Investigating Emotions as Functional States Distinct From Feelings

Abstract

We defend a functionalist approach to emotion that begins by focusing on emotions as central states with causal connections to behavior and to other cognitive states. The approach brackets the conscious experience of emotion, lists plausible features that emotions exhibit, and argues that alternative schemes (e.g., focusing on feelings or on neurobiology as the starting point) are unpromising candidates. We conclude with the benefits of our approach: one can study emotions in animals; one can look in the brain for the implementation of specific features; and one ends up with an architecture of the mind in which emotions are fully accommodated through their relations to the rest of cognition. Our article focuses on arguing for this general approach; as such, it is an essay in the philosophy of emotion rather than in the psychology or neuroscience of emotion.

Introduction

All empirical and theoretical approaches to the study of emotion face a fundamental challenge: where and how does one begin in deciding the subject matter? What counts as relevant, and as part of the phenomenon to be explained, and what should be bracketed or omitted? Our position begins with a subset of all the different phenomena related to emotions that people often talk about; we do not attempt to tackle all of them. At least at the outset, we thus offer what amounts to a promissory note. We are wagering that, if you stick with our approach, you will at the end of the day have a scientifically useful theory of emotions. It may not address all aspects of emotion that one would eventually want to understand, but it addresses enough of them that we can all agree that the wager paid off: that we have a scientific theory that is indeed about emotions (as opposed to having changed the topic to something else), and that accounts for a substantial portion of our commonsense view of emotion. This at least is our conviction.

Our wager can be introduced as follows. Suppose an alien spaceship lands on earth. The aliens, intelligent and skilled in doing science as they are, have none of the prior concepts about the mind that we use, notably also lacking any concept of conscious experience. What science would they now construct in order to explain the complex behaviors of people and animals that they find on earth? We posit that they would come up with two broad classes of explanations: those that do not require latent variables (akin to our concepts of mental states) and those that do. The former would explain many kinds of movements, like reflexes. The latter would explain complex behavior. Among the latter would be a collection of internal states (which could be partitioned in ways rather different from our current science) that are distinguished by particular features or properties that the aliens regularly observe. Whereas we have attention and memory and fear and sadness, the aliens might employ other concepts to do the job. One class of these internal states, a class prominently continuous across all animal species including humans, would have a particular set of features or properties (which we elaborate on further in what follows; see Figure 3), and these would roughly correspond to emotions. The moral of this story is that, in our view, emotions emerge as a particular class of latent variables that are required by any science of the complex behaviors observed across higher animals. The particular categories or dimensions by which to distinguish different emotions might vary (indeed, we think they likely will, as we suggest further below), but the generic category would be a requirement of any science of the mind, however mental states are conceived.

Figure 3.

Table of preliminary list of the features of emotions.

To look ahead to the rest of this article, there are three key points: (a) a science of emotion is constructed from observable data (whatever the aliens could measure), but not, at least in the first instance, by how emotions feel (the conscious experience of having one); (b) emotions are internal states (latent variables) of the same broad type as cognitions (i.e., functionally individuated states); (c) but emotions are distinguished from the rest of cognition by specific operating characteristics or features (much the same as attention, memory, and other cognitive state types are distinguished from one another).

As the starting point for a science of emotion, then, we are imagining all the same phenomena that we observe in other people and animals all the time. We are simply asking what kinds of explanations one might come up with if one did not have specific prior beliefs about emotions. We are preserving all of the objective phenomena to be explained (and, again, bracketing conscious experience), and in particular all of the behavioral and physiological observable events that we would normally use to infer emotions: facial expressions, overt motor behaviors, autonomic and endocrine activity, and so forth. In this picture, we find it plausible that emotions would emerge as a class of internal states that provide causal explanations of behaviors at a particular level of control and complexity: a level intermediate to that of reflexes and deliberated behavior. This level of control, in turn, is distinguished by a set of operating features that we can begin to list (and that the aliens would similarly list). Further below, we provide such a preliminary list (Figure 3) and suggest that it can be used to infer computations that characterize emotions.

In what follows we elaborate this view and consider some of its implications. However, we emphatically do not offer a theory of emotion in any sense. We are here interested not in arguing for specific functional roles that emotions subserve (let alone that specific emotions like fear or anger might subserve), but rather in arguing that a functional account is a promising type of account for a science of emotion in general (as an approach). We thus devote most of our article to laying this groundwork, suggesting some specifics only in the penultimate section.

Bracketing the Conscious Experience of Emotions

Our approach is of course a rather bold one, since most people, and many psychological theories of emotion, explicitly (or even exclusively) focus on conscious experiences of emotions. But as we suggest in the next section, there are good reasons to think that feelings cannot provide a principled distinction that carves out emotions from other kinds of conscious content; and feelings create other problems as well, such as methodological ones in studying emotion in animals (or constructing them in robots).

