4.1.  Functionalism does not work for taxonomy

The first objection I will tackle is the one by Kron ([2019]). Kron claims that functionalism does not provide a good candidate to construct a scientific taxonomy of emotions. In his view, empirical research must be able to select manipulations and stimuli for experimental settings. Yet, he argues, functional principles do not enable researchers to select these manipulations and stimuli. Put differently, the claim is that we cannot carry out said selection based purely on functional profiles. Instead, attempts to carry out this selection based on functional principles end up using feeling-based criteria.

Kron sets up his argument on a distinction between terminological and epistemic changes to classification systems. A terminological change is one where researchers merely change the names or terms used to refer to a class of phenomena they are interested in studying. When terminological change occurs, the extension of the phenomena of interest remains intact. An example of terminological changes to taxonomy is LeDoux’s shift from neural circuits of emotions to survival circuits (LeDoux [2012]; LeDoux and Brown [2017]). According to LeDoux, we must change our terminology when it comes to studying emotions and their underlying neural circuits. Since studying neural circuits themselves does not give us access to conscious states of feeling, which is where emotions arise in his view, we should not refer to these circuits as circuits of emotion. Instead, we should talk about these circuits as related to survival mechanisms. In this sense, we have a change in the taxonomy with which we classify neural circuits, but the extension remains the same. Whatever circuits we found before and labelled as circuits underlying fear, for example, should now be called threat-response circuits and the like.

A second kind of change that Kron considers, and which is the target of this argument, is epistemic change. An epistemic change in a classification system is one that can change the extension of the phenomena of interest. In other words, the demarcation between what counts as an instance of interest, say of fear, changes relative to the previous taxonomy. An example of epistemic change in taxonomy, as I understand Kron’s distinction, would be the change from taxonomies based on basic emotions and affect programmes (Ekman [1992]; Panksepp [1998]; Izard [2007], [2009]; Ekman and Cordaro [2011]) to a constructionist taxonomy based on domain-general states of core affect (Russell [2003], [2009]; Barrett [2017], [2018]). Under the taxonomy proposed by basic emotion theorists, emotions form discrete categories with clear boundaries. The shift from this taxonomy to a constructionist taxonomy entails a change from discrete to dimensional taxonomy, where emotions are not categorically distinct from each other but rather lie in a continuous space.

Kron ([2019]) offers the following challenge to the functionalist: propose a classification system that (1) involves epistemic and not mere terminological change and that (2) does not use feelings as features of the classification. In his words:

Can the functionalist offer a criterion that meets Kron’s two requirements? To do this, the functionalist must offer an account of fear as a ‘response to a threat’, analysing responses to threats without recourse to feeling-based criteria. In his view, this challenge cannot be met.

To argue for this claim, Kron considers four versions of functional classification. The first is individuating these stimuli and manipulations in heuristic terms, that is, based on intuition. On this taxonomy, responses to threatening stimuli are classified based on the researcher’s own knowledge. He rightly rejects this classifier claiming that intuition is fallible and is bound to introduce implicit assumptions that rely on feeling-based classification. Using a heuristic classifier, a researcher might believe that a stimulus is threatening based on how they would feel or what they would expect to feel in the presence of such a stimulus, hence adopting a feeling-based classifier.

A second alternative is to individuate these responses based on their biological markers such as neural or physiological responses. Whatever the responses are, they are responses to a threat if we see, for example, amygdala activation and increased heart rate. Against this classification scheme, Kron argues that it assumes we already know that the biological markers we choose as demarcation criteria are markers of fear responses. This makes the classifier viciously circular, as to look for these biological markers we already need a classification of fear responses versus non-fear responses. In this regard, I also side with Kron’s argument.

The third classifier Kron considers is based on approach and avoidance behaviour. According to this classifier, a response to a threat obtains when there is avoidance behaviour rather than approach behaviour. In Kron’s view, links between emotion categories and behavioural reactions are too complex to allow a tractable demarcation criterion. For example, we might find reactions to a threat that rather than leading to avoidance, lead to approach, as in attack bouts. Hence, Kron thinks, this classifier also fails.

