In his book, The Cognitive-Emotional Brain, and and in his précis, Pessoa (Reference Pessoa2013) provides a rich and cogent overview of current emotion research, with a particular emphasis on human neuroimaging work and his own important contributions to the field. A major theme addressed from the start concerns the notion of “automaticity” in emotion processing and the elaboration of a “dual competition model,” whereby emotion and motivation are proposed to interact either synergistically or antagonistically with perception and executive functions. In line with laudable efforts throughout the book to deconstruct the boundaries between affective and cognitive realms of the mind-brain, as encouraged by other scientists in the past (Gray Reference Gray1990; Ledoux Reference Ledoux, Lane, Nadel, Ahern, Allen and Kaszniak2000; Leventhal & Scherer Reference Leventhal and Scherer1987), it is notable that Pessoa is embracing a perspective on the functional relationship between “controlled” versus “automatic,” or conscious versus nonconscious, processing of emotional information that appears less dichotomous than earlier views (Pessoa & Ungerleider Reference Pessoa and Ungerleider2004), and also less crude than ways in which these terms are commonly (unfortunately) used in the literature. Hence, Pessoa puts forward a “continuous framework” for the perception of emotional stimuli and their competition with other information in capacity limited systems.

However, as Pessoa tries to accommodate these general notions with existing data and refers to general psychological concepts and terms (e.g., resource capacity, multiple waves, recurrent loops, etc.), the proposed framework often remains at an abstract level, which makes it difficult to falsify and thus lacks a strong heuristic value. Moreover, Pessoa chose to refrain from providing his own definition for several important notions (e.g., emotion information or automaticity, to start with) – which is wise to avoid a priori biases but then leaves abundant space for ad hoc interpretations and potential inconsistencies. What is needed in the field is a precise mechanistic account of these concepts and the corresponding phenomena, based on specific neuronal processes and circuits.

There is abundant evidence that some form of emotional processing can take place without conscious awareness of stimuli or despite opposing intentional goals, which cannot simply be dismissed by stating that awareness was not probed or attention “exhausted” in an appropriate manner. These effects are subserved not only by the amygdala, but also by high-level cortical areas and other brain structures, including reward pathways (Pessiglione et al. Reference Pessiglione, Petrovic, Daunizeau, Palminteri, Dolan and Frith2008; Pourtois et al. Reference Pourtois, Schettino and Vuilleumier2013; Raio et al. Reference Raio, Carmel, Carrasco and Phelps2012; Vuilleumier & Righart Reference Vuilleumier, Righart, Calder, Rhodes, Johnston and Haxby2011). What is of interest and remains challenging is not really whether or not emotion-related information is processed without awareness of its presence, but rather what are the differences in processing, nature of representations, neural substrates both in space and in time, etc. For example, fear conditioning occurs without consciousness but shows different temporal features than during conscious processing (Raio et al. Reference Raio, Carmel, Carrasco and Phelps2012). Furthermore, as underscored by several researchers (Bargh Reference Bargh1989; Moors et al. Reference Moors, De Houwer, Hermans and Eelen2005) and acknowledged by Pessoa, automaticity is a broad concept that encompasses several distinct dimensions, including lack of explicit report, intentionality, effort, or control, among others. None of these dimensions is a unique defining feature of “automaticity,” and only some of them may apply to emotion responses (e.g., in amygdala) – as the resource efficiency principle described by Pessoa; but other dimensions may not apply – as lack of control or autonomy, for example (Dominguez-Borras & Vuilleumier Reference Dominguez-Borras, Vuilleumier, Armony and Vuilleumier2013; Vuilleumier & Righart Reference Vuilleumier, Righart, Calder, Rhodes, Johnston and Haxby2011). An important goal for future research is to more precisely dissect these dimensions in relation to different kinds of emotion information and different brain circuits. In the light of past research in neuroscience and cognitive science, to which Pessoa has made great contributions, it is hard to argue against or in support of “automaticity” in emotion processing without defining more precisely what is meant by automatic and emotional.