Bracketing feelings means just that: for now, we leave them aside, as a practical matter in order to make scientific progress. We are thus proposing that there exists a level of basic mechanism that can be used to understand emotions without recourse to the feelings caused, or otherwise involved, in emotion states. We think it likely, in fact, that feelings are an important part of the overall picture, a part that eventually needs to be explained. But this does not mean that they need to figure at this stage in a science of emotion. It is even possible that a scientific theory of emotions that begins by bracketing feelings might eventually provide a new explanation of feelings. At a minimum, leaving aside conscious experience would surely seem to make any scientific investigation a more tractable task than including it: there is no agreement on any approach to understanding consciousness, and there remain serious proposals claiming that it cannot be understood at all. Leaving an obviously difficult aspect of emotion aside—again, for the time being and as a practical matter—thus seems a wise strategic choice. It should be noted that many emotion theories in fact do distinguish emotions and feelings (e.g., Damasio, 2003; LeDoux & Brown, 2017), although in general this is because they argue the two are different things. We are not advancing any argument about the possible relation between emotions and feelings—we are simply omitting feelings as one aspect for purely methodological reasons.

Bracketing conscious experience is in fact a common strategy in any project that takes mental processes and attempts a functional, mechanistic, or specifically neuroscientific explanation of them. We take this approach all the time; for instance, in vision science as in memory research (Adolphs, 2017). Visual processes in both humans and animals have usually been studied without reference to conscious visual experience (yet allowing, in principle, for an enrichment that would account for it). Why not do the same thing for emotion?

It might be argued, however, that conscious experience is more easily detached from the central function of vision or memory than from emotion. When we consider conscious visual or mnemonic experience, we are (so the argument might run) generally concerned with what Ned Block (1995) has called “access consciousness”: awareness of seeing or remembering X, and the ability to report this and to make use of it in deliberation. We are often not concerned with the qualitative feel of seeing or remembering X in these cases (what Block calls “phenomenal consciousness,” and what is often referred to as “qualia”). By contrast, when X is the object of an emotion, it seems that the qualitative “feel” that goes with emoting about X seems to take precedence.

Perhaps relatedly, it might also be argued that the aspect of consciousness that is most prominent in the case of vision and memory is a higher order awareness that one is having a visual or mnemonic experience. And again, in the case of emotion experiences, the “raw feel” of the emotion seems to take precedence over the awareness that one is emoting about X (although Joe LeDoux has recently suggested otherwise; LeDoux & Brown, 2017). Moreover, the fact that different emotions seem to carry a distinctive “feel” can be thought to be what differentiates emotional episodes from other kinds of mental episodes, and to differentiate categories of emotion, such as fear, anger, happiness, and so forth, from one another (a view whose utility we argue against further in what follows).

For now, these purported disanalogies between the prominence of conscious experience in the case of emotion, on the one hand, and in the case of vision or memory, on the other, may weaken the appeal to the success of vision and memory research, but do not directly argue against our approach. The mere fact that people tend to include qualitative conscious experiences of emotion in their concept of emotion is certainly no argument. As we will argue further in the following lines, there is every reason to think that conscious experiences of feelings cannot provide a scientific basis for distinguishing emotions from other mental states, or even different types of emotion from one another.

What Are the Alternatives?

It may help to locate our approach in relation to other views. We do not undertake a detailed review of emotion theories, but sketch the most important alternatives to our view, and the difficulties that we believe they face. Andrea Scarantino has surveyed philosophical theories of emotion as falling largely into three approaches that take different aspects of an emotion as foundational or primary in some sense (Scarantino, 2016). These three are appraisal, feeling, and motivational theories, which, respectively, emphasize the evaluative, experiential, and behavioral components of an emotion. We comment further on appraisal and motivational approaches next, which we find congenial but pegged at a different level of explanation than the present article (see From Emotion Features to Functional Role section). So far, we have argued for bracketing the conscious experience of emotions, but one might well wonder how the story would look if instead of bracketing feelings, one took them to be foundational.

Feelings are taken as the starting point for the layperson’s concept of emotion, and also in many theories of emotion. But the attempt to classify emotions in virtue of how they feel immediately brings with it a host of problems about how to measure or characterize feelings. The bulk of the difficulties in finding any kind of consistency between emotional feelings and either physiological patterns or neural activation patterns is focused on the difficulty of distinguishing among different emotional feelings—categorizing happiness, sadness, fear, anger, and so forth as distinct categories (for reviews regarding neural activations and physiological patterns, respectively, see Lindquist, Wager, Kober, Bliss-Moreau, & Feldman Barrett, 2012; Siegel et al., 2018). But there are problems even at a coarser grain: it’s also unclear how to distinguish emotions, generically, from other states. Pain and hunger, for instance, also seem to be characterized by feelings that share a lot in common with emotions: they are valenced and they motivate behavior. It is very unclear how to find a principled difference that would distinguish emotions from these other states by feeling only. In the absence of a principled distinction in feelings between these states and emotions, it becomes ad hoc to say they feel different. Indeed, the problem is deeper than that: what is it about emotional feelings that makes them emotions, as opposed to some other internal states? Saying that these feelings are accompanied by other phenomena, which in turn serve to individuate them as emotions, of course begs the question since, then, it is these other phenomena or criteria that are doing the work. The burden is on the feelings-first theorist to provide a positive account of some sort.

