emotions come prepackaged at birth. But whether affect programs are so very determined by inherited characteristics that they would occur in a recognizable form in radically different environments, or whether instead a significant degree of their inheritability arises because certain environmental features are pancultural and these help determine the program, is (though very interesting) not relevant to the discussions that follow. Similarly, whether the elements of the emotion syndromes are generated and coordinated by a central neural program, or whether they just occur together because of reliable environmental conditions (and thus, for example, could be controlled by several neural systems that could potentially operate individually, were certain unusual environmental conditions to occur), need not be answered here. I do believe that the affect programs arise from centralized neural programs, but otherwise the issue is one I leave to future empirical research. (For a discussion of these issues, see Griffiths 1997.)
Which Emotions Are Basic?
I will call all and only the emotions that are pancultural and that fall under the affect program theory the basic emotions. But there remains disagreement about what these emotions are. Ekman and others involved in facial studies have included fear, anger, joy, sadness, and disgust (Ekman and Friesen 1971). Panksepp doubts that disgust is a basic emotion; he believes that the basic emotions include at least seeking, fear, rage, and panic (a social distress system), lust, care, and play (Panksepp capitalizes all these terms to draw attention to the fact that these are technical terms related to, but still potentially distinct from, our usual uses of these terms; see Panksepp 1998). Some have also found preliminary evidence that there are pancultural expressions of contempt (Ekman 1988), and embarrassment and shame (Keltner 1995). But since fear and anger are in the intersection of all such lists (such as also Izard 1971, Plutchik 1980; see Kemper 1987 for a review of such attempted lists), for my purposes in this book I shall ensure that each argument regarding the import of basic emotions can be made with this subset alone. We can otherwise remain agnostic about the exact set of basic emotions. For the record, I opine that the union of both Ekman's and Panksepp's lists identify (not necessarily all) basic emotions: fear, anger, joy, sadness, disgust, seeking/curiosity, social distress, lust, care, and play.
Some Hypotheses Concerning Function and Eliciting Conditions
In arguing that some basic emotions are in part constituted by action programs, I have endorsed a view that these basic emotions have specifiable functions. That is, for example, if part of fear is the action program of flight, then flight is a function of fear; and if part of anger is the action program to
end p.28
 
attack, then attack is a function of anger. Although it will not be necessary for many of the arguments that follow in this book, it will at times be useful to refer to both potential roles and also eliciting conditions of the basic emotions. These are separate issues, strictly speaking; and yet, one should expect that functions that are type-specific to a basic emotion have eliciting conditions that are also type-specific. Thus, if a function of fear is to motivate flight from a dangerous object, then we expect that a dangerous object would be a typical eliciting condition.
There is growing evidence that there are some universal eliciting conditions for basic emotions, and for other affects (Boucher and Brandt 1981; Scherer and Walbott 1986; Scherer, Walbott, and Summerfield 1986). The general patterns revealed in these and other studies are quite familiar. Ekman and Friesen (1975) identify an actual or a threat of harm as an elicitor for fear; loss of an object for sadness; something repulsive for disgust; and frustration, a physical threat, insult, a violation of one's values, or someone's anger directed at oneself being causes of anger. Lazarus (1991) offers a taxonomy of "Core Relational Themes" for various emotions; these help define both function and eliciting conditions. They include a demeaning offense against me and mine for anger; facing an immediate, concrete, and overwhelming physical danger for fear; having experienced an irrevocable loss for sadness; taking in or being too close to an indigestible object or idea (metaphorically speaking) for disgust; making reasonable progress toward the realization of a goal for happiness (122).
These and other accounts suggest that, for some of the basic emotions, an abstract characterization of function and eliciting conditions is possible that will be consistent with many of the contemporary theories. Since I will work only with fear and anger as typical emotions, I suggest the following:
Fear functions to motivate flight from a threat, and is elicited by the perception of a threat.
Anger functions to motivate an attack against a defeasible enemy, and is elicited by the perception that a defeasible enemy has harmed or intends to harm the organism or something the organism values.
This list is obviously short; I could attempt an account of the functions and elicitors for many other affects (e.g., disgust functions to motivate the expelling of, or withdrawal from, potential toxins or pathogens, and is elicited by the perception that something is both potentially ingestible and is a toxin or pathogen). But the actual function and universal eliciting conditions of basic emotions and other affects is an empirical matter, and will require additional empirical investigation. This partial list will suffice to allow me to make a few points regarding function and eliciting conditions in later chapters, and so I will end with the hypothesis that these two accounts are correct.16
end p.29
 