Lastly, Kron suggests the possibility of constructing a probabilistic classifier. This classifier works on the assumption that a stimulus counts as threatening if it is likely to cause damage, injury, or harm to the reacting organism. Against this classifier, Kron considers the example of sports cars. Sports cars are bound to cause damage to an organism based on (objective) probabilistic grounds. Yet, assuming that sports cars by themselves are fear-producing stimuli seems absurd. This leads to a problematic classification of stimuli, hence failing to produce the epistemic change Kron is after. Therefore, Kron concludes, there are no good candidates for functional classifications of stimuli and responses that would enable researchers to carry out their experimental work.

In my view, Kron dismisses options three and four on weak grounds. Regarding approach-avoidance classifiers, Kron’s argument fails to establish why the presence of approach reactions leads to a false classification. In his view, characterizing fear as an avoidance response to a threat fails because there are cases in which we approach the threat rather than avoid it. However, for Kron’s argument to work, these approach reactions must first be considered fear reactions. This involves the application of a conceptual criterion to identify reactions under a given emotion category. In the picture I sketched above, this criterion can be pre-theoretical or folk-psychological, but it can also stem from previous research. In either case, there must be a way to identify responses in order for researchers to design an experiment in the first place.

Let us first suppose that the conceptual criterion does not allow the classification of approach reactions to a threat as fear reactions. In this case, Kron’s argument fails from the outset, as it shows that the initial characterization does work to discriminate between fear and (at least some) non-fear reactions. What Kron needs thus is that there is an approach reaction that we can classify as fear (that is, avoidance of a threat), thus leading to an incorrect classification of the stimulus if based on the previous classifier. In this case though, the functionalist still has the possibility of sophisticating the functional characterization of fear. As in the cases I presented above, functionalists may then specify the types of reaction involved, hence including certain forms of approach behaviour under the scientific concept of fear. In this case, all that Kron has shown is that the functional profile of fear is more complex than the mere avoidance of a threat, but the argument fails as a rejection of functionalism as a whole. In other words, this would show that we got this particular functional description of responses to a threat wrong in the first place, but not that one such description cannot be offered in principle.

With regards to probabilistic classifiers, Kron’s argument is based on the premise that functional classification using probabilities may lead to different descriptions of the adequate stimulus with no way to decide from one of them. If a probabilistic classifier uses objective probabilities, this leads to the absurd consequence that a sports car may count as a fear-producing stimulus insofar as there is a probability that a sports car may harm an individual (either by approaching the subject at high speeds or, in the stranger cases, by exploding or the like). This means that probabilistic classifiers, if they are to be useful, must use subjective probabilities rather than objective ones, that is, they must consider whether a stimulus is perceived as a threat rather than whether it constitutes a threat or not.

In Kron’s view, subjective probabilities do not suffice though. In his words: ‘How will the functionalist researcher decide if stimuli are perceived as a threat without checking the feelings they elicit? When I discuss this with my colleagues, they usually suggest using biological or behavioral markers instead of feelings. But remember that vicious circle? In a “functional world,” in order to decide between two functional arguments by a marker (e.g., behavioral, biological), the functionalist needs to know which markers indicate threat, and in order to know which markers indicate threat, the functionalist needs to select the right stimulus’ (Kron [2019], p. 230). So, in Kron’s view, probabilistic classification cannot get off the ground without proper stimulus selection, but proper stimulus selection requires already knowing which markers show that there is a fear response (and hence that the stimulus really is a fear-inducing stimulus).

In my view, Kron is appealing to a chicken-and-egg problem in this regard. What comes first? Stimulus selection or hypothesis confirmation? This resembles a number of well-known problems in philosophy: to identify one thing, we must first know how to identify it, but to know how to identify it we must already know what it is. For Kron, this constitutes a vicious circle. Yet, I believe there are reasons to make this a virtuous circle instead. It is true that in order to identify certain markers, we must already have some identification criteria for the reaction we intend to obtain experimentally. However, empirical research does not begin in a vacuum, especially when it comes to psychological concepts such as emotions. As I explained above, functionalism may begin with folk concepts and, once research moves forward, refine their scientific concepts to propose better functional characterizations of emotional reactions. Of course, we may need some ground to stand on, but we must recall that the explananda of emotion research are, at least partially, the phenomena we refer to by using folk emotional concepts. Hence, it is not true that we do not have at least a pre-theoretical criterion that may help us identify markers for emotional reactions to perceived threats. We may (and often do) assume certain concepts of emotions, along with the criteria they offer, and from there we may formulate empirically tractable hypotheses.