There is another obvious problem in taking feelings as foundational as well, methodological in nature and with severe consequences for a science of emotion: focusing on feelings makes it currently impossible to study the topic in nonverbal subjects. Infants and aphasic subjects, as well as of course all nonhuman animals, become impossible to study for a science of emotion that puts feelings first because nobody can agree on a reliable dependent measure for feelings other than verbal reports (especially in the case of more finely differentiated feelings that could serve to classify emotions). Worse than that, it becomes quite problematic how to compare measures across people who speak different languages that use different words to describe their feelings.

Indeed, we believe that if feelings are chosen as the place to start, we will be left with a hopelessly anthropocentric, culturally relativistic exercise. Even if we choose for these reasons to constrain our science of emotion just to college-educated, Western, English-speaking human adults (as much of psychology has in fact done), we end up, as a starting point, by choosing to solve what is arguably the most difficult problem that a science of the mind faces: we have chosen to aim for a theory of conscious experience. Why start there? It seems like the last place to get any traction, and the first place to run into large conceptual problems right at the start.

There is another interesting modern view, often curiously intertwined with a feelings-first emphasis: the view that neuroscience can provide the needed conceptual purchase to individuate emotions. This road is of course inspired by the success of neuroscience, and its power to disambiguate emotion theories, two facts we wholeheartedly acknowledge. For instance, neurobiologists like Jaak Panksepp link emotions in a strong sense to the operation (and phylogenetic conservation) of specific brain systems (primarily subcortical systems in his scheme, which implement his list of specific basic emotions; Panksepp, 1998). There may indeed be strong continuity in the neural systems for emotions across species if we stick to mammals, but this approach is unlikely to make sense as foundational if we wish to understand the emotions of octopuses, robots, or aliens.

A different starting point builds on the idea that our current emotion concepts are hopelessly confused and we need a fresh start. Why not look just at data from the brain, and use a more data-driven approach from neuroscience to bootstrap a theory of emotion? Insofar as most neuroscientists do not offer clarification on what they mean by the term “emotion,” they are probably guilty of fueling this approach as well. Perhaps the clearest statement of it from an emotion theorist can be found in Lisa Feldman Barrett’s work. On our reading of Barrett, a psychological theory of emotion should fall out of attention to the neuroscience facts and her recent book (Barrett, 2017) accordingly spends most of its pages detailing anatomical and physiological findings from neuroscience. She is certainly not alone in an emphasis on neuroscience, but we feel that there are a number of problems with this approach as foundational.

First, one might naturally question whether we know enough about the brain to do this. Currently, our knowledge of any brain, and certainly of the human brain, is woefully slim. But even if we knew more about the brain, as surely we eventually will, it remains quite unclear how and whether we could derive any facts about the mind, or about categories of behavior, that would in any way allow us to recover a notion of “emotions” (or, for that matter, any other cognitive states) from neuroscience data alone. The reasons are well known: knowledge of internal processes alone cannot allow us to infer what they evolved for and what they can do for an organism. It is as impossible to figure out the function of invertebrate central pattern generators merely from recording their neurons as it is to figure out a microprocessor that was designed for playing computer games merely from measuring its circuits (Jonas & Kording, 2017). Without some anchor in the world (and, we would argue, in the natural history of interactions with the world), it is either very difficult or outright impossible to make sense of what the brain does (but see below).

Second, there is the old conceptual problem of reference or intentionality. If we restrict ourselves to data from the brain alone, there is no grounding that could link the brain’s internal processing to the world. What makes the brain an information processing system cannot be understood without stepping outside the brain in order to see what the information is about. We need to know what stimuli trigger changes in the brain, and in what way behavior is more than mere motion. Ethological data that observe animals and humans in their natural environment are essential not merely for methodological reasons (as one might think of them in providing constraints for the previous paragraph), but serve a deeper need to make representation possible at all.

It should be emphasized that we are of course not saying that neuroscience is irrelevant, or that emotions do not fully supervene on the neurobiological details. But we just have no way to describe emotions as mental states unless we can refer to their function in relation to causes and effects in the environment; indeed, to the natural history of such causes and effects. Our aliens might very quickly acknowledge that all the behaviors that they observe are proximally caused by events in the nervous systems of animals; but we would argue that they would not begin by studying the brain and neglecting ethology. Philosophers have come up with various thought experiments to drive home this point. A simple one from Donald Davidson suffices here (Davidson, 1987). Consider sunburn, which is a state of my skin. The dermatologist may well be incapable of distinguishing true sunburn (caused by the sun) from an identical skin condition caused by a chemical burn. But one is sunburn, and the second is not. The fact that the sunburn is identified as the state that it is (sunburn and not chemical burn) doesn’t make it a different physiological state of the skin, but without this external reference we are unable to individuate it as sunburn—we cannot do it only by looking at a piece of skin. So, in our view, it is perfectly compatible to believe in the functionalism that we will detail further in the next section, yet also to maintain that emotions are just states of the brain. They are individuated by their function (which, in the broad external sense generally requires knowledge of their natural or engineered history) but identical with a particular physiological state of the brain.