end p.30
 
2 The Case Against Cognitivism
Abstract: Cognitive theories of emotions are criticized. Cognitivism is shown to have two forms: reductive and doxastic. Each is found inconsistent with a range of important scientific findings about affects. The affect program theory is consistent with these findings, and is consistent with a weak form of cognitivism. The failure of cognitivism about emotions is also evidence for the hierarchical view of mind.
Keywords: affect program theory, cognitivism, doxastic cognitivism, emotions, mind, reductive cognitivism, weak cognitivism
Craig Delancey
In chapter 1, I observed that one thing that might distinguish affects is their cognitive contents (or perhaps their relations to these contents). This would be an approach that is consistent with the various cognitivist theories of emotion. My purpose in this chapter is to address cognitivism and show that it is an untenable view of the basic emotions if it is meant to define them or otherwise explain their necessary nature. In recent years, many criticisms of cognitivism about emotions have been made (see Deigh 1994; de Sousa 1987; Gordon 1987; Griffiths 1989, 1997; Stocker 1987; Stocker and Hegeman 1996). These various criticisms have not, however, touched upon the important scientific evidence, especially from neural science, that is inconsistent with cognitivism about emotions. In this chapter, I use a sampling of this evidence to explore why cognitivism is inadequate. This will also, as in chapter 1, provide an opportunity to support my overarching themes: the affect program theory, a hierarchical and bottom-up view of mind, and an enriched naturalism.
Cognitivism about emotions presumably arises from the observation that affects can be about something: they can be representational states, even propositional attitudes. Some scholars have attempted to reduce affects to propositional attitudes like belief or judgment, or at least to claim that affects require these kind of states. In philosophy, the most common attempts at reduction of affects have generally been made for emotions, although some have also attempted to so reduce desire. Here I will criticize only theories that reduce the basic emotions to, or posit that they require, beliefs or other propositional attitudes (for criticism of attempts to reduce desire to belief, see, for example, Lewis 1988, 1996).
A Note About "Cognitivism"
Theories that claim emotions require or are made of beliefs have been, at least in philosophy, called "cognitive" theories of emotions. This is an unfortunate term. In contemporary cognitive science, for example, researchers
end p.31
 
freely posit mixtures of unconscious and even simple processes together with complex conscious processes into explanations of the kind of skills that would normally be called "cognitive." There is, in other words, no clear demarcation between cognitive processes and complex but non-cognitive processes; rather, the only things that would clearly be noncognitive processes would be things like very simple reflexes or activities that are not neural, such as digestion. However, in theories about emotions, the notion of "cognitive" tends to be much stronger; and subcognitivism about emotions might correspondingly involve processes much more complex than simple reflexes.
Thus, what makes some theories of emotions "cognitive" is sometimes not clear for philosophers or scientists. However, although scientists have had their own debates regarding cognitive theories of emotion (one classic debate was held between Lazarus and Zajonc; see Scherer and Ekman 1984, 221-270), a lack of clarity is sometimes not as pressing a practical problem for the scientific study of emotions since such studies often need not be explicit about what is necessary and sufficient for a process to be cognitive. This is because if a theory posits a process that is widely granted by other scientists to be cognitive, and the existence of the process can be demonstrated in experiments, then more conceptual clarification may be unnecessary. For example, if someone believes that emotions require appraisals which are by definition cognitive, and appraisals are granted to be demonstrated by the answers of subjects to certain questions, then the otherwise somewhat mysterious notion of an appraisal may not need further analysis for a hypothesis to be defended or a limited theory to be posed. Since my goals here require conceptual clarity, I will focus upon several theories that are cognitive insofar as they contain claims that one can reduce emotions to, or that emotions require, beliefs or closely related kinds of propositional attitudes. This too may suffer from serious ambiguities; for example, it could be that one kind of brain state is sometimes acting as a constituent of a propositional attitude or otherwise being used as one, and at other times it is not, so that the very idea of a state being a propositional attitude would be deceptive. However, since my goal is to criticize cognitive theories and not to endorse them or any theory of propositional attitudes, I can avoid clearing up these ambiguities any more than is necessary to provide counterexamples. That is, often a cognitive theory of emotions is stated in a way (e.g., that emotions require particular kinds of judgments or beliefs) that can be refuted without providing more clarification about the necessary and sufficient conditions for a process to be cognitive. In this regard, we can best address cognitivism by examining philosophical theories of emotions that offer clear statements of some features of such positions.
Paul Griffiths, in his criticism of some cognitivist theories of emotions (1997), has used the term "propositional attitude theory" to describe those philosophical theories that hold that these affects require, or are, propositional attitudes. This is a useful clarification, and it also touches upon related issues that I aim to criticize (such as particular views about the role in
end p.32
 
What Is a Propositional Attitude?
Propositional attitude is the philosopher's term for any mental state that is directed at, or has, a proposition. Any mental state that is best described as relating to a proposition is prima facie a propositional attitude: "Karen believes that it is raining," "Eric is angry that his landlord did not fix the water heater," or "Adam fears that he may not pass the exam." In these cases it is raining, his landlord did not fix the water heater, and he may not pass the exam are propositions "had by" the attitudes of belief, anger, and fear, respectively.
From a naturalist perspective, the notion of a propositional attitude presents at least two challenges. First, propositional attitudes have as yet no clear place in scientific theories of mind, and so we should not complacently assume that they somehow represent the fundamental form of thoughts. Second, there is a possible ambiguity between a mental state warranting the title "propositional attitude," and the ability to formulate explicitly and reflect upon the kind of contents that are objects in a propositional attitude. In the former case, we might say that a cat tracking a mouse "believes that the mouse is behind the bookshelf." This might be a correct attribution of a propositional attitude if by such an attribution one merely wants to predict in a robust way the cat's behavior. But philosophers who grant special status to propositional attitudes and to their role in behavior tend to be concerned with the role of such attitudes in deliberative reasoning, such as in inference. In such a case, it seems likely that nonhuman animals do not have them. Often, what appear to be at issue (and what shall be my primary concern in this book) are the abilities that humans have to actually, in some sense, reflect upon cognitive contents; the ability to not only believe that the mouse is behind the bookcase, but also such abilities as to understand that "the mouse is behind the bookcase" is a proposition, to be able to express it as one in a language, to be able to draw logical inferences based upon it, or at the very least to be able to recognize some of the conditions that must obtain in order for the proposition to be true. All of these seem to be capabilities that accompany adult language use, and so the ability to use language provides a sufficient condition for the ability to have propositional attitudes.
mind of certain forms of rationality). In most of this book I will take cognitivism about emotions to be the view that the relevant affects are, are in part constituted by, or require, propositional attitudes. However, I will continue to use the term "cognitive." The primary reason is that the term is already established as a label for these propositional attitude philosophical theories. But another reason is that there are some approaches that attempt to explain affects by reference to the kind of states that we might call highlevel cognitive states, but which are not based on propositional attitudes. In chapter 11, I will criticize the idea that emotions can be explained by symbolic models of the kind that have typified classical AI, and one might well call these kinds of theories cognitive. Thus, I eventually aim to expand
end p.33
 
