I am grateful for all the commentators for engaging with the ideas of the target article, and offering comments and critiques to some of The Entangled Brain's central elements (Pessoa, 2022, 2023). I learned quite a bit and appreciate their contributions to the dialogue about how we might best study the brain.Several commentators shared some of my views and in fact captured related ideas persuasively. For example, Hayden (2023) writes incisively about problems with the arealization hypothesis, namely, the idea that area borders are the most important principle of functional organization. Uddin (2023) even suggests that localization-based accounts of brain function have largely gone out of style. That is certainly not my reading of the current state of the literature and, in all likelihood, not a notion shared by the commentators contributing to this Special Focus who found problems with my framework (Boone, Van Rooy, & De Brigard, 2023; Krakauer, 2023; Knutson & Srirangarajan, 2023; Wyble, 2023). For Uddin (2023), then, some of the important questions include “What constitutes a brain network?” and “How many brain networks exist?”. Sadaghiani and Alderson (2023) discuss how the contextual dependence of networks/circuits is both important and one of the most challenging problems that needs to be addressed. Both Druckmann and Rust (2023) and Jennings (2023) applaud some aspects of the proposal, but voice concerns, too. For example, Druckmann and Rust (2023) propose that experiments that target specific brain areas can be productive, especially when accompanied by careful behavioral observations. Jennings (2023) hints at a preference to consider processes that are on a spectrum, such as perception and action, as ultimately separable, a view that is clearly “not entangled.”In what follows, I will briefly try to address some of the concerns raised. At the outset, I should explicitly apologize for the very selective references to recent work. The literature on the topics below spans decades and even includes Nobel prizes (Ylia Prigogine and Giorgio Parisi). I focus on a few studies to emphasize recent or ongoing work, and to keep this response focused.Here, I would like to discuss a key challenge that often goes unacknowledged by neuroscientists who tend to focus on brain areas—the “arealists” as Hayden (2023) puts it—as the natural unit of investigation: What exactly is an area?The challenges of adequately defining “area” are formidable enough that even some arealists have recently stated that “[d]espite centuries of brain–behavior relationships, a clear formalization of the function of many brain regions…is lacking” (Genon, Reid, Langner, Amunts, & Eickhoff, 2018, p. 362). In fact, these authors propose that the goal of their article is to “lift the conceptual fog that has clouded structure-function relationships in the brain” (p. 350). Following Brodmann, they propose that “the a priori defined construct is the brain region and the unknowns are the behavioral functions associated with it” (p. 350). In other words, the defining criteria of a brain area are structural (e.g., cytoarchitectonic). Yet, it is far from clear how Genon and colleagues' contribution is able to “lift the conceptual fog.”To appreciate some of the still ongoing problems with the notion of brain area, let us consider perhaps the quintessential region outside of sensorimotor cortex: Broca's area, which is closely associated with Brodmann's areas 44 and 45 and linked to speech processing since the classic description of patient Leborgne (“Tan”) by Paul Broca. Despite more than 150 years of research, there is no definitive basis to define Broca's area, and multiple definitions coexist (Keller, Crow, Foundas, Amunts, & Roberts, 2009). A study by Amunts and collaborators suggested that Broca's area can be subdivided into more than a dozen subareas (Amunts et al., 2010), including subdivisions of areas 44 and 45, as well as multiple small areas in the depth of the inferior frontal sulcus, the frontal operculum, and at the transition with premotor cortex (area 6).Because these cortical territories are involved in multiple language and nonlanguage functions, Amunts and Zilles (2012) concluded that there are no robust functional criteria to define Broca's area unambiguously. More recently, Fedorenko and Blank (2020) have argued that hypotheses concerning Broca's area cannot be properly evaluated because it consists of multiple functionally distinct components. In fact, for them, a brain area should be defined functionally, in direct opposition to the call by Genon and colleagues (2018) that structure is the defining feature of an area (although the latter also believe an area has a “core” function, if only we can discover it). To add to the difficulties of the area concept, there is considerable evidence that large-scale gradients in structural organization (e.g., gene expression patterns, anatomical connectivity) are important organizational principles of human cortex (Huntenburg, Bazin, & Margulies, 2018), thus making the determination of more clear-cut borders problematic.The situation is even more daunting subcortically. Research in the past decade or so has revealed that an important principle of organization of subcortex is that classical areas contain multiple neuronal populations that can be defined via their chemical and/or genetic properties. These populations can be used to define micro-circuits that support functions in a cell-type-specific manner. For example, the central amygdala contains different subpopulations of neurons, and optogentically silencing one of them impairs fear extinction (Whittle et al., 2021). Thus, working out the functions of these microcircuits is key to understanding the mechanisms of fear extinction and when it fails (i.e., fear is not properly unlearned). Whereas the concept of circuits involving subpopulations does not in itself pose a problem to arealists, it reveals yet another functional level that needs to be considered in unraveling the putative “core” process of brain areas.In summary, arealists face considerable challenges in establishing stable criteria to understand how to parcellate the brain. This is not to say that such obstacles are insurmountable, but they show that the concept of brain area is far from settled.In the target article, some forms of emergence are discussed in terms of interactional complexity and illustrated in terms of Type I and Type II networks. A complementary