Returning to Barrett’s view, we briefly comment further on three of its core features: the primacy of neuroscience data; the confusion of current emotion concepts and terms; and the flexible construction of emotions from other ingredients. We discussed before our reasons for not believing that neuroscience data can be foundational in and of itself. Yet there is one possibility that should be mentioned. It is possible that eventually we know so much about cognitive neuroscience that finding activation of particular brain systems when people are in emotion states may tell us about the functional role of such states because we know so much about the corresponding brain systems. This still raises the question of how to delineate when somebody is in an emotion state in the first place, but one could imagine a very rough characterization, or indeed one based on self-report, that is subsequently refined by the neuroscience data. It is possible that Barrett has something like this in mind in sketching brain systems out of which emotions are constructed: if we know enough about the other functions of such brain systems (appraisal, attention, motivation, etc.) we might get insight into the functions of emotion as well (Barrett & Satpute, 2013). It is possible that other thinkers, such as Luiz Pessoa, have in mind something along these lines as well in their emphasis on brain networks that mediate interactions between emotions and cognition (Pessoa, 2013, 2018). At this stage, it is unclear to us how useful such a strategy is—we would generally feel that we still know far too little about the brain to use such data in formulating a theory of emotion. It should be the other way around: the theory of emotion should make testable hypotheses, which can then be investigated with neuroscience.

Turning to a second core feature of Barrett’s view, we actually agree with Barrett and others that current emotion concepts, and in particular words and concepts for specific emotions (fear, anger, and so forth) are highly problematic. One good indication of the problem is that most emotion theorists have their own list of emotions. Ekman has his list of basic emotions; Panksepp has his; and there are other accounts as well (Tracy & Randles, 2011). We believe that we do not yet know how to classify emotions into specific emotions, or whether that will even turn out to be the most useful way to characterize their variability (perhaps dimensional approaches will save the day).

Yet our view emphatically disagrees with the third, and perhaps most fundamental, of Barrett’s views: that emotions are constructed. By this she means not only that they are compositional in some sense (a view with which we agree if it just means that their mechanisms can be investigated), but that the construction is highly variable and individual. Indeed, the view that emotions are constructed seems to be Barrett’s main reason for arguing that current words and concepts for them are problematic: if they can be constructed so flexibly that anybody can create a new emotion, then views for a scientific taxonomy seem doomed. We believe that there is such a scientific taxonomy (although it may offer dimensions rather than categories), but we also believe that we have not yet found it, and that the current schemes advocated are unlikely to be good candidates. While Barrett believes that we know so much about the neural basis of emotion that we can see that emotions are flexibly constructed in each case, we believe that we know so little about emotions that current schemes are likely wrong—but that there will be a future scheme that science can reveal.

To summarize this section: very broadly, we see three approaches to providing foundations for emotion. The most obvious one is to anchor emotions in conscious experience; we have chosen instead to bracket feelings (again, not to eliminate them or necessarily exclude them forever). Another currently popular approach is to hope that neuroscience can give us all the answers, especially if we also believe that current concepts are infused with a hopeless folk psychology and just need to be eliminated. We have argued that this is just impossible—but we are in agreement that our current terminology will likely require substantial revision. Instead of looking to neuroscience as the foundation, we advocate looking to functionalism, and subsequently using neuroscience data to test more refined functional hypotheses. Finally, a third broad approach indeed uses functional criteria: appraisal theories and motivational theories both fall into this category. However, these are functional role theories of emotion that characterize emotions as having specific functional features (and, in general, do so for specific emotions like fear, anger, and so forth). We are not proposing any such theory here, although towards the end of this article we will suggest an approach for how to further develop a specific functional-role theory for emotions. There we will argue that one could do so by listing reasonable properties of emotions (including aspects of appraisal and motivation; see Figure 3 for our provisional list). But we will also argue that it is likely premature to commit to a strong and specific theory that chooses only one or a small number of emotion properties as foundational (e.g., appraisal, motivation, valence). We think that there will be numerous features that emotions exhibit by which they achieve their functions (most or all of the ones listed in Figure 3) and the challenge will be how to combine them into a computational architecture. Again, we feel this is a large and difficult project for the future, and it is not our current project here.

Functionalism and Emotions

Our current project, then, is to provide a broadly foundational approach that is methodological. How would you go about creating a theory of emotion, how would you interpret empirical observations? (How would our aliens do it, observing behaviors amongst species on earth?). We see emotions as functional mental states in the sense intended by classical philosophy of mind: as relational, that is, individuated by the web of causal relations they have to all other internal states, as well as to stimuli and behaviors. We thus shun any views that try to locate emotions essentially as physical properties at a particular level (“essentialism,” a term we avoid here since it can be confusing). In a nutshell: emotions are individuated by what they do, not by how they are physically constituted.

We also refrain from any metaphysical claims about what emotions “really are”: we are not committed to claiming that they really are nothing but place holders in the causal web. We are not defining or characterizing emotions (again, we are not providing a theory of emotion), but describing an approach to investigating them and theorizing about them. Our stance is methodological, sparing us the need to counter the objections that have been raised against psycho-functionalism as a metaphysical thesis. Thus, for example, arguments purporting to show that a functionalist theory of mental states cannot for metaphysical reasons account for the “feel” that some of those states (such as emotions) carry, do not compromise our position. There may be features of emotions that don’t directly fall out of their functional role; we think it wise at this point to refrain from taking a bet one way or another. On the other hand, we certainly allow for the possibility that the same functionally defined emotion state could occur in animals or robots with very different hardware, as long as the structure of cause–effect relationships that defines the emotion state is preserved.