the notion of cognitive to include both propositional attitude theories and symbolic computational functionalism (and I do not claim to show that any other notion of cognitive is inappropriate for the basic emotions or any other affects).
I suspect that one might, eventually, find an even broader characterization of the cognitive that is demonstrably not a necessary condition for basic emotions. Thus, I am criticizing philosophical theories in this chapter, and in chapters 3 and 6, as my target cognitivist theories, but the results here will generalize to many of the cognitive theories of emotion held by many scientists (particularly psychologists), even though they are using the term "cognitive" in a way usually divorced from any conception of propositional attitudes. For example, in their book on the cognitive origins of emotions, Andrew Ortony, Gerald Clore, and Allan Collins have argued that emotions are "valenced reactions to events, agents, or objects, with their particular nature being determined by the way in which the eliciting situation is construed" (1988, 13). If we were to understand this to be either a definition of emotion, or otherwise as a statement of the necessary conditions of any emotion, then this might be a theory which implies that emotions require mental states, like beliefs, that are propositional attitudes or are at least of similar complexity. This is because the notion of how a situation is construed could require all kinds of abilities to recognize and categorize situations, to recall other situations, to draw inferences about them based upon our beliefs, and so on. Thus, if a certain form of the propositional attitude theory fails, then this reading of Ortony and his colleagues might also fail. Similar conclusions can be drawn for a host of other cognitive theories of emotion; I will not be undertaking a literature review of these theories, but I will be arguing for a view of emotions and of minds that is antithetical to some of the presuppositions common to some of these theories, and so it is important to recognize that the arguments against certain forms of cognitivism are meant to outline a general objection to some of these presuppositions. In this regard, it is sometimes useful (especially in building a case against the cognitive autonomy fallacy) to keep an admittedly vague contrast between cognitive processes that are typically involved in our conceptual abilities, especially language, and that perhaps arise primarily from neocortical neural circuits, on the one hand; and potentially subcognitive processes that are typically involved in perceptuomotor control and integration, and in affect, and which perhaps arise primarily from subcortical neural circuits, on the other hand. Nonetheless, let me reiterate that the general use of cognitive in the sciences of mind is definitively not the defining feature of the philosophical cognitive theories of emotion that are my present targets; a cognitive theory of emotion is here understood to be a propositional attitude theory (or, in chapter 11, also a symbolic computational functionalist theory).
I will thus use the following terminology. A representation is (in the kind of organisms that are my concern here) a brain state that stands for another (not necessarily real) state or object. Representations need not be discrete
end p.34
 
(i.e., they can be magnitudes), but they must play a role in a representational system (that is, although I do not endorse holism and so do not require that each representation requires others, each representation must be part of a system that "consumes" that representation appropriately). I grant that affects use or are constituted in part by representations, and so have no objection to those theories (which, in some contexts, might be called cognitive theories) that take affects to use representations. Symbols are discrete representations that function in a representational system in a way that can be properly modeled by a combinatorial syntactic system. An example of a symbolic system is a natural language. Propositional attitudes are representations of events or states of affairs, and they have the special property that they are normally true or false (thus, these mental states can play a role in logical inference that a proposition can play—this is important when I discuss matters of rationality), and they are articulate representations formed of symbols or other representations. Thus on this terminology a cognitive state, or a cognitive system, is one that requires propositional attitudes—although, as stated, I shall later weaken this to include complexes of symbols, and argue that we can also reject this weaker form of cognitivism.
In recognition that the term cognitive is difficult to pin down, and also that the line between cognitive and other representational processes is likely not clear (there is probably no clear line between symbols and non-symbolic representations, for example), I will use the term subcognitive instead of noncognitive for those processes that are not propositional attitudes or complex symbolic representations. Subcognitive processes may be conscious but do not need to be so; and they can be representationally rich but are not propositional attitudes or otherwise propositions, and they do not all require language. Evidence that a kind of process is potentially subcognitive will include any of the following: it is shared by nonhuman animals of presumably simpler mental abilities (e.g., rats); it develops in humans before language and other complex cognitive abilities; it does not have to be learned; it happens or is elicited very quickly (e.g., in a few tens of milliseconds); it is enabled by neural wiring that is subcortical or otherwise can operate independently of the kind of neural structures that enable abilities like language. Also, a process may be suspected to be subcognitive if it is the case that assuming the process is a propositional attitude explains nothing more than would assuming it to be a more basic representation; if assuming the process is a complex of symbols explains nothing more than would assuming it to be a more basic representation; or if the agent is unable to report accurately or at all on the process or its object or cause.
Two Kinds of Cognitivism: Reductive and Doxastic
Philosophical theories that associate emotions with cognitive states like beliefs are usually of two kinds. Some identify emotions with other propositional
end p.35
 