way to illustrate some of the key points is to consider distributed processing across (potentially large) sets of brain regions.In a recent computational study, Mejias and Wang (2022) simulated neuronal activity across 30 cortical areas across the four lobes. Interareal connectivity was defined based on quantitative connectivity data from tract-tracing studies of the macaque brain. Parameters regulating the dynamics of individual areas were defined such that they did not generate sustained activity by themselves (this is consistent with empirical data from the mouse, in which thalamic input is required for frontal cortex to produce sustained activity; Guo et al., 2017). Nevertheless, after (simulated) stimulus withdrawal, activity persisted in multiple areas across frontal, temporal, and parietal lobes in a manner that agreed with empirical data. To further understand how distributed interactions support sustained activity, the authors induced simulated lesions. Inactivating most areas had a limited impact on the number of regions exhibiting sustained activity. However, the impact was large with virtual lesions of some temporal and frontal areas. These are areas that are particularly well-connected, and in fact have been proposed to be part of a “connectivity core” based on computational analysis (Markov et al., 2013).Overall, the computational model from Mejias and Wang (2022) shows how sustained activity can arise as a result of the dynamics of individual regions and multi-area interactions. This is a simple example of a property—sustained activity—that is novel with respect to what the individual regions are able to generate alone. Multi-area computational models also provide powerful ways to generate experimental predictions. For example, if an area displaying sustained activity is silenced, the activity levels of sustained activity in other areas should decrease (roughly, because sustained activity in all areas results from “mutual support” between areas). Empirical failure of predictions like this would provide support for the alternative hypotheses that the implementation of sustained working memory is more area-specific. In all, a multi-area proposal is no different from different proposals of how single-area mechanisms operate, and subject to potential falsification in the best Popperian fashion. Such proposals are, then, not an exercise of “seduction of pure philosophical speculation,” as suggested by Knutson and Srirangarajan (2023).To illustrate the high degree of interactional complexity in the brain, the target article describes some of the regions involved in fear extinction. The goal of the example is to suggest that processes like fear extinction should be understood at the network level because of the rich, bidirectional connectivity of the circuits that support it. The key point being made is not that the basolateral amygdala, say, functions all of a sudden in an entirely differently manner, the presumed “shapeshifting” property that Krakauer (2023) alludes to. Instead, the working hypothesis is that the contribution of the basolateral amygdala to overall brain function will be different in the context of fear learning versus when engaged in processes supporting fear extinction.Finally, what does it mean to say that regions do not instantiate specific functions? What do they do then? The argument is that they provide constraints such that the system-level property arises under specific circumstances (Bishop, Silberstein, & Pexton, 2022; Raja & Anderson, 2021; Juarrero, 1999). Consider the model by Mejias and Wang (2022) discussed previously. The physiological properties of the temporal, parietal, and frontal regions that most easily exhibit sustained activity are not “working memory,” although, under the right conditions, they exhibit this property. Individually, it is productive to think of the regions and their connections to other regions as providing constraints to the overall system: The system-wide activity (such as persistent activity in some areas but not others) is determined in particular ways by the system.How can researchers proceed? From the standpoint of studying specific tasks or conditions, building from Passingham, Stephan, and Kötter (2002), I have suggested that multivariate and distributed activity fingerprints provide useful summaries of evoked responses or states (Pessoa, 2017; Figure 1A). This type of summary description can be potentially very rich and immediately shifts the focus from thinking “this region computes X” to “this region participates in multiple processes.” By studying multiple related tasks/conditions and determining the relative commonality of engagement across regions, one can even test the extent to which “core” functions are instantiated by brain areas, for example, showing that regions RA and RB tend to participate together across some tasks/conditions (Figure 1B). As an additional recommendation, a complementary summary can be generated by characterizing the (multiple) functions of specific circuits of interest that can be summarized via a functional diversity profile (Anderson, Kinnison, & Pessoa, 2013; Figure 1C). For example, in the case of the basolateral amygdala, it would involve arousal, vigilance, novelty, attention, value determination, and decision making, among others.Boone and colleagues (2023), Knutson and Srirangarajan (2023), Krakauer (2023), and Wyble (2023) all express concern about the lack of compositionality of the Entangled Brain framework. Krakauer (2023) explicitly compares brain areas to brain organs (such as the heart and liver), whereas Wyble (2023) compares them to words. As Wyble (2023) states, “language is compositional, such that the meaning of a sentence is strongly (though not completely) determined by the individual words and the way that they are combined, using the rules of syntax and the semantics of the words.”Compositionality is certainly one way that individual elements of a system can be combined. In the case of language, with rich syntax and semantic properties operating, individual words can combine in exceedingly powerful ways. Despite that power, what this form of compositionality misses is what we have learned about complex, nonlinear dynamical systems in the past decades. These systems exhibit phenomena such as bifurcations, static and dynamic attractors, and self-organization, which can give rise to a wide range of dynamic behaviors (Strogatz, 2018). Nonlinear systems can also