This classic view has an immediate and attractive consequence for the scientist: it tells us how to infer an emotion, namely, from observing its causes and effects. Thus, physiological dependent measures, including neurophysiological ones, as well as facial expressions and overt behaviors can all serve as evidence on the basis of which an emotion state can be inferred. Similarly, particular circumstances and environments can all serve as part of the evidence on the basis of which an emotion state can be inferred. And this is exactly what we do all the time in everyday life: we regularly infer emotions in other people and in animals from their observed behavior, or from observing the situations in which they find themselves, or both. This skeleton schematic is shown in Figure 1. Needless to say, the inference of emotions from observable measures is shared with many emotion theories, even though they may not be functionalist in nature.

Figure 1.

Skeleton schematic for a functional role of emotion states. The figure underemphasizes the full functional role of emotions, which would include causal connections (both as inputs and outputs) not only to many other cognitive states, but also to other emotion states.

Note. Modified from (Adolphs & Anderson, 2018).

Of course the story is not always so simple, and there are at least three complicating factors. First, the behaviors caused by an emotion can, at the next time step, become stimuli in their own right; so there are feedback loops to incorporate into the picture—emotions are temporally extended processes. Second, there is substantial regulation, at least in adult humans, that complicates how emotions are caused by stimuli. (It is a large and interesting further question to characterize how such regulation might also be found in other animals, such as trained dogs, and how it might differ from the volitional regulation of emotion seen in adult humans.) Third, the links from stimuli to emotions, and from emotions to behaviors, are extremely diverse: many kinds of stimuli can cause a certain type of emotion state, and a given type of emotion can cause many different behaviors. This “fan-in, fan-out” architecture is exactly the reason why behaviorism was unable to solve stimulus–response relations, and exactly why science (and evolution) needed internal states such as emotions that could serve to bind together a large range of different stimuli and a large range of different behaviors.

Psycho-functionalism,¹ then, provides a place to start, but it does not yet say anything useful about how exactly to distinguish emotions from any other cognitive state, nor how to distinguish different types of emotions. So the next question is: what is the functional role of emotions, and what is the functional role of, say, fear, that makes it a different emotion than, say, disgust? We do not answer these questions here but sketch how one might go about answering them. One would need to focus on the function within the general scheme of the creature’s life and ecology. In other words, we need to understand the biological significance of the emotion viewed as a set of relations in the web of mental states, stimuli, and behavior. Our usage of the term “biological significance” here is also functional, but in a different sense than we have been using “functional” so far. Biological significance requires a functional account that is broad and external: what functional role does the emotion, operating in an organism, subserve in order to produce behavior that is adapted to that organism’s environment? Evolutionary psychology has attempted some such characterizations, but these are difficult questions for which we desperately need more empirical data. As so often is the case in biology, the analysis of proximal causes, in Ernst Mayr’s terminology (Mayr, 1961), needs to be complemented by the consideration of distal or ultimate causes, those of evolutionary history (these are also what could explain mistakes or errors in how an emotion functions in any given instance). Whatever functions it is that emotions carry out in this broad and external sense, they likely evolved, they show considerable phylogenetic continuity, and they were mostly adaptive in the ancestral environment in which they evolved.

There has to be a story to tell about how simpler organisms like worms and flies first evolved states that might be considered the precursors to emotions, and how these evolved into the emotion states that we see, with remarkable homology, across many mammals and other higher animals today. A functional role from the perspective of evolution, then, is what careful comparative and ethological work would bring to the table. Using such data to construct psychological theories is extremely difficult, a hurdle with which the entire field of evolutionary psychology has needed to battle. But the fact that it is difficult should not detract from its necessary programmatic role in a theory of emotion. It is only by gleaning an account of the natural history of emotions, however incomplete and inaccurate it might be, that we could begin to answer the broad functional question about emotions, what they are “for.”

To date, there are merely a handful of functional accounts along these lines that have been fleshed out in any detail. For instance, there are models of disgust and models of fear, each of which are grounded in specific aspects of adaptive function (avoiding toxins, evading predators), and that can then be elaborated as a complex web of subprocesses. Indeed, for each of these emotions, the models typically fractionate the initial emotion into subtypes: there are varieties of disgust (for food, for other people, for moral acts; Tybur, Lieberman, Kurzban, & DeScioli, 2013) and varieties of fear (such as anxiety, fear, and panic in threat imminence theory; Mobbs, Hagan, Dalgleish, Stilson, & Prevost, 2015). There are also broader accounts that cut across emotions, such as motivational and appraisal theories, which differentiate emotions according to their focus on these properties.

Is it a requirement of our view that all emotions must have evolved? Of course not; we are merely saying that this will be one large aspect of understanding the functional role of emotions in humans and animals. If we want to understand the emotions of robots (which, being functionalists, we certainly believe are in principle possible) we would not look to evolution—but the kind of explanation to the question of what the function is “for” would have the same flavor. Need all functional roles of biological emotions have evolved? No, even this is not necessary—we explicitly note that culture and learning are a very major feature of emotions. But even here, the learning that takes place, and the constraints on it, link back to an evolutionary story that provides the grounding.

