have just indicated, normativity applies in many ways, corresponding to the many different values that might be realized by behavior or, what is much the same, the many different reasons why behavior should be guided one way rather than another. In this article, we will be focusing on basic “prudential norms” related to values of organism integrity, health, and survival (p. 93). ⁴ But regardless of how the most basic norms of life are defined, we suggest that the main problem is not to determine the scope of normativity but to explain its influence on behavior. This is what Barham calls the “ground problem”: “understanding how something that is not an ordinary cause or physical cause can nevertheless have a real influence or determinative power over events in the world” (ibid.). Arguably, the ground problem should have priority, not only because it is more challenging, but also because it can be used to define the effective scope of normativity: practically speaking, normativity applies only to things capable of normatively guided behavior.

We are proposing that normatively guided behavior depends on the capacity to make discriminations of value, but this is merely a signpost indicating the direction in which inquiry should proceed. How should we understand the discrimination of value in naturalistic terms? How does this capacity emerge within rudimentary forms of life?

To clarify the challenge before us, let us briefly consider the normative guidance of conscious behavior. This may seem like an unnecessary detour, given that our aim is to describe the minimal conditions for normative behavior that we are presuming to be nonconscious. But to undertake this task is to assume the existence of a general category of normative behavior that includes both conscious and nonconscious varieties, and this category needs to be defined. We are seeking a definition of normativity that distinguishes traits that are essential to normativity per se from traits that are unique to conscious normativity. To formulate a hypothesis about the former, it makes sense to begin with our experience of our own behavior as normatively guided, as this is the only behavior we can safely assume to be genuinely normative in this respect. Accordingly, in the next few paragraphs we briefly examine our conscious experience of normativity with the goal of clarifying minimum conditions that could be applied to normativity in general. ⁵

Unfortunately, when we turn to our experience of normativity—the only kind of normativity with which we are directly acquainted—we find that the influence of values on our behavior is by no means easy to describe. Some values and norms are consciously considered and voluntarily chosen while others are more implicit and habitual; some are conventional and culture-specific while others are species-universal. Nevertheless, it can be argued that our behavior is directly influenced by values only insofar as they are present to our experience as felt inclinations or preferences. ⁶ Because normativity is not a matter of what can be done but rather what should be done, its influence on behavior must be the sort that inclines us to act, that attracts and repels rather than strictly determines. Moreover, this influence is not simply a brute force that pushes or pulls, but is always grounded in some prima facie reason, however inarticulate, for acting. It seems that felt preferences have this kind of influence, as they constitute reasons that influence our behavior without forcing and without needing to be articulated as reasons. A basic manifestation of this influence is a felt preference for the act itself. We sometimes feel that it would be better to act a certain way, even if we cannot say why or what we hope to accomplish thereby. An even more basic manifestation is when we feel inclined to do something without yet knowing what we want to do.

These common experiences suggest that conscious behavior can be normatively guided without having any specific goal or plan, and without any articulate motive. The minimal condition for conscious normative behavior is simply a felt inclination or influence that we feel as normative, that is, as roughly indicating that a more favorable state of affairs should be brought about. This felt inclination, even if no more than a vague nudge, entails an implicit discrimination of value—of better or worse—that is arguably essential to normative influence. It seems that discrimination of value and normative influence are mutually entailing: just as there cannot be normative influence without some discrimination of value, value cannot be discriminated without influencing our behavior. It is intrinsic to the nature of value that its discrimination constitutes a reason and impulse to act, even if the reason is not made explicit and the act is not specified.

In our experience, then, it seems that there is a minimal sense in which value must be recognized for behavior to be normative. Most of the time we do not consciously articulate the values that influence our behavior—we do not usually recognize values in this sense—but nevertheless we do recognize the normative influence of values insofar as they determine our felt preferences. In short, as encountered in experience, the distinctive influence of normativity on behavior has the character of motivation (Barrett, 2020). Accordingly, even if conscious motivations cannot be ascribed to primitive forms of life, perhaps we should consider the possibility that for a cause of behavior to count as normative it must share certain traits of motivation—it must be akin to a felt preference. But what could it mean to be akin to a felt preference without actually being conscious? We propose to define the relevant aspect of felt preferences as follows.

Even in the minimal case, felt preferences constitute a value gradient with a real influence on behavior. This, we suggest, is one of the essential traits of normativity per se that we are seeking to generalize beyond conscious behavior and describe in its most rudimentary form. Felt preferences respond to value gradients without needing to articulate the values that define their gradients, and without defining specific goal-states or behaviors. They merely detect “local differences” of favorability and push, strongly or weakly, for adjustment toward a more favorable state. This is a very minimal form of normative guidance but it is far from nothing.

Indeed, we cannot be sure that this form of normative guidance can be generalized beyond conscious behavior. Nothing that we have said so far rules out the possibility that differences of value have to be felt as such—they have to be experienced as conscious motivations—for us to speak of normative influence. But most would say that making conscious motivation a condition for normative guidance sets the bar too high, as it likely excludes many if not most forms of life, not to mention many if not most of our own cognitive processes. In any case, the point of our brief discussion of felt preferences is only to indicate that values do not have to be recognized as such to guide behavior; they only have to be recognized in the weaker sense of a discernible gradient-like influence. Some such influence—though still in need of explanation—seems to be a more appropriate minimal condition for normative behavior. Otherwise accidentally favorable behavior would count as normative, which sets the bar too low.

In sum, we are proposing that for any behavior to count as normatively guided, there must be some way—not just in reference to past history, but in the present—to show that this behavior is performed (or performed a certain way) because it is normative. But if conscious motivation cannot be a condition for normative behavior, how should we understand the discrimination of value in more general terms without reducing it to something else? This is the central question posed by the challenge of normativity, and it remains unanswered.

Now, having clarified this challenge, we need to strengthen the motivation for a naturalistic but non-reductive account of normatively guided behavior. Why struggle with this problem? What lies behind the intuition that normativity is essential to life and mind in its most basic form?

First, evidence for normative guidance can be gathered from observations of primitive forms of life. Although we will not detail this evidence here, it should be kept in mind as both motivation and testing ground for theories of normativity. Consider, for example, the following description from biologist Herbert Spencer Jennings’ Behavior of Lower Organisms (1906, pp. 338–339; cited in Withagen, 2023, p. 49):

The organism moves and reacts in ways that are advantageous to it. If it gets into hot water, it takes measures to get out again, and the same is true if it gets into excessively cold water. If it enters an injurious chemical solution, it at once changes its behavior and escapes. If it lacks material for its metabolic processes, it sets in operation movements which secure such material. If it lacks oxygen for respiration, it moves to a region where oxygen is found. If it is injured, it flees to safer regions. In innumerable details it does those things that are good for it.