This may sound to some as a fallback to analytic functionalism, the view that functional descriptions come from the analysis of folk concepts, a view that functionalists like Adolphs and Andler reject. I believe this interpretation can be resisted though. By claiming that folk emotional concepts may provide a starting point for empirical research, I am not suggesting that the correctness of a functional description lies in it being a proper analysis of a folk concept itself, but rather that we may start by invoking folk concepts and propose testable scientific functional characterizations in the future, the correctness of which is up to empirical research rather than conceptual facts. Hence, on the picture I propose, functionalists may still appeal to subjective probabilities by recourse to pre-theoretical criteria to select stimuli and, once research has advanced, examine previous intuitions.

Another possible source of resistance to this view is that it may fall into a feeling-based classifier. Specifically, it is worth considering what happens in case folk emotion concepts involve feelings. As I have developed this account, what we need to reply to Kron is that we can advance empirical research by recourse to pre-theoretical criteria based on these folk concepts. Yet, even if folk concepts involve feelings, it is not the case that they are exclusively constructed on feelings. Folk emotion concepts also include information about expressions, behaviour, and contextual cues that help us identify a given emotion. Hence, we can appeal to such non-feeling-based pre-theoretical criteria to individuate emotional reactions without falling into a feeling-based classifier. In any case, the extent to which folk concepts involve feelings is a matter of empirical investigation and at the very least cannot be decided a priori as falling back into feeling-based criteria. Furthermore, what we need to resist Kron’s scepticism is that we can advance a scientific research programme on the basis of functional and non-feeling-based descriptions, a possibility that I take to be open in this case.

As a result, I submit overall that Kron’s arguments do not show that functional classifications fail. What they show is that functional classifications (1) require paying attention to the functional characterization at play, and (2) constitute empirical hypotheses rather than conceptual claims. Whether a classifier is adequate requires empirical testing, as conceptually there are a number of ways of sidestepping Kron’s criticism.

4.2.  Functionalism is teleological

Another objection put forward by Barrett ([2016]) is that functionalism appeals to teleology. Against Adolphs ([2016]), Barrett claims that teleology is an act of mental inference, not of identification of a phenomenon. When we attribute a teleological function to an emotion, or any other psychological phenomenon for that matter, we are inferring what that emotion or phenomenon is adaptive for, rather than describing the actual phenomenon itself.

As I understand Barrett, she argues that a science of emotions must deal with what she calls ‘action identification’ rather than inference. To explain this distinction, she contrasts the claim that ‘eyes widen in fear to increase vigilance’ with ‘eyes widen to expand peripheral vision’ (Barrett [2016], p. 34). The first claim, a teleological claim, makes an inference about the purpose of the eyes widening, but it does not describe the action that leads to the satisfaction of that purpose. The second claim does describe the action, without any reference to purposes. It is merely the claim that eyes widen and, as a result, peripheral vision is enhanced. For Barrett, since functionalist models such as Adolphs’s make claims of the first sort, instead of the second sort, they fail as models that identify what is going on and instead constitute models of a speculative nature.

Barrett’s presentation of the argument is quite unclear. First, the assumption that science must only deal with action identification and not with what she calls mental inferences strikes as implausible. This is because ‘mental inferences’ are quite ambiguous here. Barrett claims that teleological language is metaphorical, a claim she attributes to Mayr ([2004]).⁴ What she means by ‘metaphorical’, however, is not specified. As I read Barrett, what she has in mind is that attributing a teleological function to a phenomenon is speculation. She holds that teleology involves a ‘metaphorical language that cannot be verified in physical terms [and therefore it involves] mental inferences or attributions […] of psychological functions’ (Barrett [2016], p. 34).

There are two ways in which one can interpret Barrett’s objection. On a naive reading, Barrett’s objection is weak insofar as all of our scientific hypotheses are inferential in some sense. When we formulate scientific hypotheses, we do not know yet whether the hypotheses are well confirmed. All we know is that given a phenomenon, it is plausible that some other phenomenon is going on. Criticizing teleology for being speculative in this sense is confusing at best. On a more charitable reading, what she is objecting to is rather that there are no criteria to determine which teleological claims are true. This is why she appeals to the idea that teleology cannot be verified in physical terms. On this interpretation, all there is is our speculation about what might be the purpose of a given phenomenon. This would mean that teleological vocabulary is unsuited for scientific theorizing and would make her case at least plausible.