An account of the provenance of the functional role of an emotion, whether evolutionary, evolutionary + learned, or by human engineering, is required to know what that functional role is. Consequently, a molecule-by-molecule duplicate of one of us would not have emotions because there is no history whose reference could provide an answer to the question of what any functionally individuated state is for. Our molecular duplicate would be indistinguishable in all behavioral and physiological respects, and would of course claim to have emotions, to remember things, and to recognize things. But we would be unable to tell if it is remembering or recognizing anything for the same reasons that we could not tell if it is emoting: it never encoded or cognized anything in the past, and its emotional behaviors never had any past stimuli to cause them or past consequences (this situation might quickly change once our duplicate becomes embedded in a history of causal interaction with the world, however brief).

Finally, we again follow the tradition of classical cognitive science in thinking of mental states in general, and emotions in particular, as computational states, so that we can help ourselves to a multilevel description along the lines proposed by David Marr (Marr, 1982): at the highest level of abstraction, the function of an emotion is the biological role it plays within the web of mental states, stimuli, and behaviors (“broad” or “external” functional role; what we called “biological significance”); at the next level, that function is computationally realized by a certain algorithm or set of algorithms (our focus on a functional approach to emotion); and further down, that algorithm is actually implemented by some physical process that very much depends on the physical substrate in which the process is taking place (in most cases, neuroscience; see Figure 2). Each of these levels can be further decomposed into sublevels of increasing concreteness and so this scale is more continuous than tripartite.

Figure 2.

Levels of abstraction in a functional account of emotion.

Note. Note that it is possible to view all of these as functional through and through. Even the implementation level could be construed as functional: it is solely through their causal properties that any neurobiological events implement algorithms. Modified from Adolphs and Anderson (2018).

Finally, it is important to note that emotions are states of an entire organism. Single brain structures, such as the amygdala, cannot possibly implement an emotion, since that is not their function. The functions of the amygdala can only be defined in relation to the inputs to, and outputs from, this structure—and those inputs and outputs are not sensory and behavioral, but rather pertain to other brain regions. A neural system that processes emotion would then be analyzed compositionally, with each neural component contributing a particular functional, algorithmic, and implementational part that, in the whole system, functions to instantiate an emotion. Thus the Marr-like scheme shown in Figure 2 can be applied across all levels of scale, from organisms (or indeed societies) to brain networks to brain structures to microcircuits.

Challenges

There are a number of challenges faced by the approach we have sketched. One common misconception is that a theory of emotion should be sufficient, by itself, to explain how stimuli result in behaviors. That is, if our aliens came up with a theory of emotions in their science, following the functional approach we just discussed, should they now be able to explain emotional behaviors solely with this theory of emotion? Clearly not, and no theory of emotion could, or should, provide that kind of explanation. We do not require this of theories of memory, perception, attention, or even of decision-making. None of these processes, by themselves, are sufficient to explain behavior: they each contribute a part. The same thing holds for emotions. There are a host of enabling conditions, and a host of other cognitive processes with which emotions interact, that make the links from stimuli to behavior possible. Only a full mental architecture, complete with many other cognitive and physiological processes, could produce a full mechanism for the behavior.

This point may also speak to our disagreement with Lisa Barrett’s view of emotions as constructed from many other cognitive processes. Arguably, if one feels the need to have an emotion to be a state so comprehensive that it is sufficient to explain behavior, one would need to incorporate all other relevant processes. But this seems counterproductive to us: of course, emotions interact with all other cognitive states (see Pessoa, 2018). But they are not constituted by these processes—we want to individuate them as distinct from other cognitive states. Making distinctions is essential if we want to understand mechanisms.

Consider for instance the fleeing from a predator caused by a state of fear. It makes sense to explain an animal’s (or a person’s) fleeing as caused by the fear, and it makes sense to look in the brain for the neural mechanisms of fear processing. But the state of fear merely causes a motivation to get away from the predator (amongst other effects). It does not itself provide the detailed description of how this should be done. Depending on the terrain, the lighting, obstacles in the way, I may turn left or right, go fast or slow, and the details of how I flee will be offloaded to other sensorimotor processes and are not a part of what the emotion state needs to explain.

Another challenge is generalizability to other emotions. The good specific examples we alluded to in this article seem restricted to a few emotions like fear and disgust. What about love, embarrassment, pride, awe, and many others? Does our functionalist approach work for all emotions? This is a challenge only in the sense of saying that there’s more work to do, but does not seem to us to provide an actual argument for why it should not work (again, bracketing conscious experience). In Figure 3, we outline features of emotions generically, not with any specific emotion in mind. Are there clear counterexamples, a state that everybody agrees is an emotion but which does not have most or any of the properties listed in Figure 3?

