Once consciousness is analyzed as “efficient network activity,” it is manifested across a broad range of systems, and its meaningfulness as a concept becomes diluted. That is, as Merker et al. (forthcoming) successfully show, a fundamental challenge for the integrated information theory (IIT) of consciousness (Tononi, Reference Tononi2008; Tononi, Boly, Massimini, & Koch, Reference Tononi, Boly, Massimini and Koch2016).
As a way forward, we propose that measures of information integration can more straightforwardly be interpreted as measures of agency rather than of consciousness. Because agency can be defined in terms of behavioral patterns, it avoids the problems arising from quantifying “first-person perspective” properties by means of “third-person perspective” measures. As the conceptual dual of consciousness, agency may, thus, deserve a more prominent place in consciousness research.
Agency and integration. Agency is increasingly of interest to biologists, as many developmental patterns and behaviors (including those of plants) are characterized in agential terms. It is part of a trend to assign a greater theoretical role to organisms as such in our understanding of evolution: Organisms do not simply passively undergo evolutionary processes but actively shape their selective environment (Laland, Matthews, & Feldman, Reference Laland, Matthews and Feldman2016) and respond in a goal-directed manner to opportunities or “affordances” in their environment (Walsh, Reference Walsh2015). At a very general level, “agency” refers to how organisms exhibit goal-directed behaviors in response to environmental change.
Talking about goals in this way activates old worries about teleology and anthropomorphism (i.e., agency as human-like intentionality). However, in practice, the pay-off for explaining behaviors as goal-directed lies in accounting for patterns of behavioral robustness: An organism's “goal” is simply what it attempts to achieve through various means, even when it is perturbed or challenged by an environmental change. In other words, agency refers to how a (1) small number of goals can account for patterns of connectivity between (2) a large number of possible environmental states, and (3) a large number of possible behaviors.
This explanatory structure describes a bow-tie architecture (see Fig. 1) where environmental states and behaviors are integrated in virtue of the “goals” present. Note that the figure does not illustrate any fine-grained connections between environmental states and behaviors. What it does illustrate is the explanatory general structure of agency, where the “goals” are used to explain how environmental states and behaviors are informationally integrated (for an information-theoretic treatment of scientific explanation, see e.g., Desmond, Reference Desmond2019).
Fig. 1.The bow-tie architecture of agency.
When the issue is put in this way, suggestive parallels with theories of consciousness emerge. The IIT posits consciousness as the integration of various experiential properties (Tononi, Reference Tononi2008), but also global workspace theory (Dehaene & Changeux, Reference Dehaene and Changeux2011, p. 11) posits a similar structure, where consciousness is a global broadcast mechanism, integrating input and output systems.
However, unlike theories of consciousness, an integrated information theory of agency would not need further justification of why such bow-tie architectures should be identified with agency. Goal-directedness is a third-person concept and in this way it can be unproblematically fleshed out in terms of input–output patterns. Agency as a concept just is a type of pattern of connectivity between environmental states and behaviors. There is no need to posit a counterpart to “qualia” or some ineffable subjective quality.
The underlying reason for this is that agency is an explanatory concept rather than one that refers to an empirical state-of-affairs. Informally, agency could be said to be more similar to “Newton's law of inertia” rather than to “white snow.” If one is habituated to thinking of agency in terms of intentionality (or the presence of some form of mentality), this view of agency may require a gestalt-switch. The concept of agency imposes a structure on observed behavior, and if the observed behavioral patterns do not exhibit this general structure, there is simply no need to describe them as “agential” (see discussion in Desmond & Huneman, Reference Desmond, Huneman, Altobrando and Biasetti2020).