attitudes; I will call these reductive cognitive theories. Others may not identify an emotion with these other mental states but claim that emotions require beliefs of particular kinds; I will call these doxastic cognitive theories. The claim that the beliefs need be of certain kinds is necessary. Most of us, for example, might think that human minds must have some beliefs, and since human emotions are mental states they would require beliefs in this sense. But the doxastic cognitivist means something stronger than this; what is of importance in the doxastic theories is that emotions require beliefs which are instances of particular kinds specific to the emotion. This requirement will be made clear for each doxastic theory as needed. The reductive cognitive theories are usually trivially doxastic cognitive theories, but doxastic cognitivism need not be reductive. In the rest of this chapter, I shall show that these theories, when construed as universal claims about all instances of basic emotions, are false.
Reductive cognitive theories generally have similar presuppositions, and here we can get a sufficient characterization of them by reviewing just a few of these presuppositions. Most of the reductive cognitivist theories in philosophy are of two general kinds:
(1)  
Judgment theories. Robert Solomon claims that "an emotion is a judgment (or a set of judgments)" (1977, 185). Not all such judgments result in emotions, but rather, "Emotions are self-involved and relatively intense evaluative judgments. . . . The judgments and objects that constitute our emotions are those which are especially important to us, meaningful to us, concerning matters in which we have invested ourselves" (187). Martha Nussbaum reconstructs, and endorses a version of, the view of the stoic Chryssipus that emotions are judgments of value concerning something that is essentially related to the eudaimonia—the well-being—of the subject. Nussbaum writes, an "emotion is itself identical with the full acceptance of, or recognition of, a belief" (1990, 292). This phrasing makes it seem that Nussbaum has a second-order theory (where emotion is a belief about belief), but as I understand her, the actual formation of the relevant belief is the emotion (albeit the belief may be one we resist, and so second-order epistemic matters are involved).
(2)  
Reduction to belief and desire. Joel Marks proposes that emotions "are belief/desire sets . . . characterized by strong desire" (1982, 227), and thus "emotion reduces to belief plus strong desire" (240). Ronald Alan Nash gives a slightly more sophisticated version of this in a theory he calls the "new pure cognitive theory" (1989). He holds that beliefs and desires give rise to a dispositional state that results in a desire upon which the subject has an unusual degree of focused attention and (potentially obsessive) overvaluation. The emotion is this state of having and focusing upon an overvalued desire (or perhaps desires).
end p.36
 
We can, without loss of accuracy, group these views if we recognize that the cognitive element is the formation of the right kind of beliefs given certain desires. There are many situations in which we form or assent to some belief (that is, in which we make a judgment) but do not feel an emotion (such as, forming beliefs unconcerned with our self-esteem or eudaimonia). Clearly the particular kinds of beliefs, not the act of judging, is the operative notion.
Doxastic cognitive theories are more homogenous, and can be treated as of one kind:
(3)  
Doxastic cognitivism. Radford (1975), Walton (1978), Shaffer (1983), and many others hold that emotions are caused by certain beliefs and desires. Shaffer gives an example: "I am driving around a curve and see a log across the road. . . . I turn pale, my heart beats faster, I feel my stomach tighten. . . . I slam on the brakes and stop before I hit the log. I acknowledge that when I saw the log I felt afraid" (161). In his analysis, an emotion is "a complex of physiological processes and sensations caused by certain beliefs and desires. Thus, seeing the log, I believed that bodily harm was likely and I desired not to be harmed." (161)
This view is quite similar to that of Marks and Nash, but the two kinds are distinct in that doxastic cognitivists take other bodily responses to be essential to the emotion—even to be the emotion—whereas Marks, Nash, and the other reductive belief/desire views take the physiological response to be an inessential consequence.
Empirical Evidence Against Reductive and Doxastic Cognitivism
In chapters 3 and 6 I introduce arguments against both reductive and doxastic cognitivism that appeal to philosophical notions of rational action and to common platitudes about emotional behavior. These approaches are important, since philosophical differences about the import of scientific results can mean that the vast empirical evidence available to us is moot. Here, however, I will go straight to the scientific evidence, which, for at least the basic emotions, effectively demolishes both views. I will review 6 objections here.
1. The confusion of cognition with affect. One problem with reductive cognitivism is that it does not capture what is specific about affects and separates them from other cognitive states like beliefs or merely entertained ideas. For example, basic emotions are characterized by autonomic body changes (one can of course deny this, and may have to, in order to defend a reductive cognitive theory of emotion). But judgments or beliefs are not so
end p.37
 
characterized. For Marks, these body changes are just features of desire: for him it turns out that an emotion is not just a combination of beliefs and desires, but of beliefs and "strong" desires. This is perhaps nothing more than a terminological difference from the kind of taxonomy I introduced here: whereas I doubt that there is anything like desire, and so I separate basic emotions from desires, Marks would group desires and basic emotions together, calling the latter "strong desires," and then construct cognitive emotions out of beliefs and the strong desires. Much the same could be said about Solomon's notion that the judgments that constitute an emotion are "intense." But if all that was intended were a terminological change, such an approach could at best be called misleading, since: (1) there is surely something distinct about anger and desire, or anger and judgment, as these are usually understood; (2) emotions do not operate like desires are supposed to do (see chapter 3); and (3) this would make it impossible to distinguish the different emotions, since they would all be instances of a generic notion of desire; so that (4) this would make the position merely a form of doxastic cognitivism, since it would presumably be the beliefs which distinguish the emotions. Nash, on the other hand, is explicit: emotions normally have but do not need their physiological correlates. "What I deny is that bodily changes constitute being emotionally upset or perturbed, or are even necessary to such a state" (1989, 497). Nash's theory sounds like a cognitivist theory that reduces emotions to more than just beliefs and desires, since it introduces these elements of focus and overevaluation. However, since presumably one can be focused upon other things and value other things without having an emotion, these are not sufficient to have an emotion; and as for their being necessary, there appears to be no way to distinguish desires that one is focused upon and values highly from other desires, except of course to call the former "emotions" (a parallel to the problem discussed below regarding Marks's "strong desires" and normal desires).
Even if we avoid talk about beliefs, and instead reduce emotions to judgments, the result is similarly problematic. What separates judgments that are emotions from other kinds of judgments? The answer is the content of the judgment: the belief that is formed. In Solomon, emotions arise when we are making judgments about ourselves, the content of which matters to our self-esteem. For Nussbaum's Stoic, they arise when we are making judgments about things we value. This characterization could be circular, given that much moral theory attempts to explain value—or at least valuing—in terms of emotion; if we were to accept such claims and then argue that emotions are judgments about what we value, the theory would be quite vacuous. But these theories fail on more explicit grounds. Since emotions are identified with judgments, the relevant judgments should always be accompanied by the proper emotions, but they are not. This has been shown by, to pick just one example, Antonio Damasio's studies of prefrontal cortex damaged patients who show no measurable loss of cognitive
end p.38
 