This passage clarifies that it is the capacity of organisms to adjust behavior “in innumerable details” that most strongly suggests an open-ended form of normative guidance rather than a fixed repertoire of behavioral responses. Although we will not argue this point here, it is the open-ended nature of normative behavior that tells against representational accounts of normativity—for example, those that posit homeostatic mechanisms with predefined goal-states (e.g., Lyon & Kuchling, 2021).

In addition, it can be argued that the capacity to make normative discriminations is so important to survival and reproductive success that we should expect it to have emerged along with the first self-replicating lifeforms. Here is how Daniel Dennett makes this point (1991, pp. 174–175):

[I]f these simple replicators want to continue to replicate, they should hope and strive for various things; they should avoid the ‘bad’ things and seek the ‘good’ things. When an entity arrives on the scene capable of behavior that staves off, however primitively, its own dissolution and decomposition, it brings with it into the world its ‘good.’ That is to say, it creates a point of view from which the world’s events can be roughly partitioned into the favorable, the unfavorable, and the neutral. And its own innate proclivities to seek the first, shun the second, and ignore the third contribute essentially to the definition of the three classes. As the creature thus comes to have interests, the world and its events begin creating reasons for it – whether or not the creature can fully recognize them…. The first reasons preexisted their own recognition. Indeed, the first problem faced by the first problem-facers was to learn how to recognize and act on the reasons that their very existence brought into existence.

Dennett’s use of anthropomorphic language in this passage—replicators “should hope and strive”—is presumably intended for rhetorical effect: he means to highlight the fundamental importance of the capacity to “recognize reasons,” that is, to discriminate value. What Dennett calls the “first problem” of life is more basic than any particular problem faced by an organism. After all, without the discrimination of value there can be no preferences, and without preferences there can be no problems. The “first problem” of life is therefore to be able to recognize the preferable and the problematic, that is, to discriminate between favorable and unfavorable.

Moreover, this passage indicates not just the priority of the capacity to discriminate value, but also its enormous cognitive utility or power. Once attained, the capacity to discriminate value can generate an indefinite range of adaptive behaviors, endowing life with the capacity to adapt to novel situations. Admittedly, this capacity still needs to be explained, but to motivate work on the challenge of normativity it is crucial to recognize its generative power. Two further points can help to bring this power into focus.

First, the discrimination of value is not the same as, and need not entail, the discrimination of meaning. Indeed, it is the difference between these that makes the former so powerful. This point was made implicitly in our brief discussion of felt preferences. There we pointed out that the discrimination of value always entails a minimal inclination or impulse to act, but it does not always specify a particular action or goal. For example, we sometimes experience a vague discomfort without knowing why or what exactly to do about it, although we are sure we want to do something. Our discomfort then initiates and guides a search for a satisfactory behavioral adjustment.

Second, as just indicated, the discrimination of value should not be separated from behavior, but rather should be conceived as a basic kind of embodied know-how (cf. Kauffman, 2000, p. 110–117). The discrimination of value and the normative selection of behavior are one and the same. However, although the normative selection of behavior requires behavioral variability, it cannot be defined as a set of behavioral proclivities or a list of things to be avoided and sought. It is, rather, that which guides the generation of behavior in any and all situations; the set of normatively guided behaviors cannot be predefined.

There still remains, of course, the task of defining this capacity in minimal, nonconscious terms. The simplest replicating systems may not literally “hope and strive for various things,” but if they are to survive and reproduce, they should act as if they did. They should “avoid the ‘bad’ things and seek the ‘good’ things,” and they should do so without needing to know beforehand what these things are. How can we understand the emergence of this motivation-like tendency without positing conscious motivation? This is the challenge of normativity.

Minimal Normativity in Enactive Theory: The Autopoietic Model

Of the three perspectives to be discussed in the remainder of this essay, enactive theory is the most actively and directly involved with the challenge of normativity, and has produced the most extensive literature on the topic. ⁷ Rather than attempt a comprehensive review, in this section we examine an influential “minimal model” of normative behavior—what we call the autopoietic model—presented by Xavier Barandiaran and Matthew Egbert in their article, “Norm-Establishing and Norm-Following in Autonomous Agency” (2014). As indicated by its title, this model clarifies a critical distinction between norm-establishing and norm-following, roughly corresponding to the “scope” and “ground” problems discussed earlier.

The autopoietic model belongs to what Barandiaran and Egbert call the “organism-centered or autonomous organizational approach” to normativity (p. 8). Following key precedents such as Georges Canguilhem and Hans Jonas, this approach focuses on the organism as a special kind of individual—an autopoietic individual—which can be characterized as follows. An autopoietic individual is a network of processes that are distinguished as a set by their self-producing interdependence or “operational closure” while also drawing a constant supply of materials from the environment. The tension between closure and openness makes the autopoietic individual inherently precarious and unstable, such that to continue existing it must continually re-make itself (Di Paolo, 2018). This concept of the individual organism as a precariously enduring pattern of interdependent processes is at the core of enactive thinking about normativity during the past several decades. How does a minimal form of normative behavior emerge on this view?

Let us grant that the preceding definition picks out a class of individuals that includes all living things. ⁸ A critical feature of these individuals is the inherently precarious nature of their existence: they require self-maintenance to persist, and this self-maintenance is not occasional self-repair but a continually active process of self-production. Enactive theorists argue that any such individual, by its very existence, brings forth a set of reasons or norms related to the conditions for its persistence, aka “viability conditions” (Weber & Varela, 2002; Di Paolo, 2005; cf. Dennett, 1991). Thus, the autopoietic model clarifies the intrinsic normativity of life by showing how norm-establishing is inherent to the constitutive dynamics of a class of individuals that includes all living things (Barandiaran & Egbert, 2014, pp. 8, 12–13).

Now, when such norm-establishing individuals not only emerge but also manage to persist in a changing environment, we have reason to suppose that they are able to adjust their dynamics in conformity with their own norms: they are capable of norm-following adaptive behavior (Di Paolo, 2005). ⁹ Indeed, in the case of living things, evidence for norm-following is often so striking that we may take this adaptive capacity as an established fact. Our task, however, is to “reverse engineer” this capacity and provide a minimal description of the kind of dynamics that makes it possible. In this respect, Barandiaran and Egbert’s minimal model is especially illuminating, as it clarifies that the dynamics of norm-establishing and norm-following are not necessarily the same (see esp. pp. 11–12, 24). Not only is it possible for the former to obtain without the latter, both can obtain through separate dynamics without any causal connection. Indeed, as they point out, for many models, including their own (see p. 24), “the organismic dynamics that define viability and the dynamics that control adaptive behavior remain dissociated” (p. 12).