Agency and consciousness as duals. Whether or not the apparent isomorphism between theories of consciousness and the structure of agency is more than skin-deep is not a question we systematically address here. Instead, we offer a general rationale why the appearance of such isomorphisms should not be surprising. If agency refers to the “activity” of the organism in relation to the environment, consciousness in its broadest sense denotes the “passivity” of the organism. A synonym for consciousness – sentience – makes this passivity more apparent: The capacity of “feeling” refers to how an organism “undergoes” its environment (think also of “e-motion”: being moved). Agency and consciousness are different sides of the same fundamental coin of organism-environment relationality. One cannot have activity without passivity, and vice versa.
In mathematics, dual concepts are used to integrate two different ways of looking at a same object (Atiyah, Reference Atiyah2007). Similarly, agency and consciousness can, at a fundamental level, be viewed as “duals.” And just as invoking the dual operator in mathematics may help solve otherwise intractable problems, perhaps some of the challenges facing our understanding of consciousness can be addressed by invoking agency. For instance, it is likely that the evolution of consciousness can only be understood by simultaneously understanding how agency evolved. This is reflected in how greater sensorimotor control has evolved in tandem with various proxies of consciousness such as cognitive systems (Godfrey-Smith, Reference Godfrey-Smith2020; van Duijn, Keijzer, & Franken, Reference van Duijn, Keijzer and Franken2006).
Could an IIT of agency avoid the equivalent of panpsychism, which seems unavoidable once consciousness is naturalized and/or de-anthropomorphized? Panpsychism's dual is “panagentialism”: seeing agency everywhere. In other contexts, this has been called hyper-agency detection (cf. Atran, Reference Atran2002). We believe panagentialism can be more easily defused, because of a subtle asymmetry between agency and consciousness (at least as the latter is typically understood). Attributing agency is an explanatory strategy to make sense of behavioral complexity – not a statement about the ontological makeup of the world. Panagentialism is thus simply a (poor) explanatory practise. Note that there may also be “no facts of the matter” regarding consciousness (Carruthers, Reference Carruthers2020). In that case, panpsychism would also be a poor explanatory practise, and agency and consciousness would be true duals.
Once consciousness is analyzed as “efficient network activity,” it is manifested across a broad range of systems, and its meaningfulness as a concept becomes diluted. That is, as Merker et al. (forthcoming) successfully show, a fundamental challenge for the integrated information theory (IIT) of consciousness (Tononi, Reference Tononi2008; Tononi, Boly, Massimini, & Koch, Reference Tononi, Boly, Massimini and Koch2016).
As a way forward, we propose that measures of information integration can more straightforwardly be interpreted as measures of agency rather than of consciousness. Because agency can be defined in terms of behavioral patterns, it avoids the problems arising from quantifying “first-person perspective” properties by means of “third-person perspective” measures. As the conceptual dual of consciousness, agency may, thus, deserve a more prominent place in consciousness research.
Agency and integration. Agency is increasingly of interest to biologists, as many developmental patterns and behaviors (including those of plants) are characterized in agential terms. It is part of a trend to assign a greater theoretical role to organisms as such in our understanding of evolution: Organisms do not simply passively undergo evolutionary processes but actively shape their selective environment (Laland, Matthews, & Feldman, Reference Laland, Matthews and Feldman2016) and respond in a goal-directed manner to opportunities or “affordances” in their environment (Walsh, Reference Walsh2015). At a very general level, “agency” refers to how organisms exhibit goal-directed behaviors in response to environmental change.
Talking about goals in this way activates old worries about teleology and anthropomorphism (i.e., agency as human-like intentionality). However, in practice, the pay-off for explaining behaviors as goal-directed lies in accounting for patterns of behavioral robustness: An organism's “goal” is simply what it attempts to achieve through various means, even when it is perturbed or challenged by an environmental change. In other words, agency refers to how a (1) small number of goals can account for patterns of connectivity between (2) a large number of possible environmental states, and (3) a large number of possible behaviors.
This explanatory structure describes a bow-tie architecture (see Fig. 1) where environmental states and behaviors are integrated in virtue of the “goals” present. Note that the figure does not illustrate any fine-grained connections between environmental states and behaviors. What it does illustrate is the explanatory general structure of agency, where the “goals” are used to explain how environmental states and behaviors are informationally integrated (for an information-theoretic treatment of scientific explanation, see e.g., Desmond, Reference Desmond2019).