skills but who have, as a result of their brain damage, emotional defects. One such subject, EVR, was studied extensively (Damasio, Tranel, and Damasio 1990). This subject has an IQ of 135, and passes all the usual neuropsychological tests like a normal. But he came to the attention of Damasio and his colleagues because he showed deficits in rational decision making. In one experiment, EVR was shown pictures of disturbing and provocative scenes. These pictures cause in normals a skin conductance response—a clear measure of the autonomic signs of affect. But EVR showed no significant response—he literally flatlined on his polygraph when he merely looked at the pictures and was not asked to describe them. This subject even reported after the test that he had noticed that he did not have the kind of feeling that he thought he ought to have for some of the pictures. He has the cognitive ability to recognize and describe the phenomena, but he does not have the appropriate emotional responses to them. Damasio's explanation of EVR's lack of reaction, and of his impaired rationality, is his own somatic marker hypothesis: Damasio argues that the bodily reaction that a normal subject has for the affect-evoking stimuli acts as a marker of that stimuli, and we sometimes depend upon this marker in making rational decisions. But regardless of whether the somatic marker hypothesis is true, EVR is a clear counterexample to reductive cognitivism, and perhaps even to doxastic cognitivism. He has intelligent, seemingly rational judgment-making abilities, makes the correct kinds of judgments, and not only has little or no affects in some of these cases, but in his everyday life performs so many irrational tasks that he is essentially disabled.
2. The inexplicability of direct neural stimulation and of abnormal cases. Other kinds of evidence of basic emotions without the kind of content as constituent or cause that cognitivism requires include the generation of basic emotions through direct stimulation of the brain by electrodes, or by what is believed to be direct stimulation from defects like epilepsy. Direct electrical stimulation of particular subcortical areas of the brain can yield affective states in humans and nonhuman animals (see King 1961, Gloor 1990; Fish et al. 1993; for review, see Frijda 1986, 381-386). Recall also that (as we saw in chapter 1), the neuroscientist Jaak Panksepp has argued that the basic emotions are identifiable by the criterion that they can be generated by direct electrical stimulation of the brain (1998, 52). Also, brain damage can result in spontaneous and excessive affect. Specific emotional reactions often accompany the onset of seizures for epileptics (Ervin and Martin 1986). It has long been known that lesions in parts of the hypothalamus can cause rage in human and nonhuman animals. The classic studies of decorticate cats also first led to such observations (Cannon and Britton 1924; Bazzett and Penfield 1922; see also Bard 1928). To sustain a reductive or doxastic cognitive theory given such observations one must either deny that these are real emotions, contrary to all the behavioral evidence that is available; or somehow claim that these lesions and direct stimulation first,
end p.39
 
or at least simultaneously, generate the required beliefs of the organism. This is possible but implausible; at least, the burden of proof is surely with these cognitivists.
A related and noteworthy fact is that some emotions seem to be more easily triggered by features which are not in any relevant way beliefs. For example, R. B. Zajonc has argued that failure to cool the brain properly (which can happen, for example, if your sinuses are very severely clogged) can cause anger (Zajonc, Murphy, and Inglehart 1989). And we recognize that things like being too hot, loud noises, an uncomfortable chair, and other environmental factors can predispose us to certain emotions.
3. The problem of homology. If we accept evolutionary theory, we should expect there to be homologs of many capabilities between organisms, where more nearly "related" organisms share more common features. Thus, we should expect affects to most likely exist in other species of animals, and to be more similar to our own affects as those animals are more closely related to ourselves. And we do in general talk this way, and most scientific understanding of emotions has these states as being present in many species of nonhuman animals. We do not usually attribute fear to worms, but we do usefully attribute fear and a host of other emotions to cats and dogs; and many scientists readily study fear by using cats or rats or other organisms as models. Are we mistaken to do this? It would seem on a doxastic or reductive cognitive theory of emotions that we are, since presumably a cat or rat does not have the kind of cognitive capabilities necessary for an emotion on such a view. As already observed, we can weaken the sense of emotions being cognitive, so that a cat's fear is said to be merely representational. The cat is afraid of an approaching dog because undoubtedly it recognizes and categorizes the dog as a threat. But such a weakening of the requirements of what will make an emotion cognitive will fail to satisfy some of the goals of having a doxastic or reductive cognitive theory of emotion. One of the principal motivations for a doxastic or reductive cognitive theory of emotions has been to make emotions a part of rational action by having each relevant emotion be a state with content that itself can be part of a rational "belief-desire system"; a foremost feature of this is the formation of propositions and some minimal proper logical procedures upon them (drawing inferences, expunging contradictions). Presumably mere representations, which are not part of reflectively propositional contents, do not qualify: mere representations are not true or false, for example, so cannot be consistent or inconsistent; they cannot alone play the same kind of role in an inferential system that propositions can; we cannot revise them in the same way; and so on. Similar problems arise for emotional evaluation. At the very least, doxastic or reductive cognitivism is going to have to be supplemented with a powerful theory of representation if it is going to explain how both rats and humans can have emotions that are to be reductively or doxastically construed.
end p.40
 