Another issue is that many facets of automaticity and their underlying substrates are not necessarily specific to emotion, but also apply to nonemotional material. Various forms of nonconscious, unintentional, efficient, and fast processing have been observed in some degree for a large variety of stimuli, for example, ranging from subliminal priming with words or numbers, classic Stroop effects, through to motor preparation and mimicry (Dimberg et al. Reference Dimberg, Thunberg and Elmehed2000; Eimer & Schlaghecken Reference Eimer and Schlaghecken2002; Kouider & Dehaene Reference Kouider and Dehaene2007). These effects imply activations of high-level brain areas several synapses away from primary sensory areas, including motor cortex, hippocampus, language, or even prefrontal executive control areas (Henke et al. Reference Henke, Treyer, Nagy, Kneifel, Dursteler, Nitsch and Buck2003; Nakamura et al. Reference Nakamura, Dehaene, Jobert, Le Bihan and Kouider2007; van Gaal et al. Reference van Gaal, Ridderinkhof, Scholte and Lamme2010). There is no necessity to attribute a “special” privilege to the amygdala in this respect, neither to deprive it.

What might be “special” is a particular threshold to respond to a particular kind of information and the functional impact on particular downstream areas, for example, associated with attention or autonomic arousal (Vuilleumier Reference Vuilleumier and Phelps2009). Nonconscious or preattentive processing should not be equated with the implication of a subcortical route specialized for emotion information – just as nonconscious word priming does not imply direct subcortical connections to Wernicke's area. It has long been proposed that such effects may reflect efficient, fast, feed-forward sweeps of inputs into pathways overlapping at least in part with (cortical and subcortical) pathways subserving conscious processing, but presumably unaccompanied by other neural processes engaged in the latter case such as reentrant feedback, sustained activity, rhythmic oscillations, and/or additional inputs through the same or parallel pathways (Vuilleumier Reference Vuilleumier2005). This echoes the notion of “multiple waves” of processing put forward by Pessoa and Adolphs (Reference Pessoa and Adolphs2010), although the latter account is rather broad and needs to be more specific in order to be testable. The role of a specific subcortical route (e.g., through colliculus, pulvinar, or visual thalamus) is necessary to explain affective blindsight in patients without visual cortex (Hamm et al. Reference Hamm, Weike, Schupp, Treig, Dressel and Kessler2003; Morris et al. Reference Morris, DeGelder, Weiskrantz and Dolan2001; Pegna et al. Reference Pegna, Khateb, Lazeyras and Seghier2005), but its role in the healthy brain, exact anatomy, and informational content remain to be elucidated (Schmid et al. Reference Schmid, Mrowka, Turchi, Saunders, Wilke, Peters, Ye and Leopold2010).

In any case, it must be underscored that preattentive or nonconscious processing (cortical or subcortical) should not be expected to be similar to “full” conscious processing of the same sensory inputs. It might be limited to coarse or partial information, such as low-frequency cues or eye features in faces (Whalen et al. Reference Whalen, Kagan, Cook, Davis, Kim, Polis, McLaren, Somerville, McLean, Maxwell and Johnstone2004) (among other possibilities), and thus possible for some stimuli but not others. For example, emotional meaning might be extracted efficiently from faces because of particular features or configural properties present in a single stimulus, thanks to specialized neuronal populations tuned to faces that exist throughout the visual systems. This seems much less likely for visual scenes where emotional meaning derives from specific layouts of more complex, multipart objects. Accordingly, there is evidence for coarse representations of faces (Gabay et al. Reference Gabay, Nestor, Dundas and Behrmann2014; Maior et al. Reference Maior, Hori, Tomaz, Ono and Nishijo2010; Nguyen et al. Reference Nguyen, Matsumoto, Hori, Maior, Tomaz, Tran, Ono and Nishijo2014) or biologically salient stimuli such as snakes (Van Le et al. Reference Van Le, Isbell, Matsumoto, Nguyen, Hori, Maior, Tomaz, Tran, Ono and Nishijo2013) in subcortical visual pathways, although their functional role is unresolved. This accords with the notion that nonconscious processing might be “dumb,” as Pessoa puts it (see also Vuilleumier et al. Reference Vuilleumier, Armony, Driver and Dolan2003; Vuilleumier & Righart Reference Vuilleumier, Righart, Calder, Rhodes, Johnston and Haxby2011). An important goal for research is to pinpoint more precisely which information is encoded in which brain region, for which function, and under which conditions.