This distinction needs special emphasis, as it arguably points to an essential criterion for all accounts of normative behavior. Even if we take norm-following to be an established fact for living things, we still need to understand how this capacity emerges from the same dynamics that constitute the organism as an individual. The dynamics of norm-establishing and norm-following need to be integrated or else autonomous agency is lost. As pointed out by Barandiaran and Egbert, when norm-establishing and norm-following dynamics are dissociated, the entity in question may happen to be configured so that it follows norms but it will “do so blindly” (p. 24): the normativity of behavior is accidental, not genuine.

By insisting that norm-following cannot be blind, it may seem as if Barandiaran and Egbert are calling for something like conscious awareness of norms, but that is not the case. In another key move, they propose that “any causally relevant effect of norm-establishing processes upon behavior generating mechanisms” (ibid.) may be sufficient for genuine autonomous agency. However, although they indicate how this condition could be fulfilled by the integration of chemotaxis with metabolism (see Egbert et al., 2011), to our knowledge this work has not led to a description of its realization by a “real-life” system. Without such an account, the integration of norm-establishing and norm-following remains an unfinished task for enactive theory.

Another feature of Barandiaran and Egbert’s model that deserves special mention is their introduction of the concept of normative field. The normative field is a “virtual space” that is generated when every point of the state space of an autopoietic individual is given a value—a ranking of better or worse—in relation to the viability conditions of that individual (pp. 16–17, 22–23). For example, the normative field of Barandiaran and Egbert’s minimal model is a vector field defined in a simple two-dimensional space of “metabolic dynamics” in which vectors at each point indicate the metabolic requirements needed at that point to return to the viable region (p. 17, see Figure 1). ¹⁰ Roughly speaking, the smaller the vector, the more favorable the state of the system at that point.

Figure 1.

Representation of a normative vector field. Vectors indicate direction and minimum amount of constant increase in [F], a food variable, required to reach viable region. The variable [A] stands for autocatalysts and represents concentration of chemicals involved in an autocatalytic network. Reproduced from Barandiaran & Egbert 2014 (p. 17) with permission from MIT Press.

The concept of normative field illustrates how normativity can be conceived as a kind of gradient, such that normative behavior can be conceived as a kind of gradient-following behavior. Moreover, using the concept of normative field, we can define a minimum requirement for normative behavior: the capacity to follow contours of a normative gradient by discriminating local differences of value. Thus, the concept of normative field helps us to visualize a minimal form of nonconscious but “felt-preference-like” behavior that pursues basic prudential values related to integrity, health, and survival.

Of course, we still need to know how local differences of value are effectively discriminated. As Barandiaran and Egbert point out, although the normative field of an autopoietic individual is defined non-arbitrarily by viability conditions with real consequences for that individual, it remains “virtual” insofar as it lacks any causal connection to the behavioral dynamics of the entity in question. Without this connection, “it makes no sense to claim that the norms themselves have any causal or regulatory effect” (p. 24). This is another critical point that needs special emphasis.

Although in principle every autopoietic individual establishes its own normative field—defined by its constitutive dynamics in relation to relevant aspects of its environment—and although differences of the normative field have real consequences for the autopoietic individual, there is no reason to presume that any such individual is actually sensitive to the contours of its own normative field. Sensitivity to even the most basic norms of viability would seem to require sensitivity to all of the relevant variables, which for real organisms are highly complex and interactive. Consider our own normative field as defined by viability: although we are fairly sensitive to many variables such as metabolic levels and ambient temperature, we are completely oblivious, for example, to deadly levels of radiation. Accordingly, although the normativity of radiation levels for us is quite real—there is no question that we should avoid high levels of radiation—it remains “virtual” as long as we cannot discriminate these levels.

Barandiaran and Egbert are able to work around this problem because for their model it happens that an effective approximation of metabolic norm-following can be achieved if the individual simply acts to increase a single variable, the “food variable” [F], whenever possible. Even in their simple metabolic model, however, the tendency to increase this variable is a feature that they must add on. As pointed out earlier, to achieve genuine autonomy this behavioral dynamic needs to be integrated with the constitutive dynamics of the autopoietic individual, as they themselves acknowledge (p. 24). Still, we suggest that this jerry-rigged solution is itself an important step forward. The key feature of the food variable is that its maximization constitutes a form of gradient-following behavior that can be used by the individual to approximate “true norm-following” (as defined by its viability conditions). It suggests, therefore, that a preexisting capacity for gradient-following can be effectively repurposed by an individual so as to mimic norm-following.

In the next section, we examine a real-life example of how norm-following could be grounded in this way. But first, in the remainder of this section, we wish to indicate how the gap between norm-establishing and norm-following remains problematic for the autopoietic model.

First, for the autopoietic model, norms are defined by the formal identity of the system, which is essentially an abstract pattern of interaction, and therefore exists at a longer timescale than norm-following behavior, which presumably belongs to specific component processes. ¹¹ For example, in the case of metabolic autopoiesis, a network of interdependent chemical reactions constitutes a pattern—a metabolic individual—that exists at longer timescale (that of a metabolic cycle) than its component reactions. Presumably a similar difference of timescale also exists in the case of sensorimotor and social individuals and their component processes (Di Paolo et al., 2017; Di Paolo et al., 2018; Mojica, 2021). It needs to be explained, therefore, how component processes can become responsive to norms that belong to the pattern of their interaction over some minimally extended period of time.