Yet, Barrett is confusing two senses of functions that it is vital to keep separate. These are what Godfrey-Smith ([1993]) calls ‘Wright functions’ and ‘Cummins functions’ (named after Wright [1973] and Cummins [1975]). Wright functions are the sort of functions that appeal to teleology, and that are used in evolutionary biology and other related disciplines. They are initially analysed in the following terms:

This analysis, as Godfrey-Smith presents it, is meant to account for explanations in evolutionary biology. For example, evolutionary biology ascribes functions in virtue, not only of their effects but how those effects explain why the function has been passed on across generations. To say, for instance, that the function of claws is to aid in hunting is useful insofar as the effects of having claws for an organism that hunts explains how that organism has survived and the trait has been passed on. In this sense, the standard interpretation of Wright functions is as answers to ‘why?’ questions (Mayr [1961]).

On this reading of functions, functions are indeed teleological as Barrett thinks. In Barrett’s example, the claim that the function of eye-widening in fear is to enhance vigilance aspires to be explanatory in that the ascription of said function to eye-widening would presumably explain why organisms have evolved in such a way that their eyes widen in fear, namely, because it has enhanced vigilance in the past and led to better chances of surviving.

Here we can introduce a first counterargument against Barrett: despite Barrett’s conviction, it is not clear that this analysis of functions is not scientifically useful. As the example shows, biologists do make use of these patterns of explanation successfully. Moreover, they do not need to reduce them to physical vocabulary in order to render them empirically testable. What a biologist would need to show to support their hypothesis that the eye widens in fear to enhance vigilance is to show how increased vigilance aids in the survival and subsequent reproduction of ancestor species. The explanation only requires that we have good evidence that this effect of the widening of the eyes leads to increased survival. This is a matter of empirical fact, not mental inference or speculation, as Barrett would have it. Yet, let us examine the second sense of functions: Cummins functions.

Cummins functions are more akin to what Barrett calls ‘action identification’. They are functions ascribed in terms of how an effect contributes to a more complex capacity of an overall system. For instance, saying that the function of the heart is to pump blood is to ascribe it a role in the overall working of our body as a system. This need not require any story about how that function came to be or aids in the survival of an organism. All that it requires is that we identify how a phenomenon contributes to a system, that is, identify its actions. According to the standard interpretation, Cummins functions respond to questions of the form ‘how?’ (Mayr [1961]).⁵

If we can ground functions in this second sense, this is a final blow against Barrett’s objection. A functionalist approach to emotions can be developed focusing on how emotions, understood as dispositions to behave in certain ways due to relations between stimuli and other mental states, contribute to the overall behaviour of an organism. In other words, we can cash out emotions in terms of how, and not why, they have an influence on thought and action. If this strategy is plausible, it is simply not true that functionalism is inherently teleological since it can appeal to functions in terms of roles and capacities rather than purposes. At best, Barrett would have to offer a stronger argument against teleology in science in general and offer reasons why functionalism applied to emotions is inherently teleological. These two claims I find highly implausible.

In Barrett’s favour though, there is an important caveat that this reply entails. Adolphs and Andler’s particular account, as Scarantino ([2018]) shows, oscillates between these two senses of functions. This is a problem since they lead to very different empirical predictions and each account is suited for different types of explanation. As Scarantino ([2018], p. 203) puts it: ‘[Adolphs and Andler ([2018])] cannot have it both ways: either the relevant functional roles emerge from an observation of current capacities and duplicates have our emotions since they share such capacities, or the relevant functional roles emerge exclusively from a past history of selection, but then the current capacities we observe cannot shed light on functional roles’. What Scarantino correctly points out is that a functional model of emotions must decide between teleological (Wright) and dispositional (Cummins) accounts of functions. As I said before, this is because these accounts lead to different empirical predictions, thus shaping the empirical content of the theory proposed and because each account of functions has different explanatory demands and offerings.

Making such a decision lies outside the scope of this work since it requires an in-depth discussion of functions. Nevertheless, let me offer arguments in favour of applying a dispositional account of functions to emotions rather than a teleological account.