To be sure, there are unclear cases, such as spontaneous moods or the feelings induced, say, by listening to music that moves us. These may well lack some of the features listed in Figure 3, and they certainly seem to lack functional descriptions from an evolutionary perspective. It is hard to say what adaptive role they serve, for example. In our view, it is quite unclear whether to call such states emotions; but we are also not strictly wedded to the requirement of a functional role that needs to be explicable in terms of evolution (see previous lines). If they do not exhibit the functional roles in Figure 3, and if they do not play an adaptive functional role that evolution, or our alien visitors, would have seen—well, then the only basis for calling them emotions would seem to be how they feel to us, and we are back to the problems of conscious experience. One could conclude from this that our approach failed to include an important set of emotions, because it bracketed conscious experiences. Or one could conclude that feelings are not a good basis for determining category membership for emotion states, and that, in the absence of other evidence, a feeling of awe induced by music may simply not be an indicator of an emotion state. We are inclined to believe the latter, but more empirical work is needed.

From Emotion Features to Functional Role

The task ahead for a science of emotion is progressively to unpack the picture we sketched here. Needless to say, most of the work remains to be done. While most functionalist theories of emotion in fact already propose specific functional roles for emotions (or for specific emotions), we have not done so here. Here we briefly want to sketch a possible approach to this question. We return to our thought experiment of the aliens observing animal behavior on earth. How might they begin to approach the question? If they located emotions as functional states that explain behavior, without mention of conscious feelings, what properties would these functional states exhibit? In particular, what properties of emotions would distinguish them from all other mental states?

The broadest way to frame this question is to ask how emotions carry out the ecological roles that contribute to the survival of an animal; that is, their “broad” or “external” functional role that gives them biological significance. As we noted, answers to this question are extraordinarily difficult to come by. Another way to frame the question is to ask what properties emotions as commonly conceived show (i.e., start with some clear, ordinary examples). There are of course many properties of emotions that they share with cognitive states in general, but the idea would be that one could come up with a list that would serve as a sort of package that could help to individuate emotions as distinct from other cognitive states. Ideally, this list would span some of the levels shown in Figure 2: they would speak to the ecological (broad) function of emotions, they would suggest algorithms by which that function could be accomplished, and they would motivate hypotheses that could be tested with neuroscience methods.

In Figure 3, we have proposed a preliminary list of features or properties that are intended to apply generically to all emotions (although the particular values some of these parameters take may serve to distinguish different emotions). This list of features is intended to distinguish emotions, generically, from other cognitive states. The list is likely incomplete, and not every instance of an emotion needs to exhibit all of the features in the list (but it should exhibit most of them; more detailed discussion can be found in Adolphs & Anderson, 2018; Anderson & Adolphs, 2014). We generated this list by searching through examples of what we would normally consider bona fide emotions, and asking what properties they exhibit empirically. Consequently, the features are a rather heterogeneous set of attributes. Some of them are probably higher order properties that emerge from the functional architecture in which an emotion state is embedded. Some arise from physical constraints at the implementation level—from how neurobiological systems can actually operate to implement a computation.

This set of features motivates specific neurobiological hypotheses: they suggest ways of implementing the features. Thus, we could search in the brain for specific features, such as persistence. And we could see if we can distinguish different types of emotions on the basis of their implemented features. For instance, we might expect neural systems that process fear to have shorter time constants than those that process sadness, and we might search for emotions that are hypothesized to differ in their degree of coordination and generalization.

Our list of features and their emphases bear some resemblance to other conceptual work on emotion. For instance, Andrea Scarantino’s view, the motivational theory of emotion (Scarantino, 2014), derives partly from earlier work by the psychologist Nico Frijda (Frijda, 1986) and sees emotions as tendencies or dispositions for action. This view is not uncommon, and has a lot of prima facie appeal. Like the list of features we provided, Scarantino locates emotions as motivating behaviors while exhibiting certain properties: flexibility, impulsivity, and bodily underpinnings. The motivational theory of emotion, like appraisal theories of emotion, fleshes out detailed stories that serve to distinguish different emotion categories. The particular relational goals towards which emotions motivate behavior serve a similar purpose in Scarantino’s theory as the core relational themes do in Richard Lazarus’s appraisal theory (Lazarus, 1991; see also Sander, Grandjean, & Scherer, 2018).

However, unlike these other thinkers, we do not take our list of emotion features to be foundational. Generically, we take the functional approach we described before to be foundational for emotion. Specifically, we take the list of emotion features as helping to point us in the direction of a theory of emotion, and of specific emotions. Particular functional roles and particular computations will correspond to the entries in Figure 3.

We briefly elaborate on one broad feature that can be abstracted from Figure 3 and that may be the most diagnostic of emotions, that helps to situate them at a level of behavioral control between that of reflexes and volitional behavior, and that motivates neurobiological hypotheses about what to look for in the brain. This broad feature might be called “semiflexibility” and to some degree subsumes the features of persistence, generalization, and automaticity in Figure 3. One might also call it “weak modularity” under an updated version of Fodor’s notion of modularity (Spunt & Adolphs, 2017). (Although we do not elaborate on it here, the rough characterization would be that emotions are strongly domain-specific, weakly cognitively penetrable, and not at all informationally encapsulated.)