Another issue is the kind of normativity established by autopoietic individuals, and the question of whether this is sufficient for effective norm-following. One of us has argued that the autopoietic model is limited to a “purely proscriptive” or negative norm that pushes away from precarious states (Barrett, 2015, 2017). This limitation might be overcome by a model that maximizes variables related to viability, as suggested recently (García & Barandiaran, 2025). But whether they are proscriptive or prescriptive, all autopoietic norms are defined in relation to the enduring formal identity of an autopoietic individual: a pattern that persists over time. As a result, all such norms are essentially conservative, that is, identity-conserving (Barrett, 2015, p. 237), and this restriction complicates, if not precludes, the creation of new identities and the transition to new forms and patterns. ¹²

Enactive theorists have responded to these criticisms in a number of ways: by indicating how complex kinds of normativity emerge from multiple overlapping or embedded individuals (Mojica, 2021), by incorporating an inherent tension and instability into the very nature of autopoietic individuals (Di Paolo, 2018; see also Di Paolo et al., 2017; Di Paolo et al., 2018), and by emphasizing the “messiness” of living bodies and suggesting that organisms are subject to multiple kinds of normativity that are never perfectly unified (McGann, 2024a). However, while each of these developments may constitute an improvement on the autopoietic model, none of them questions its fundamental premise, namely that normativity emerges in relation to a precariously enduring individual. Following this premise, enactive normativity is confined to the viability conditions of an enduring entity that already exists. Consequently, all autopoietic norms are norms of endurance, and all attempts to revise this model are confined to the logic of self-preservation. Could considerations of a different kind of normativity help enactive theory to overcome its limitations with respect to norm-following?

A number of recent articles inspired by the philosophy of Gilbert Simondon (Dereclenne, 2021; Di Paolo, 2021, 2023; García, 2023) explore the process-relational ontology that implicitly underlies autopoietic individuation. Although still in the early stages of development, the orientation of this “Simondonian turn” is clear: “to think of the (onto)genetic process by which individuals come into being instead of focusing on finished, static, and constituted entities” (García, 2023, p. 4). This shift of focus seems to call for a different set of norms that can apply to “novelty and transformation rather than stability and identity” (ibid.), that is, to the realization of concrete individuality rather than the maintenance or propagation of an already formed, abstractly defined individual. It remains to be seen how these norms will be defined and incorporated into the autopoietic framework. And, of course, a naturalistic account of norm-following will still be needed. Even so, the turn toward a more genetic, process-relational perspective offers the enticing prospect of explaining how norms of endurance might be realized through processual norms of non-enduring individuation. ¹³

Minimal Normativity in Ecological Psychology: The Thermodynamic Model

Although ecological psychologists have not prioritized normativity in the same way as enactive theorists, it is arguably no less important to their understanding of mind, even if it remains largely in the background. In this section, we briefly trace the outlines of an ecological approach to the challenge of normativity, with special attention to recent research on end-directed behavior in dissipative systems.

Toward the end of his life, James Gibson described his work as an attempt to move toward “a psychology of values rather than a psychology of stimulus” (Reed, 1988, p. 296; cited in Hodges & Baron, 1992, p. 263). Suggestive as this statement may be, there is no simple and straightforward way to describe ecological psychology as a “psychology of values.” In Gibson’s own writings, the centrality of value might seem to be implied by his definition of the concept of affordance as what the environment offers to the animal “either for good or ill” (1979, p. 127). Notice, however, that the phrase, “for good or ill,” is ambiguous, as it could be taken to refer simply to a kind of meaning rather than a normative influence on behavior. Granted, most of the time these are not separable aspects of experience. But for present purposes it is crucial to distinguish meaning from value. Defined in purely semantic terms, a discrimination of meaning, even if it includes the possibility of benefit or harm, does not necessarily include any influence on behavior. In contrast, the discrimination of value always entails some inclination to act (see discussion above). Gibson recognized this difference, but perhaps because he was unsure how to conceive of value in ecological terms, he equivocated about whether to include the influence of value in his concept of affordance. ¹⁴ In short, Gibson did not disambiguate the discrimination of value from the discrimination of meaning and, subsequently, insofar as ecological psychology has focused on meaning, it has not yet realized its promise of becoming a “psychology of values.” ¹⁵

An important exception is the “ecological values theory” of Bert Hodges (Hodges, 2023; Hodges & Baron, 1992; Hodges & Rączaszek-Leonardi, 2022). Hodges has shown that it is possible to make a solid case for the distinctive influence of value on human behavior even if we cannot explain its physical basis. He does this by calling attention to the way in which actions are shaped by “proprietary” constraints. For example, it can be shown that objects of similar size and weight—such as a baby and a bag of groceries—are carried differently so as to conform to different proprietary constraints (Hodges & Lindheim, 2006). Hodges argues that these constraints cannot be attributed to natural laws, intentionally chosen goals, or socially constructed rules, and in so doing he reveals a value-shaped hole, so to speak, within the science of human behavior. Once brought to our attention, the influence of value on behavior emerges as a highly complex, situation-specific, ecologically constituted set of constraints that cannot be reduced to preferences of the subject or properties of the environment. According to Hodges, values define the fields within which actions are formed as well as the boundary conditions for their proper execution (Hodges, 2023, p. 5). For example, when driving a car, our behavior is shaped by multiple values, including “accuracy, tolerance, speed, safety, freedom, trust, comfort, equality, justice, stability, and flexibility” (p. 2).

In light of this complexity, it seems that ecological values cannot be objects of full conscious awareness. Rather, insofar as they are present to awareness, they belong to the complex and continually changing “demand character” of our current situation, a pre-reflective dimension of experience that Kurt Lewin described as a “hodological” space or field, from the Greek hodos for path (see Lambie, 2020). For the Gestalt psychologists, this hodological field constitutes a real normative field, that is, one that exerts a real causal or regulatory effect on behavior. Hodges has given new life to this view by pointing out tell-tale signs of the influence of proprietary constraints on the organization of actions like carrying. However, for ecological values theory to become more firmly established, we need a naturalistic account of this influence, or at least a plausible idea of what such an account would look like. How can we account for the subtle “forcing” of our behavior by ecologically constituted normative constraints?

To explore a minimal version of this question, we turn to a line of research within ecological psychology that focuses on experimental studies of nonliving dissipative structures (Kondepudi et al., 2020; Dixon et al., 2015; Kondepudi et al., 2015; De Bari et al., 2019; De Bari et al., 2023). These “bio-analog systems” (De Bari et al., 2024) provide what appears to be a kind of “minimal model” of norm-following behavior: a system whose adaptive behavior, or “dissipative adaption” (England, 2015), seems to be shaped by a thermodynamic version of a normative field—a field shaped by “thermodynamic reasons” (Swenson & Turvey, 1991).

Here, we will limit our discussion to studies of electrical dissipative structures (EDS) made of metal beads in a thin layer of oil subjected to high electrical voltage. As electric current flows through the system, the beads spontaneously form a tree-like structure that attaches to the wall of the containing dish (see Figure 2). Once formed, these structures display a remarkable array of life-like behaviors. They grow and repair themselves when damaged, and they are capable of various behaviors that resemble “foraging”: they move toward energy sources, change their shape to get around obstacles, and sweep back and forth to harvest electrical charges that accumulate on the surface of the oil (De Bari et al., 2024).