The bow-tie architecture of agency.
When the issue is put in this way, suggestive parallels with theories of consciousness emerge. The IIT posits consciousness as the integration of various experiential properties (Tononi, Reference Tononi2008), but also global workspace theory (Dehaene & Changeux, Reference Dehaene and Changeux2011, p. 11) posits a similar structure, where consciousness is a global broadcast mechanism, integrating input and output systems.
However, unlike theories of consciousness, an integrated information theory of agency would not need further justification of why such bow-tie architectures should be identified with agency. Goal-directedness is a third-person concept and in this way it can be unproblematically fleshed out in terms of input–output patterns. Agency as a concept just is a type of pattern of connectivity between environmental states and behaviors. There is no need to posit a counterpart to “qualia” or some ineffable subjective quality.
The underlying reason for this is that agency is an explanatory concept rather than one that refers to an empirical state-of-affairs. Informally, agency could be said to be more similar to “Newton's law of inertia” rather than to “white snow.” If one is habituated to thinking of agency in terms of intentionality (or the presence of some form of mentality), this view of agency may require a gestalt-switch. The concept of agency imposes a structure on observed behavior, and if the observed behavioral patterns do not exhibit this general structure, there is simply no need to describe them as “agential” (see discussion in Desmond & Huneman, Reference Desmond, Huneman, Altobrando and Biasetti2020).
Agency and consciousness as duals. Whether or not the apparent isomorphism between theories of consciousness and the structure of agency is more than skin-deep is not a question we systematically address here. Instead, we offer a general rationale why the appearance of such isomorphisms should not be surprising. If agency refers to the “activity” of the organism in relation to the environment, consciousness in its broadest sense denotes the “passivity” of the organism. A synonym for consciousness – sentience – makes this passivity more apparent: The capacity of “feeling” refers to how an organism “undergoes” its environment (think also of “e-motion”: being moved). Agency and consciousness are different sides of the same fundamental coin of organism-environment relationality. One cannot have activity without passivity, and vice versa.
In mathematics, dual concepts are used to integrate two different ways of looking at a same object (Atiyah, Reference Atiyah2007). Similarly, agency and consciousness can, at a fundamental level, be viewed as “duals.” And just as invoking the dual operator in mathematics may help solve otherwise intractable problems, perhaps some of the challenges facing our understanding of consciousness can be addressed by invoking agency. For instance, it is likely that the evolution of consciousness can only be understood by simultaneously understanding how agency evolved. This is reflected in how greater sensorimotor control has evolved in tandem with various proxies of consciousness such as cognitive systems (Godfrey-Smith, Reference Godfrey-Smith2020; van Duijn, Keijzer, & Franken, Reference van Duijn, Keijzer and Franken2006).
Could an IIT of agency avoid the equivalent of panpsychism, which seems unavoidable once consciousness is naturalized and/or de-anthropomorphized? Panpsychism's dual is “panagentialism”: seeing agency everywhere. In other contexts, this has been called hyper-agency detection (cf. Atran, Reference Atran2002). We believe panagentialism can be more easily defused, because of a subtle asymmetry between agency and consciousness (at least as the latter is typically understood). Attributing agency is an explanatory strategy to make sense of behavioral complexity – not a statement about the ontological makeup of the world. Panagentialism is thus simply a (poor) explanatory practise. Note that there may also be “no facts of the matter” regarding consciousness (Carruthers, Reference Carruthers2020). In that case, panpsychism would also be a poor explanatory practise, and agency and consciousness would be true duals.
Financial support
This research was funded by the ANR-DFG funded project “The Explanatory Scope of Generalized Darwinism: Towards Criteria for Evolutionary Explanations Outside Biology.”
Conflict of interest
The authors declare that there is no conflict of interest.