Emotion states are semiflexible in the sense that they accept a limited variety of inputs innately (but a much larger variety of inputs through learning), span a limited time interval, and produce a limited variety of outputs. As we stated, this is in contrast with both reflexes (one kind of input, one kind of output, the entire process occupying a fixed, very brief interval that dispenses with the need for an internal state altogether) and fully reflective behavior (no preset limit on kinds of inputs or outputs, no bound on the duration of the process). It is a large open question, requiring considerably more empirical data, to understand what sets the permissible range of these parameters for emotions (and hence, what, at least in part, helps to distinguish emotions from either reflexes or fully flexible cognition).

We can distil some further points from Figure 3. First, the range of permissible values that these features can take is not rigidly preset, but is itself subject to some flexibility, as learning from experience, education, and volitional effort and training can, and do, modify the initial values to some degree. Second, the set of permissible values depends on the specific emotion and is subject to interindividual differences. This second point offers substantial purchase for neuroscience: we can, in part, distinguish different types of emotion states on the basis of the kinds of inputs they can take or how long they last (e.g., sadness lasts a long time, whereas surprise or fear are more ephemeral), and we may, in part, be able to characterize mood disorders by their abnormal parameter values (they might accept inputs that they should not and last longer than they should, for example). Third, as a first approximation, we may think of a type of emotion state as something akin to the “scripts” or “scenarios” postulated by early work in artificial intelligence and cognitive psychology, which were computational schemes in which certain typical inputs generated certain typical outputs, in a typical time sequence (Shank & Abelson, 1977). For instance, such scripts comprised rules for what to do when going to a restaurant. “Typical” allows for quite a lot of variety: there are indeed many kinds of restaurants and many ways of complying with the explicit and implicit rules of going to a restaurant. But that variety is restricted—this was in fact held against the idea of using such scripts as a reasonable model of cognition; but it is exactly what we think characterizes emotions. One might thus think of these scripts as similar, at the algorithmic level, to the “core relational themes” that classical appraisal theories have put forth (Lazarus, 1991).

In summary, while provisional, a list of emotion features derived from observation (like in Figure 3) could be used to begin assembling a theory of emotion that explicates the functional role of emotions. Such a theory might add or jettison specific entries in Figure 3 as more data are accumulated, and would aim to sculpt a core set of features, a package, that can serve to individuate emotions.

The Task Ahead for Philosophy and for Cognitive Neuroscience

The science of emotion has followed a somewhat erratic path over the last decades, leading to an inordinate amount of confusion even by the standards of an emerging discipline. Can we hope now to make progress towards a shared framework from which to undertake the empirical work? From the perspective we have suggested here, it seems, first, that philosophy of emotion should take the naturalistic turn more resolutely. While it has directed much of its attention to the phenomenology of emotion experience (for a summary, see Solomon, 2009) and to the conceptual analysis of the structure of emotions (for a recent lucid analysis and illustration, see Tappolet, 2016), it would profit from a deeper engagement with the relevant sciences, on a host of issues. These include the provenance of emotions, at both the developmental and evolutionary scale; an account of how errors might arise, for example, in psychopathology; and a taxonomy of emotions based on distinctions in functional role. Perhaps most importantly, we need to aim for a theory that explains not only the distinction between, but also the interaction among emotion and cognition, types of mental states that, we suggest, are both functionally individuated (see Pessoa, 2018).

Neuroscience, in turn, should focus on how the processing features of emotions listed in Figure 3 are implemented (and perhaps help to revise the list). The cognitive neuroscience of emotion needs to link broad functional accounts with algorithmic ones, aiming to provide computational accounts of what emotions do and how they do it (see Bach & Dayan, 2017). Such accounts can then be used to propose neurobiologically testable hypotheses; and of course, the neuroscience findings will help constrain the accounts. However, we resist the idea that neuroscience data, in and of themselves, and especially at this stage of our meager knowledge of brain function, can provide a foundation for a theory of emotion. The foundation has to come from elsewhere—although, of course, neuroscience should be in the dialogue. Success in forging a science of emotion will require tapping the resources of ethology, evolutionary biology, and paleo-anthropology, so as to pinpoint homologies across species and construct evolutionary scenarios. While the foundation for a theory of emotion should lie in a science of behavior, and thus psychology, our (and the aliens’) version of this has a different emphasis from most current versions, in particular in ignoring conscious experience.

We thus advocate a thoroughly interdisciplinary science of emotion that includes philosophy, ethology, psychology, neurobiology, and cognitive science. Calls for interdisciplinary work are notoriously easier to issue than to follow, and this is where philosophy can help, this time in the guise of philosophy of science rather than philosophy of mind. Its role here is not, as previously stated, to propose a theoretical framework in the tradition of philosophical psychology, but rather to inspect the actual procedures and results of the research programs relevant to the study of emotions, and bring to light the structure of the field, its tenets, its inferential structure, its conceptual repertoires, its real or apparent disagreements. Philosophy of science can contribute in this way to a dissolution of misunderstandings and a gradual convergence, and clear articulation, of the research programs involved, as it has in other scientific areas. Nor is this task the exclusive province of professional philosophers: the entire community can and should contribute. Naturalism also consists in fostering the direct involvement of philosophers in the scientific process, and of scientists in the philosophical search for intelligibility.