Figure 2.

Electrical dissipative structure (EDS). Top image (a) depicts the experimental set-up, in which conducting beads are placed in 60 mL of oil and subjected to voltage V = 26 kV across source and ground ring-electrode. Photos (b–e) show the time evolution of the tree structure. Graph (f) and red arrows indicate how formation of tree structure, occurring between 40–45 seconds, corresponds to a sharp increase of entropy production ∑ (mJ/K/sec). Reproduced from Dixon et al. 2015 (p. 190) with permission from World Scientific Publications.

To an observer who is unfamiliar with the experimental set-up, this behavior would appear to be normatively guided toward the preservation and maintenance of these tree-like structures. And, in fact, the authors of this work suggest that the EDS adjusts itself in accordance with a simple norm: “behave so as to persist” (p. 2). It would appear, then, that this system constitutes a minimal case of normatively guided behavior in which the relevant norm is defined in relation to an enduring individual—the tree structure.

However, on closer inspection, we find that the end-directedness of this system might be other than what it appears. The flow of electric current through the EDS can be measured, and it can be shown that its behavior tends toward states with higher rates of entropy production (REP), which is related to the flow of energy through the system. For example, as shown in the figure (see (f) in Figure 2), a marked increase of REP coincides with the formation of the tree structure. One could argue, therefore, that the persistence of the tree structure is not the “true end” of this system’s behavior, but rather a means toward the increase of REP. If the system has a norm, it is simply to increase REP whenever possible, a spontaneous tendency of non-equilibrium systems known as the Principle of Maximum Entropy Production (PMEP).

PMEP is a variational principle of non-equilibrium thermodynamics. Whereas the second law of thermodynamics dictates that total entropy of an isolated system increases and evolves toward a maximum, PMEP dictates that systems maintained away from equilibrium by an energy gradient will evolve toward states that dissipate that gradient at a faster rate. Some have argued that PMEP expresses a fourth law of thermodynamics (Swenson, 2024). Although we will not enter into this question, for our purposes we can consider PMEP to be law-like at least insofar as it applies to “non-equilibrium systems with nonlinear forces and flows” (De Bari et al., 2024, p. 7), a class that presumably includes all living systems.

For the sake of argument, let us suppose that PMEP explains the dissipative tree structures by providing “thermodynamic reasons” (Swenson & Turvey, 1991) for their emergence and behavior. ¹⁶ On this view, the “reasons” of PMEP constitute a thermodynamic version of a normative field. To every configuration in the state space of the dissipative system, we can assign a value constituted by the system REP in that state (Dixon, 2024). ¹⁷ Following PMEP, dissipative structures are “sensitive” to local differences of REP and adjust themselves so as to increase REP whenever possible, and over time these adjustments constitute a kind of gradient-following behavior that effectively approximates the norm “behave so as to persist.” Accordingly, even if we cannot say that the EDS is directly sensitive to its normative field, it still might be capable of a minimal form of norm-following, insofar as its pursuit of higher REP effectively functions as a proxy for the norm of self-preservation. But are we really justified in calling this behavior normative if its “true norm” is simply the law-like dictates of PMEP?

Certainly, the system acts as if it were normatively guided. What gives it this appearance is the spontaneous responsiveness and adaptivity of the tree structure. Looking at videos of the electrical dissipative system in real time, we seem to be witnessing the emergence of a collective unity—a kind of individual—that is capable of responding to its environment in accordance with its own norms, moving toward more favorable states whenever possible: “In innumerable details it does those things that are good for it.”

However, in this article we have argued that the question of normativity rests not just on the favorable consequences of behavior but also on the manner of its determination. In the previous section, we drew from Barandiaran and Egbert’s model (2014) the criterion that a minimal form of autonomy is achieved only if the dynamics of norm-following are somehow integrated with the constitutive dynamics of the norm-establishing individual. They propose that “any causally relevant effect of norm-establishing processes upon behavior generating mechanisms” (p. 24) may be sufficient for genuine autonomous agency. For the EDS, this condition seems to hold, insofar as its constitutive dynamics and its various adaptive behaviors (foraging, etc.) are both driven by PMEP. Strictly speaking, however, the behavior of the tree structure is directed not by PMEP alone but by the way PMEP is conditioned and constrained by a number of variables, some of which belong to the tree structure, and some of which belong to the entire experimental set-up: the constant flow of high-voltage electrical current, the size of the container, the viscosity of the oil, and so forth. Once all of these factors are given their due, can the tree structure still be singled out as an individual, autonomous agent?

When answering this question, we should take care not to impose a concept of agency that is inconsistent with the relational frame of ecological psychology. For the ecological approach, behavior is a function of the entire organism-environment system, not the organism taken by itself. Nevertheless, in relation to the behavior of the tree structure, it is possible to make a distinction between exogenous and endogenous variables: the former are variables that affect the tree structure without being affected by it in return, while the latter both affect and are affected by its appearance and behavior (Dixon, 2024). Notice that this distinction does not belong to any simple partition of tree structure and its environment; it is not a simple question of being internal or external. For example, the source of high-voltage current is an exogenous variable, while the distribution of charges on the surface of the oil is an endogenous variable, even though both would appear to be external to the tree structure. With this distinction in mind, we can pose the question of autonomy without presuming to separate the tree structure from its environment. And we can say that the tree structure is autonomous insofar as the variables that direct the influence of PMEP toward its preservation are endogenous. Conversely, insofar as these variables are exogenous, what decides the normative character of the behavior is the experimental set-up, not the tree structure itself.

We will not try to settle this question here. Also, for similar reasons, we will not try to determine whether the tree structure is the kind of individual that distinguishes itself as such. Although all dissipative structures are characterized by a kind of self-sustaining circularity, “naked” dissipative structures such as whirlpools do not maintain or construct any distinction or barrier between themselves and their environment, and it is not clear that the electrical dissipative structures discussed here are any different. Certainly, in comparison with the autopoietic model and the autogen model (discussed below), the tree structure lacks the more robust kind of closure that is characteristic of autocatalytic sets. It also lacks a physical barrier that separates “inside” from “outside.” Again, our goal is not to settle these matters, but rather to indicate that genuine norm-following turns on the question of whether the entity to which normative behavior is attributed can be understood as an autonomous individual. If the tree structure is not an autonomous individual, we cannot claim that it uses PMEP to achieve the end of self-preservation. Rather, preservation of the tree structure is simply a fortuitous consequence of PMEP under the conditions established by the experimental set-up.