Furthermore, the existence of nonconscious and unintentional responses does not preclude that they may be modulated by top-down factors, including expectations, context, goal-driven control, and so forth. Subliminal priming is malleable to current task demands and attentional control even when stimuli remain nonconscious (Bahrami et al. Reference Bahrami, Lavie and Rees2007; Fischer et al. Reference Fischer, Schubert and Liepelt2007; Martens et al. Reference Martens, Ansorge and Kiefer2011; Nakamura et al. Reference Nakamura, Dehaene, Jobert, Le Bihan and Kouider2007). Even exogenous attention, which typically operates in a reflexive (rapid and involuntary) manner, can be modulated by top-down factors related to task goals or readiness to respond to particular events (Ansorge et al. Reference Ansorge, Kiss and Eimer2009; Folk et al. Reference Folk, Remington and Johnston1992). Likewise, expectations and affective states influence emotion processing for stimuli outside attention (Bishop et al. Reference Bishop, Duncan and Lawrence2004; Pichon et al. Reference Pichon, Miendlarzewska, Eryilmaz and Vuilleumier2015) and can abolish the effect of attention load on emotional response (Cornwell et al. Reference Cornwell, Alvarez, Lissek, Kaplan, Ernst and Grillon2011). These top-down modulations do not necessarily imply that stimulus processing depends on consciousness and voluntary attention and hence cannot be interpreted against “automaticity” without distinguishing more precisely between different sources of modulation and different components of automaticity.

Although a continuous framework of resource competition neatly account for number of phenomena (Pessoa Reference Pessoa2013), it is not sufficient to explain how reduced resources as a result of increased attention load can affect emotion-specific responses in some brain areas (e.g., prefrontal cortex or visual cortex) without affecting others (e.g., amygdala) (Shafer et al. Reference Shafer, Matveychuk, Penney, O'Hare, Stokes and Dolcos2012; Vuilleumier et al. Reference Vuilleumier, Armony, Driver and Dolan2001), unless one postulates dissociable sensitivities to these effects. Moreover, in some cases, increased attentional load or suppression from awareness may actually increase emotional responses (e.g., in amygdala) to neutral or positive stimuli relative to low-load conditions (Silvert et al. Reference Silvert, Lepsien, Fragopanagos, Goolsby, Kiss, Taylor, Raymond, Shapiro, Eimer and Nobre2007; Williams et al. Reference Williams, Morris, McGlone, Abbott and Mattingley2004). Further, a continuous framework does not take into account that key aspects of selective attention and awareness are nonlinear in nature (Dehaene et al. Reference Dehaene, Charles, King and Marti2014; Sergent & Dehaene Reference Sergent and Dehaene2004) (and presumably emotion too; Sander et al. Reference Sander, Grandjean and Scherer2005). Finally, the central concept of resource is relatively vague and lacks precise neural substrates. By focusing on effects rather than causal mechanisms, a resource account runs into the risk of circularity, for example, when appealing to “residual resources” to explain why emotional effects on behavior or amygdala activity are observed under “high load” conditions, without a more direct measure of resource. Whereas in vision, competition for resources can be mapped neurally onto overlapping receptive fields (Desimone Reference Desimone1998), other forms of competition may exist in other brain systems and be resolved by distinct mechanisms. Models of emotion and cognition interactions need to consider that multiple processes operate in parallel and produce distinct (linear or nonlinear) effects on different nodes within distributed networks (Pourtois et al. Reference Pourtois, Schettino and Vuilleumier2013; Sander et al. Reference Sander, Grandjean and Scherer2005).

Lastly, caution must be taken when drawing conclusions based solely on human neuroimaging studies, which have several limitations because of their poor resolution (spatial and temporal) and vascular origin (BOLD contrast in fMRI). This makes it hard to compare processing conditions when they differ in terms of the onset or duration of neuronal responses (Pourtois et al. Reference Pourtois, Spinelli, Seeck and Vuilleumier2010) or recruit partly distinct neuronal subpopulations within the same brain structure (Zhang et al. Reference Zhang, Schneider, Belova, Morrison, Paton and Salzman2013). In particular, the amygdala is not homogenous but made of several subnuclei, each containing multiple type of neurons, which might be sensitive to different sources of sensory inputs and top-down modulations (Vuilleumier Reference Vuilleumier and Phelps2009). Conscious (vs. nonconscious) and attentive (vs. preattentive) processing might also be characterized by distinctive patterns of rhythmic oscillatory activity or connectivity within and/or between subregions, whose impact on BOLD fMRI or other electrophysiological measures is unresolved. Answering these questions will require finer investigations in animal models or novel technologies in humans.

The issues reviewed by Pessoa (Reference Pessoa2013) should encourage researchers to go beyond simplistic dichotomies (such as automatic vs. controlled processes). However, it is important to be careful in how psychological terms are used, defined, and related to specific neural substrates. Time is now ripe to formulate precise mechanistic hypotheses in order to elucidate the exact functional circuits implicated in emotion phenomena and define them in terms of information processing systems (i.e., by determining which type of information is represented and/or transmitted in specific pathways – rather than just the “amount” of information as in a continuous resource model). This is an exciting prospect for future research.