Even setting aside these concerns, it seems clear that some nonhuman animals emote and do not have the same kinds of content that we do when we have what is purportedly the same kind of emotion. Since the state of the "fearing" cat can share many of the physiological and behavioral features that our own emotions do, we are again confronted with the question of why we would take the cognitive aspects of the emotion as more important than these other features. Taking evolution seriously suggests that the other features should be primary, such as the kind of behavioral responses (in this case, flight) shared by these animals. Finally, our growing understanding of some of the neural circuitry enabling some emotions and other affects often includes the identification of crucial roles for subcortical structures that are widely shared across mammals, and some of which may even have homologs in more distantly related species.
4. The problem of early development of the emotions. Human beings show a development of some emotional capabilities from infancy (see Scherer and Ekman 1984, 73ff) to mature adulthood, and some affective capabilities develop prior to our cognitive abilities. An infant can show some of the facial expressions of emotions, and after only a few weeks exhibits many of the behavioral features of some emotions—signs of anger at being frustrated, or fear when confronted with strange stimuli, or pleasure when they see a mother's face. Surely such infants do not have the developed cognitive skills, however, to allow them to have the attitudes like belief and desire that a doxastic or reductive cognitive theory require (and consider also Eibl-Eibesfeldt's research discussed in chapter 1). Our best understanding of development suggests that affects like the basic emotions are capabilities that are inherited, and which can be changed by learning, including eventually being directed or caused by propositional attitudes. This is a view contradictory to reductive or doxastic cognitivism, in which the abilities to entertain propositional attitudes of the relevant kind would have to precede the ability to have the relevant emotions.
5. The problem of neuroanatomical differentiation. There are structural distinctions in the neuroanatomy underlying basic emotions and some other affects that are not consistent with cognitivism. This is a point well illustrated, for example, by recent research by Joseph LeDoux, who has worked to map out the neural pathways of fear and show that there is functional and anatomical separation between affective and cognitive processing systems (for an overview see 1996). LeDoux has shown that there are neural pathways involved in fear conditioning which link to both cortical and subcortical areas. In particular, using fear potentiation studies of rats, he found that the aural cortex could be ablated and the fear-conditioned response could still be shown, working through the subcortical pathways. What was lost when the aural cortex was ablated was tone discrimination: a rat would show fear response to any tone, where before it could discriminate
end p.41
 
the tone to which it had been conditioned. In human beings there are also a host of complex pathways that operate for basic emotions, including connections between the amygdala and other subcortical structures believed to be essential to basic emotions, and also connections to various cortical areas, including polymodal and supramodal areas. The proper picture of the relation between affects and content therefore seems to be that affects can have varying degrees of cortical contribution. If any one of these cortical areas that was connected to the amygdala and other relevant subcortical structures was lost, we can expect that an affective ability could in some specific way be impaired, but that it would still remain.
The subcortical pathway that LeDoux identified for fear (and presumably such pathways could be present for other basic emotions) is similar to the kind of system that is suggested by Zajonc's research on the mere exposure effect. The affective results of these pathways are not best called cognitive, or at the very least they are surely not best identified as operating by way of generating propositional attitudes: they are faster than high-level cognition, less discriminating, and not open to report. LeDoux's subcortical pathway is also consistent with the findings of Arne Öhman and Joaquin Soares (1993), discussed in the last chapter, which provide some evidence for the theory of M. E. P. Seligman (1971) that some subjects are biologically predisposed for fear conditioning for some stimuli, such as snakes. Also, since Öhman and Soares's findings were shown to be independent of lateralization, and since many cognitive functions, and especially language, are highly lateralized, this suggests that the relevant fear conditioning or recognition in question is subcortical.
6. Displacement. Finally, there is a phenomenon that has in part been studied by scientists in terms of generalization and second-order conditioning (and which may also have an analogue in theories of emotional congruence in perception and attention; see Niedenthal and Kitayama 1994), and that is part of our folk preconception of emotions. It is common folk psychology that an emotion can, as it were, go searching for an object. Eric can start out angry at his landlord, and end up angry at his boss for reasons that at some other time would not make him angry at his boss. Our common understanding of such events is that Eric is in a state of anger, caused by beliefs about his landlord, and this state can take different objects. But if reductive cognitivism were true, then such displacement should be impossible; instead, in having two different sets of beliefs or judgments, we would have two unrelated emotion events. And similarly for doxastic cognitivism: if an emotion requires a belief, then either we have two emotions here (because two different beliefs) or we have one emotion with two different beliefs. If the latter were the case, we could rightly ask what in the emotion is shared between these two doxastic states, and this unchanging element would seem to be more essential to the emotion than the fungible beliefs that are said to be required. The former case is ruled out by the conditions of the thought experiment: we supposed that the displacement results in an emotion
end p.42
 