Notice that we have returned to the problem of integrating norm-establishing and norm-following discussed in the last section, but the challenge is now reversed. The autopoietic model offers a promising account of how norm-establishing is grounded in the constitutive dynamics of a certain class of individuals (i.e., autopoietic individuals) that includes living things. The challenge from that perspective is to explain norm-following behavior in a way that is integrated with this account of norm-establishing. Now, taking up a thermodynamic perspective, we have a promising candidate mechanism for gradient-like norm-following behavior—the influence of PMEP—which belongs to a different but overlapping class of systems (i.e., dissipative structures) that also includes living things. But now the challenge is to show that this behavior belongs to an autonomous individual.

Minimal Normativity and Teleodynamics: The Autogen Model

We now turn to consider Terrence Deacon’s theory of teleodynamics (Deacon, 2012; 2021; Deacon & García-Valdecasas, 2023; García-Valdecasas & Deacon, 2024). Originally developed to explain how the teleological dynamics of life are causally generated, teleodynamics also proposes a testable model to explain the emergence of normative behavior from non-normative processes. In the present context, it provides an ideal third perspective on the questions discussed so far, as it combines ingredients of the autopoietic and thermodynamic models and proposes to make significant modifications to both. For the sake of continuity, the following exposition focuses on Deacon’s view of (1) the individual to which norms apply and (2) the causal account that suffices for behavior to be considered truly normative.

In agreement with the autopoietic approach, Deacon affirms that a precarious form of self-maintained existence is a basic condition for normativity. However, he questions whether the operational closure of an interdependent network is sufficient to define a genuine individual, or at least the kind of individual that can serve as the locus and ground of normativity. Although operational closure is not an observer-dependent feature, it could be that only an observer can discern its pattern as an individual identity that endures over time. It could therefore be argued that the individual defined by this pattern does not exist for its component processes, which effectively means that it does not exist for itself. Moreover, as we pointed out above, if a network of chemical reactions does not exist for its component processes, whatever norms are established by the former cannot directly influence the latter. To be clear, for a network to “exist for” its component processes, there must be some sense in which it not only regulates their behavior but also constrains them to cohere as a unified network. Deacon calls this kind of constraint “hologenic” (whole-creating) as it acts specifically to unify processes so that they constitute and maintain a concrete individual—and also explains how such an individual comes to exist.

In addition, Deacon proposes that the very same hologenic constraint needed to constitute an individual is also required for the normative regulation of its behavior. The reason, he argues, is that for behavior to be truly normative it must be teleological in the following sense: its causal basis must involve an intervention, that is, a constraint that counteracts or otherwise alters spontaneous tendencies so that they are directed toward the capacity of self-preservation. This constraint is needed to distinguish genuine teleology from spontaneous tendencies toward increased entropy, that is, those dictated by the second law of thermodynamics and PMEP. As discussed in the previous section, within certain parameters these tendencies can be directed toward the maintenance of a self-organized structure. However, in a recent article, Deacon and García-Valdecasas argue that this alignment is fortuitous, and its short-term benefits should not obscure the fact that all self-organized structure is ultimately degraded by these very same tendencies. To highlight their true end, or rather to distinguish them from genuinely self-directed or “targeted processes,” they refer to such tendencies and the behavior they produce as “terminal processes” (García-Valdecasas, 2025; García-Valdecasas & Deacon, 2024).

Like the constitution of a genuine individual, then, genuinely normative behavior requires internalized control over the variables that determine how the spontaneous tendencies of an individual’s component processes are directed. In other words, these variables must be endogenous to the individual entity in question, as discussed in the previous section.

Teleodynamics, in a nutshell, is a theory of the conditions for meeting this requirement. As indicated in the figure below (see Figure 3), Deacon proposes that these conditions can be met if two complementary self-organizing processes become co-dependently coupled, ¹⁸ so that each produces the enabling and limiting boundary conditions for the other’s existence. Two self-organizing processes that meet these criteria are reciprocal catalysis, also known as an “autocatalytic set” (Kauffman, 2000), and molecular self-assembly, a crystallization-like process responsible for virus shell and cell membrane formation. The interplay between these two self-organizing processes is mutually reinforcing: reciprocal catalysis generates a high molecular concentration that facilitates self-assembly while, in turn, self-assembling processes constrain the diffusion of catalysts, maintaining the integrity of reciprocal catalytic cycles. Deacon calls this co-dependently coupled system of self-organizing processes autogenic because it generates its own conditions of existence and can reconstitute them when disrupted. The process is also teleodynamic—that is, genuinely teleological in the sense just defined—because it arises from a constraint that reorients its own spontaneous tendencies toward system maintenance and reconstitution amid constant material and energetic turnover.

Figure 3.

Image on the left (a) shows reciprocity of constraints produced by and required for two self-organizing processes: reciprocal catalysis (below) and self-assembly (above). Image on the right (b) shows how these processes are strongly coupled if reciprocal catalysis produces a side product that acts as a capsid forming molecule, preventing each from reaching terminal equilibrium. The constraint is called “hologenic” because it integrates and unifies processes as dynamic whole with properties of self-individuation and self-repair. Reproduced from García-Valdecasas and Deacon (2024, p. 75) under terms of the CC BY 4.0 license.

In relation to the two models discussed previously, Deacon’s autogen model of normative behavior can be viewed as a special case of both autopoietic individuality and PMEP-driven behavior. With respect to the former, it imposes the additional requirement of a strong, co-dependent coupling between two self-organizing processes, each constituting a boundary condition for the other, and constituting a “hologenic constraint” that holds together the entire system so that it resists disintegration. With respect to the latter, it imposes the requirement of internalized or endogenous control of boundary conditions, which is achieved by the very same co-dependent coupling. The “interventionist” nature of the resulting causal dynamic is clear, as each component prevents the other from advancing toward a “terminal” state (i.e., degradation) and directs the entire system toward self-preservation, thus constituting a causal dynamic that, according to Deacon, is teleological in nature.

It would seem, then, that Deacon’s autogen model offers a solution to the puzzle described at the close of the last section, as it brings together the constitutive dynamics of a norm-establishing individual with an “interventionist,” norm-following causal regime created by a new constraint. However, without detracting from the promise of this approach, we wish to point out two possible weaknesses of the autogen model.