First, in my view, dispositional accounts better capture our folk-psychological concepts than teleological ones. Consider an everyday case of emotional attribution. When we say that someone is afraid, angry, happy, or sad, we mean that they have dispositions to entertain certain thoughts and to behave in certain ways. This makes no reference as to the adaptive purpose of the emotion but to its role in how the subject as a psychological system behaves. This way of talking about emotions functionally appeals to the emotion’s contribution to the system’s (that is, agent’s) behaviour.⁶

Second, the question about how to individuate emotions scientifically is not the same as the question of why emotions have evolved. Emotions might have evolved for a number of reasons, but, in my view, they are individuated by how they contribute to an agent’s overall behaviour. In other words, emotions help explain how an agent behaves, rather than why it has acquired those forms of behaviour. Hence, I submit that Cummins functions are a better account than Wright functions, at least when it comes to cashing out emotions as functional kinds.

Nevertheless, it is worth pointing out that this does not mean that teleology is irrelevant for a science of emotions. As I have claimed before, the question of what adaptive value emotions have is an important question in its own right, and it may call for a teleological answer. To make things clear, all I want to claim is that for the purposes of characterizing emotion kinds, we should apply a causal, Cummins-style account of functions rather than a teleological one. For the purpose of integrating emotions into an evolutionary framework, however, we may go with a teleological account.

4.3.  Functionalism is postulated

One last objection against functionalism is that it cannot be falsified in principle. The main reason for this, sceptics claim, is that we can describe anything functionally. Hence, no amount of empirical evidence can falsify a candidate functional description. This makes functionalist frameworks scientifically unattractive.

This objection was framed by Churchland ([1981]) against functionalism in general. In his presentation, he invites us to consider a functional description of alchemical kinds such as the ‘spirit mercury’. Substances that were ‘ensouled by mercury’ could be characterized functionally in terms of their disposition to reflect light or liquefy under heat. If positing functional kinds were a promising scientific strategy, there would be good reasons not to eliminate these kinds, Churchland argues.

A similar argument can be put forward using the case of phlogiston and other eliminable kinds. In Churchland’s ([1981], p. 81) words:

The functionalist stratagem, as Churchland calls it, cannot save these kinds from elimination. In Churchland’s case, it cannot save folk-psychology. Applied to our case, it cannot save emotions.

The application of this argument to emotions has also been put forward by Barrett ([2016]). She claims that under the functionalist framework, definitions are stipulated, not discovered. She argues that an emotion can be associated with a variety of functions. If this is so, we can ask: how can we find the correct functional profile that corresponds to an emotion? According to Barrett, this can only be done by stipulation. As a result, claims about emotions corresponding to functional profiles are not empirical claims, but conceptual ones that are immune to falsification.

This line of argument depends on the premise that functional characterizations are always stipulated and cannot be subject to empirical investigation. According to this objection, appealing to a functional profile is an ad hoc move that cannot lead to a progressive scientific research programme. Yet, this argument confuses the two senses of functionalism I introduced above, namely, analytic functionalism and psycho-functionalism. When psycho-functionalists propose a functional characterization of an emotion, they are not defining the concept associated with said emotion, and hence they are not claiming that the emotion is characterized in terms of the candidate functional profile as a matter of conceptual truth. Instead, psycho-functionalists propose the functional profile as an empirical hypothesis, one that can be revised in the face of conflicting evidence.

Consider again the case of characterizing fear in terms of avoidance responses to threatening stimuli. On a psycho-functionalist approach, whether this characterization works will depend on whether it can capture instances that can be called instances of fear in their own right. If we observe avoidance responses to threatening stimuli that do not fit the pattern, we must abandon this characterization and replace it with a different one. In most cases, scientists will engage in a sophistication of the proposed characterization, such as adding that they must be rapid, automatic avoidance responses, for example. Again, whether the new version is successful will depend on empirical facts, and not on conceptual analyses.

As a result, Barrett’s accusation that functionalism about the emotions is stipulative and therefore not falsifiable is unwarranted. If functionalists about emotions were proposing functionalism as a way of defining emotion concepts, the objection might have some traction. Yet, this is not what Adolphs and Andler claim. Additionally, not only are Adolphs and Andler explicit about their commitment to psycho-functionalism but they also propose a picture that is at least plausibly falsifiable.