in cases that otherwise would not give rise to the emotion. If Eric's belief that his landlord is charging him too much money is necessary for Eric to be angry at his landlord, how can it then be that Eric ends up angry at his boss for reasons that normally would not cause him to be angry? The anger in the latter case would seem to be better explained not by the beliefs involved, since these can sometimes fail to cause an emotion, but by some other factors. Thus, if emotional displacement of this kind occurs, it poses a counterexample to both doxastic and reductive cognitivism.
Weak Cognitivism
In arguing against doxastic or reductive cognitivism, I do not deny that, in humans, a basic emotion might often have some kind of belief or propositional content accompanying it; all evidence indicates that emotions in humans often are guided by propositional contents in a way that merits being called "cognitive." It is also possible that doxastic cognitivism could be true of some of the things we call "emotions"; that is, some of the things that we call "emotions" may be distinguished, by reference to related beliefs, from affects which on any other scientific measures of the individual are relevantly of the same kind; such a thing might even be because social standards play a role in the concept of what that emotion is (I return to this theme at the end of chapter 4). Also, as already stated, given a weak sense of "cognitive"—so that, for example, a mental process is cognitive if it is representational—then all emotions might come to be necessarily "cognitive." Finally, of course, one is free to chose any taxonomy she desires; so we could strengthen our definition of basic emotions to make something like doxastic cognitivism true.
As we have seen, however, one reason for choosing against doxastic and reductive cognitivism is that they fail to distinguish basic emotions from other kinds of cognitive states. Our goal should be an understanding of basic emotions that is as broad and as rich as possible, and doing this requires that we look not for what is normal for, but rather for what is necessary for (or at least, most common to), the relevant emotions.
These cognitive theories are perhaps most compelling when they are used to account for those features of emotion that ally them with what would normally be called cognitive features. These include the intentionality of emotions (the fact that they are often in some sense "about" something), their evaluative nature (they are often like judgments, which can be seen as evaluations made by the subject), and their interesting connections to rationality (some see emotions as necessary to rationality, others see them as antithetical to rationality, but most see them as having a complex and significant relation to rationality). These are all features for which any theory of emotion should account, and a doxastic or reductive cognitive theory can make a quick and plausible job of this by making emotions into judgments or having them require beliefs. Beliefs are by definition intentional,
end p.43
 
and they can be evaluations; on most accounts of rationality these are going to be the elements of rational thought. In chapters 5 through 8, I shall show that there are other equally plausible explanations for these features of some emotions—explanations that have more explanatory power.
Finally, these observations are not meant to be arguments that there is no place for a cognitive theory of emotions. In fact (as I will argue again in chapter 12), if our goal is to understand the cognitive structure of emotions, then one approach should be to study emotions directly in terms of their cognitive causes and their cognitive structure. That is, the denial of strong forms of cognitivism like doxastic and reductive cognitivism does not entail that any study of emotions in terms solely of beliefs and similar kinds of cognitive states is erroneous. Given how incomplete our present understanding of the brain and mind is, one might make little or no progress near term in understanding the cognitive structure of emotions by any other method. Again: I do not reject the goal of understanding the cognitive structure of emotions in terms of their cognitive contents, nor even the claim that emotions are often cognitive in some robust sense; rather, it is the separate claim that the basic emotions necessarily are cognitive in a strong sense such as, for example, we find in reductive or doxastic cognitivism, that is false.
An alternative to doxastic and reductive cognitivism is a view I will call weak cognitivism: the hypothesis that the occurrent instances of relevant emotions are for humans often, but not necessarily, highly integrated with cognitive states (including propositional attitudes). This integration can include beliefs and other cognitive states causing, determining the expression, the eliciting conditions, or the intensity of, the relevant basic emotion. I endorse a form of weak cognitivism (but, as I will show in chapter 6, we need to weaken this even further by explicitly disavowing that beliefs are even normally necessary for cognitive instances of emotions). Weak cognitivism is consistent with the affect program theory.
Summary: The Hierarchical Model of Mind
If I am going to review what lessons some of the emotions can hold for the problems of intentionality, rationality, and consciousness, and for AI, it will be sufficient to stop the taxonomic investigations here with the notion of the basic emotions. This is hardly the last word on emotions—it leaves most of those things we call emotions uncategorized, and it raises as many questions as answers—but it is enough to start some explorations that will reveal much about the importance of the basic emotions and the views of mind with which this understanding is consistent.
I have been concerned to describe occurrent affects, and have proposed the thesis that they are motivational states. Affects can be characterized by such properties as their duration, physiological correlates, conscious experience,
end p.44
 