First, it preserves a key assumption of the autopoietic model, namely that the most basic norms of life are norms of endurance. Strictly speaking, in Deacon’s model, the main locus and object of normativity is self-producing capacity of the hologenic constraint rather than the concrete individual that benefits from this capacity. ¹⁹ But the normativity of this capacity—and of life in general—is still related to endurance, and rooted in the logic of self-preservation. As we pointed out above, this assumption has been subjected to criticism (e.g., Barrett, 2017; Moss, 2024) and, in response, some enactive theorists have moved toward a more pluralistic, processual and “messy” view of normativity (Di Paolo, 2023; McGann, 2024a), but without venturing beyond norms of endurance. To be clear, no one doubts that norms of endurance are essential to life. The question is whether this is the only kind of normativity operative in life, and also whether it is the most basic. An ontogenetic perspective that turns to “novelty and transformation rather than stability and identity” (García, 2023, p. 4) would seem to call for a different kind of normativity. However, although Deacon’s autogen model accounts for the emergence of an individual as well as its self-maintenance, like the autopoietic model it does not indicate any way for the genesis and transformation of individuals to be normatively guided.

Second, although PMEP-driven behavior is included in the autogen model insofar as it contributes to the “terminal processes” of reciprocal catalysis and self-assembly, it does not seem to play any role in the norm-following behavior of the individual that emerges from their coupling. More precisely, it does not seem that PMEP is directed by internal boundary conditions to do anything except hold the system together. Accordingly, although the individual resists disintegration, it does not seem to be capable of making and acting upon finer discriminations of value. What is missing from this account is the remarkable capacity for spontaneous, gradient-like “dissipative adaptation” displayed by EDS and other dissipative structures. These structures do more than hold themselves together and repair themselves: they move about, coordinate, change shape to squeeze past obstacles, and carry out a variety of foraging behaviors (De Bari et al., 2024). Although the morphogenetic capacity of PMEP is acknowledged by Deacon in his discussion of Benard cells, his autogen model does not exploit this capacity as mechanism of adaptive morphological change.

These two shortcomings point to a potential problem with Deacon’s “interventionist” definition of the teleology required for normative behavior. For Deacon, the behavior of an entity is teleological only if it arises from constraints imposed by that entity on the spontaneous tendencies of its component processes. Once constrained, these spontaneous tendencies can do the work of norm-following, but constraints are responsible for the beneficial effects of their influence and it is this “intervention” that makes them normative. In the autogen model, these constraints are also responsible for the constitution of the system as a self-distinguishing individual, so that dynamics of norm-establishing and norm-following are integrated, as required for autonomy (Barandiaran & Egbert, 2014, p. 24). But it is still not clear that the resulting behavior is performed in the present because it is normative. We have established the criterion of present responsiveness to norms to distinguish genuine normative behavior—behavior that is normatively guided—from behavior whose normative status is merely accidental and from behavior whose normative status derives from a past history of natural selection (e.g., Millikan, 1989). Deacon’s account clearly excludes the former, but if on closer inspection it is revealed to be a sophisticated version of the latter, it would fall short of our definition of normative behavior.

Once again, the spontaneous tendencies of the autogen are normative—insofar as they contribute to maintenance of the whole—because of constraints belonging to the same autogen. But what accounts for the normative (self-preserving) nature of these constraints? Why these constraints and not others? At this point, Deacon’s explanation of the emergence of normativity turns to an evolutionary account, which goes something like this: Given a large enough population of molecular reactions, eventually a reciprocal relationship between processes of autocatalysis and self-assembly will obtain (as described above), and the resulting self-constrained unity will be “selected,” that is, become more prevalent in the population, because of its special capacity to maintain and regenerate itself. It is crucial to recognize that the “because” in the last sentence does not indicate any responsiveness to norms. On the contrary, like other evolutionary accounts, it indicates how behavior with normative status (with respect to self-preservation) can derive from entirely non-normative mechanisms of selection. We do not wish to dispute such evolutionary explanations as far as they go. However, our objective in this article is not to explain the normative status of certain behaviors, but rather the capacity to be normatively guided in the present, and this capacity seems to require a responsiveness to reasons that cannot be explained by evolutionary accounts alone.

Cannot we say that the autogen is constrained to be responsive in the relevant sense? Given that it effectively responds to disintegration by regenerating itself, the answer would seem to be Yes. But it is not clear that this responsiveness is sufficient. As we have just pointed out, although the autogen model includes an account of how PMEP is constrained to support this regenerative capacity, it seems to be missing the more versatile gradient-like responsiveness displayed by dissipative structures like the EDS, and which we take to be an essential trait of genuine normative behavior. Admittedly, this point is open to debate, as it implies a higher standard of responsiveness that we have not clearly defined. The best we can do is suggest that normatively guided behavior should do more than steer an entity away from disintegration: it should include an indefinite range of configurations and behaviors within the viable regime, all of which are responsive to norms.

The picture emerging from this discussion suggests that normative behavior depends on a combination of the spontaneous tendencies that constitute gradient-following behavior and the right constraints. Without spontaneous tendencies, we lose the characteristic responsiveness of living things, the generativity that allows them to do “in innumerable details” those things that are good for them. On the other hand, without proper determination of the constraints that give direction to these tendencies, whatever beneficial effects they have are fortuitous, and fortuity is not enough for genuine normativity. The same point applies to the constraints themselves, as they must be created so as to direct spontaneous tendencies in the right way. Insofar as constraints are themselves products of spontaneous tendencies, we seem to have a chicken-and-egg problem on our hands. Deacon’s autogen model appears to get around this problem, but it does so only by positing the emergence and selection of a molecular system that happens to have the right constraints. Perhaps this entity is a critical stepping stone toward the kind of adaptive behavior that is characteristic of life. But it has no means of adjusting its own constraints according to normative criteria, and its spontaneous responsiveness seems to be limited to regeneration.

Conclusion

Let us return once more to the challenge of normativity and consider what lessons can be drawn from the preceding analysis.

A basic kind of normativity that is essential to life—prudential norms related to endurance—is arguably intrinsic to precarious autopoietic individuals whose continued existence depends on the capacity to make continual adjustments to their dynamics of self-production. Note that it is not just precariousness but also the possibility of adaptive behavior that establishes this kind of normativity (Di Paolo, 2005). Practically speaking, norms of endurance only exist for precarious individuals capable of norm-following, that is, of discriminating and responding to differences of value or favorability with regard to their continued existence. An intrinsic connection between this kind of normativity and life is indicated by the way in which the simplest forms of life seem to “hope and strive for various things”: “in innumerable details,” they do those things that contribute to their persistence, integrity, and health. The challenge is to understand how norm-following works, to understand how this discriminatory capacity emerges, and to do so without positing consciousness or representational machinery.