exhibit sensitivity to initial conditions, meaning that small differences in the initial state of the system can lead to vastly different outcomes over time (as in so-called chaotic systems). That is, in a nonlinear dynamical system, isolatable elements do not have invariant properties in the way that the semantics of a word are invariant (even in language, the role of pragmatics, including the context in which words are used and the intentions of the speaker, is very important). Thus, whereas some version of compositionality may prove satisfactory in explaining how the brain supports (some) behavior and mental states, it leaves out the possibility of richer scenarios. In fact, historically, computational neuroscientists have emphasized how the brain is profitably viewed in terms of complex dynamics (Grossberg, 2021).Given limitations in recordings, evidence that complex dynamics are observed in the brain is limited to examples like central pattern generators, or dynamics involving populations of cells in a specific locale in the brain; for a detailed review, see Khona and Fiete (2022). Nevertheless, a growing body of work supports the notion of complex dynamics in the brain. This work is computational in nature but makes close contact with experimental data, including in visual awareness and decision making. For example, consider a study by van Vugt and colleagues (2018) on the threshold for conscious report of a visual stimulus. Stimuli reported by the animals were associated with strong sustained activity in the frontal cortex, whereas frontal activity was weaker and quickly decayed for unreported stimuli. The experimental data were well explained by a simple computational model that considered both feedforward and feedback connections between visual areas and pFC (van Vugt et al., 2018). Critically, the model displayed emergent dynamics not localizeable to any one node. The large-scale model of Mejias and Wang (2022) described in the context of working memory was also used to simulate the awareness-related data. In the simulations of Mejias and Wang, the magnitude of the input to area V1 was used to reflect visual stimuli with varying physical contrast. Importantly, the simulations revealed graded responses in visual cortex as a function of input strength, but all-or-none activity in the pFC. In addition, analysis of the temporal evolution of activity revealed a dynamic signature of the onset of “ignition” that paralleled results observed with EEG in a human auditory detection experiment (Sergent et al., 2021). According to Wang (2022, p. 550), “[t]he global ignition phenomenon associated with access consciousness represents but one example of a variety of distributed brain processes that can now be rigorously studied experimentally and theoretically.”To conclude, I would like to discuss a final point, namely, that of the status of traditional labels and constructs in psychology, including standard mental domains (e.g., “perception,” “cognition,” “emotion”). Skimming through neuroscience textbooks, we often see chapters on perception, attention, memory, learning, language, motor control, executive functions, emotion, and so on. In The Entangled Brain, I proposed that we need to dissolve boundaries between mental domains, and in recent work, I have more forcefully discussed problems with common psychological constructs (Pessoa, Medina, & Desfilis, 2022). So, how should we organize neuroscience textbooks?A fruitful approach is the one adopted in Striedter's (2016) Neurobiology: A Functional Approach. The book focuses on problems that brains help animals solve, with an emphasis on neural circuits and systems, highlighting important evolutionary considerations. For example, in the chapter “Remembering Relationships,” sections cover what makes some memories stronger than others, how animals learn what is dangerous, and what happens when memories conflict with each other (such as the conflict of habit memories and episodic ones).I believe targeting problems that brains help animals solve holds great promise. Consider a potential chapter on “Selecting information from the world.” The benefit of this strategy is that a construct like selection can be applied across multiple traditional domains, including perception, cognition, and emotion. Doing so allows us to conceptualize the underlying processes as inherently cutting across domains. Complex, naturalistic behaviors also offer fruitful ways to organize material, such as in the case of threat assessment (Pessoa et al., 2022).If this type of approach is adopted, the overall organization of book chapters could be substantially more diverse than that found across many current neuroscience textbooks. This is not a problem, however, as there is no unique decomposition of brain and mind. As in textbooks, I propose that this framework offers a better way for neuroscientists to organize how they think about the functions of the brain and how the nervous system supports complex behaviors and mental states. For too long, our discipline has used constructs that emphasize independence and separation. It is now time to adopt a language that emphasizes interaction and integration so that we can advance understanding of the entangled brain.The author is grateful for support from the National Institute of Mental Health (MH071589 and MH112517) and Brad Postle for constructive feedback.Reprint requests should be sent to Luiz Pessoa, University of Maryland at College Park, 1147 Biology-Psychology Bldg., College Park, MD 20742-5031, or via e-mail: pessoa@umd.edu.Luiz Pessoa, National Institute of Mental Health (https://dx.doi.org/10.13039/100000025), grant numbers: MH071589, MH112517).Retrospective analysis of the citations in every article published in this journal from 2010 to 2021 reveals a persistent pattern of gender imbalance: Although the proportions of authorship teams (categorized by estimated gender identification of first author/last author) publishing in the Journal of Cognitive Neuroscience (JoCN) during this period were M(an)/M = .407, W(oman)/M = .32, M/W = .115, and W/W = .159, the comparable proportions for the articles that these authorship teams cited were M/M = .549, W/M = .257, M/W = .109, and W/W = .085 (Postle and Fulvio, JoCN, 34:1, pp. 1–3). Consequently, JoCN encourages all authors to consider gender balance explicitly when selecting which articles to cite and gives them the opportunity to report their article's gender citation balance.