behavioral correlates, and content. All of these elements play an important role in our understanding of emotions, but of these only physiological and behavioral correlates appear to be potentially sufficient to identify and distinguish an emotion. Given that there is a class of affective states that appear to be pancultural, based in inherited biological capabilities and characterized by recognizable behaviors, I concluded that these are the basic emotions. These basic emotions include at least fear and anger, and probably many other affects. These are the emotions described by the affect program theory.
The affect program theory is consistent with or explains all of the objections raised in this chapter against doxastic and reductive cognitivist theories of emotion. Cognitivism is understood as the view that emotions are constituted by or otherwise require beliefs or other propositional attitudes (and subcognitive states are therefore any representational states that are not propositional attitudes); this terminology is standard to much philosophy of emotion, but not to the sciences, so we must be careful to remember that cognitive here is used in this strong sense. The basic emotions are clearly distinct from beliefs and other cognitive contents in a fundamental way. Thus no problem arises from those abnormal cases of spontaneous emotions or the direct stimulation of the brain; we should expect it to be possible to stimulate the neural substrates of the affect programs directly, without having to stimulate the cognitive centers that would often be responsible for their elicitation. Nonhuman animals show these behaviors because the affect programs evolved and so likely have homologs in other related species. The development of the affect programs is also no problem. Blind children, even blind children who are retarded, need not learn, but already have, these programs. The existence of subcortical emotional pathways and the extrapyramidal enervation of affective facial expression is consistent with this, and actually suggests that it is because the affect programs of the basic emotions are phylogenetically older than our cognitive abilities that they are in part independent of these abilities. Finally, displacement of emotions is at least potentially explicable, since the affect program itself does not depend for its actual existence upon a single intentional object of the relevant kind.
These findings provide us with a powerful way to view the human mind, when affects are properly accounted for: the human mind has a hierarchy of differentiable systems. These are not only modular systems, in J. A. Fodor's sense (1983; see also Griffiths 1990, and 1997, 91-97); some of them are also more fundamental in that they are required for, and constitute part of, the function of many other systems. Thus, for example, a basic emotion that has propositional content will require capabilities that underlie the possibility of instances of that basic emotion without the cognitive content. Echoing Howard Leventhal, who hypothesizes that two distinct but parallel systems are involved in affect (1984), I can in a preliminary way illustrate the feature of a hierarchical view of mind that is important to my goals here by making a simplified, but very useful, distinction between two gross
end p.45
 
supersystems. On the one hand, there are the subcognitive affective systems (among many other subcognitive systems, such as primary perceptuomotor control systems) which include the capabilities that constitute the basic emotions, and which can operate independently of many or most of the capabilities that typify "high cognition." In terms of Paul MacLean's distinctions, this would include both the "reptilian" and the limbic systems; for Leventhal, this is the emotional or affect control system. These subcognitive systems function faster than most instances of deliberative reasoning; need not be available to report (and thus are not, in this sense, necessarily conscious or cognitive); do not require intentional content sufficient for propositional attitudes; and are intimately related to homeostatic and motor control systems (such as maintaining set points in body states, and motivating actions, including the emotional actions). On the other hand, sitting (perhaps literally, in a neuroanatomical sense) above these systems are the cognitive systems, some of which may be able to operate independently of the subcognitive systems but many of which appear to need them to function properly. These are the systems that constitute the capabilities that typify "high cognition": language, the ability to plan, the ability to report on one's deliberations, and so on. Leventhal calls this the "problem control" system.
It is also tempting to assume that the affective systems are largely or wholly subcortical, and the cognitive ones largely or wholly cortical. Although there is perhaps some truth in this, it is not necessary to assume this, since the distinction is primarily a functional one; and even some phylogenetically ancient functions have been "rewired" in primates to involve neocortical structures. The functional notion of subcognitive capabilities need not correspond to this basic anatomical distinction.
This two-tier distinction is too simple: a mature science of mind will find it more useful to refer to many systems, not easily grouped into two sets, but nonetheless clearly hierarchically arranged. However, even roughly hewn into two groups, the hierarchical view of mind is useful for drawing out a number of issues. First, it points us toward a very different way of thinking about mind, and therefore a very different kind of theory of mind, than is typical in contemporary philosophy, where critical issues are often framed in relation to propositional contents or lack thereof. The basic emotions and many other affects are clearly able to operate independently of many cognitive skills, and the neural circuits that constitute some of them appear to be centered in subcortical regions or in brain structures that are functionally independent of the kind of abilities that enable propositional attitudes. Furthermore, our evolutionary understanding of the basic emotions is encouraged by the observation that other mammals, which share with us strikingly similar subcortical anatomies, also exhibit many of the same affects, including some of the basic emotions. This is all consistent with a bottom-up view of mind, in which affects and perceptuomotor abilities are understood to be phylogenetically and functionally prior to, and likely necessary for, cognition.
end p.46
 
Second, this simplified perspective on the hierarchical view of mind also helps us to clarify where disagreements about the taxonomy of affects are, and are not, substantial. There is a great deal of implicit agreement among many scientists for the essential features of the affect program theory. Disagreements tend to arise about how much we need to add to get a full-bodied "emotion." Roughly, and using again the simplistic two-tier idealization, it may be that for some basic emotions we could outline two kinds of definitions, or identity criteria. The first, of the kind I use here, would refer primarily to the subcognitive systems to identify the capabilities and neural circuits that constitute the basic emotions. It would expect the exercise of those emotions not to require the kind of cognitive skills that are special to humans, since homologs of these emotions exist in other animals. The second kind of definition would refer also to cognitive systems, and thereby make use of a broad, or "thick," notion of the basic emotions, perhaps construing them as necessarily conscious, or necessarily propositional attitudes. Which kind of definition one should use is not an issue we need spend much time debating; I have argued that something quite like the former is a richer notion, which avoids the fundamental confusions encouraged by the latter. But the latter notion is wholly consistent with the substantive claims made throughout this book, as long as it is recognized that affective systems that are not necessarily propositional attitudes are necessary to the emotion in the thick sense. Given this, I hold no disagreements with anyone who accepts that the kind of things that happen in the affect program theory are necessary to the relevant basic emotion, but then defines that emotion in a cognitive way or even a necessarily social way. Disagreements arise, instead, with those who either (1) deny that the subcognitive elements on the hierarchy are necessary, or (2) define the basic emotions as cognitive and then use such a definition in too general a way. (The first disagreement is what we saw in reductive cognitivism: the view that the beliefs and other kinds of cognitive states are alone necessary, and the other features picked out by the affect program theory are unnecessary.) Given this understanding, a very great deal of agreement should be possible between what is said here and the majority of views on the relevant emotions.
end p.47