Barandiaran and Egbert’s concept of normative field (2014) clarifies this challenge by defining relevant differences of value as a kind of gradient, allowing us to conceive of normative behavior as a kind of gradient-following. This is a promising move, as we are reasonably confident that a variety of gradient-following behaviors are achieved by relatively simple organisms without consciousness or representations. The normative field is not just any gradient, however: it is like a “master gradient” superimposed on the entire state space of the autopoietic individual. And although the consequences of norm-following are quite real, the normative field of an autopoietic individual remains “virtual” as long as it has no inherent causal or regulatory connection to the self-producing dynamics of that individual. Without this causal connection, we cannot account for the apparent responsiveness of organisms to the local contours of their own normative fields. The problem of norm-following is compounded by the fact that component processes of an autopoietic individual must be responsive to norms that apply at a different timescale, to the enduring pattern of their interrelation.

A possible way forward is indicated by the use of a “proxy value”: a variable to which component processes of autopoietic individual are, in fact, responsive and whose incremental increase effectively approximates the gradient of the normative field, even if only within a limited range of contexts. For example, in Barandiaran and Egbert’s model (2014), a simple metabolic variable suffices as a “proxy” for viability. In such cases, however, genuine normativity is achieved only if this gradient-following is autonomous, that is, if it is self-directed in the minimal sense of being integrated with the self-producing dynamics of the autopoietic individual.

Something like this solution seems to be realized by the dissipative “bio-analog” systems studied by ecological psychologists Benjamin De Bari, James Dixon, and their colleagues in the experiments discussed above. Simply by moving toward states with higher rates of entropy production (REP), in conformity with the Principle of Maximum Entropy Production (PMEP), these structures engage in a variety of adaptive behaviors—movement, coordination, foraging—that contribute to their own self-preservation. Thus, although these entities cannot be said to be responsive to their own normative field (as defined by prudential norms of endurance), they are responsive to a gradient that functions effectively as a proxy for this field.

There remains, however, the question of whether this thermodynamic gradient-following behavior is really self-directed and, in addition, the closely related question of whether it can be attributed to a genuinely self-distinguishing individual. According to García-Valdecasas and Deacon (2024), genuine normativity requires that spontaneous tendencies responsible for normative behavior are directed by constraints that are partly “internalized” by an autopoietic individual, such that we can say that the individual in question “intervenes” so as to alter the consequences of these tendencies. In addition, they argue that genuine individuality requires a stronger form of co-dependence between the component processes that make up an autopoietic individual: a “hologenic” (whole-producing) constraint that causes component processes to regenerate their coupling when it breaks down. As we point out above, however, it is not clear whether Deacon’s autogen model entirely captures the spontaneous responsiveness that is characteristic of life, and which we are proposing to describe as a kind of gradient-following behavior.

In sum, although each model makes important contributions to our understanding of normativity and the conditions for its emergence, each raises questions that remain unresolved. For our purposes, it will suffice to articulate the shared criteria that emerge from this comparison. We suggest that a minimal form of norm-following is achieved if: (1) a spontaneous gradient-following tendency belonging to the constitutive dynamics of an autonomous individual effectively approximates responsiveness to the contours of that individual’s normative field and (2) this adaptive gradient-following is directed by variables that are endogenous to the constitutive dynamics of the individual, so that the resulting guidance of behavior is genuinely autonomous.

The main contribution of this analysis is the suggestion that a basic kind of normativity that is essential to life—prudential norms related to endurance—can emerge “by proxy,” that is, by appropriating the spontaneous adaptivity of preexisting gradient-following behavior. With these criteria, we do not claim to have provided an authoritative account of the emergence of normativity. That depends on settling the questions that we have examined here. Nevertheless, we believe that these criteria, if accepted, considerably lower the bar for such an account. As argued here, the key trait of normativity is its spontaneous adaptivity: the capacity to avoid “bad things” and seek “good things” without needing to know beforehand what these things are. A proxy gradient achieves an approximate version of this capacity: it preserves the spontaneous responsiveness of normative behavior, albeit within a limited range of contexts, and it does so without requiring an implausible degree of sensitivity to relevant variables.

It could be objected that by turning to a proxy gradient we have given up on our thesis that normativity requires the capacity to discriminate value. What is discriminated by the organism is not really value (defined by the organism in relation to its viability conditions), but rather a variable that happens to function effectively as a proxy for value. However, if this appropriation is autonomous, we believe that our thesis can still be upheld. As long as responsiveness to the proxy gradient is directed by the self-producing dynamics of the individual, we can still say that the individual establishes this gradient as a value-proxy, much in the same way that it establishes its own normative field. Insofar as it can be said that an autopoietic individual enacts values related to its endurance, it also enacts potential proxies for these values. In the case of spontaneous tendencies that can serve as proxy value gradients, what is enacted is not the tendency but rather its “functional value” for the autopoietic individual.

It might also be objected that by resorting to proxy gradients we have embraced the kind of “representational machinery” that we were supposed to avoid, in conformity with the non-representationalist stance of enactive theory, ecological psychology, and other perspectives that make up the immediate context of our inquiry. But we think it can be shown that self-directed appropriation of gradient-following behavior is not representationalist in the relevant sense. Self-directed gradient-following is a form of active and direct engagement with the environment—a form of self-regulated dynamic coupling—and involves no predefined goal-states. Behavior driven by PMEP, for example, seeks only to increase entropy production whenever possible; it does not aim at a global maximum.

Finally, we wish to highlight a subtle ambiguity of our proposal with respect to the question of emergence. In keeping with enactive theory and teleodynamics, both of which define normativity in terms of endurance, it is possible to construe self-directed gradient-following as the emergence of normativity from non-normative physical processes. Alternatively, if we adopt a wider view of normativity, it is possible to construe this same phenomenon as the emergence of an important kind of normativity that is essential to life rather than the emergence of normativity per se. For the purposes of this article, the difference between these interpretations may not be very significant, as our stipulated goal was to explore the minimal conditions for behavior that is guided by prudential norms of endurance. But the difference could be more consequential for wider considerations of normative behavior in animals and humans, not to mention the problem of how to include experience within a naturalistic view of normativity. Even if norms of endurance are essential to life, that does not mean that they suffice to explain all living behavior, or that they are adequate to our experience of normativity.