Much of Tamar Gendler’s dense and engaging book argues for the emotional, cognitive and motivational power of imagination, which is presented as a central feature of human mental architecture.1 But in the final chapters Gendler argues that some of us have over-exploited this resource, too easily assuming that, if belief cannot explain a class of human behaviours, imagination will do the job. She gives a number of examples of problematic behaviours (‘Gendler cases’, as we shall say), which in her view can be explained only by appeal to a previously unrecognized mental state: alief, different from belief and from imagination, and from any other mental kinds we are familiar with.
We argue that it’s a mistake to explain Gendler cases in terms of a single mental state of the kind alief is supposed to be; we should appeal instead to a variety of representational states, including familiar ones such as belief, desire, imagination and perception. While a few of these cases do plausibly require us to acknowledge representations at levels other than the personal one, none require us to acknowledge the existence of aliefs, at least as those states are officially characterized by Gendler. We then turn to one of Gendler’s more general arguments for the new category of alief: the argument from hyperopacity. We reject that argument. But all this is not simply die-hard conservativism: we conclude by elaborating the idea (somewhat in the spirit of Gendler’s proposal) that various representational states not acknowledged by folk-psychology have a role to play in explaining behaviour, emotion and cognition.
An alief is a mental state with three associatively linked components: a representational component R (‘the representation of some object or concept or situation or circumstance, perhaps propositionally, perhaps non-propositionally, perhaps conceptually, perhaps non-conceptually’), an affective component A (‘the experience of some affective or emotional state’), and a behavioural component B (‘the readying of some motor routine’) (263–4). The link between these three consists in the fact that they are co-activated by some feature of the subject’s environment, and this co-activation is difficult to break. While traditional objectual or propositional attitudes are two-place relations between a subject and a simple representational content (naturally expressed as S believes that R), alief is a four-place relation between a subject and a threefold associatively linked content of the kind just described (S alieves R–A–B).
The difference between alief and traditional cognitive attitudes is a radical one: alief is ‘developmentally and conceptually antecedent’ to such attitudes, and it is not definable in terms of them (262; 288).2 Indeed, alief’s three components don’t constitute a mere cluster of different mental states; rather they form a unified representational-affective-behavioural state, that ‘there is no natural way of articulating’ (289). Nor is the representational component of an alief a belief, or an imagining, or some other recognized state.
2. Gendler cases
Gendler cases are, roughly speaking, cases of behaviours not characterizable as entirely automatic, but also not easily thought of as actions motivated by beliefs and desires; they are often cases where habit or a quick and dirty evolved response subverts the promptings of reason. Of the many examples Gendler gives, we start with one, from which we draw a general lesson.
Take first someone reluctant to step onto the solid clear glass floor of the Grand Canyon Skywalk (call this case ‘Skywalk’). For Gendler, this is a paradigmatic instance of behaviour explained by an alief, the threefold content of which can be expressed (in an avowedly rough and ready way) as: ‘Really high up, long, long way down. Not a safe place to be. Get off!!’ (256). We agree that in this as in many other Gendler cases, three elements are crucial: a representation, an affect and something like ‘the readying of a motor routine’ (264). Assuming that the subject believes the glass floor to be safe (and why else would she even consider getting onto it?), a natural thought is that her visual perception, representing the environment as dangerous, exerts an influence on her emotions and behaviour: an influence which her rational belief in the safety of the device may not be able to overcome. But we do not see any advantage in packaging, à la Gendler, this pattern of representation, emotion and behavioural propensity as a new tripartite entity. We say:
The three elements here in play are kinds of things we have all long believed in: a visual perception, which activates a cascade of affective responses and the readying of a behavioural routine.
Those three elements are causally related, but do not make up any substantial entity.3
Both these claims may be questioned. As a response to (i), the friend of aliefs might say that a full explanation of Skywalk would require a further level of representation (the representational component of an alief) intermediate between perception and behaviour. Or she might agree – as Gendler seems to do – that the representational component here is perceptual, but insist, contradicting (ii), that the three components we all acknowledge constitute a single, previously unacknowledged object (264, note 16; 288). Against the first response, we simply ask for a reason why we should postulate an intermediate representation, when perception seems to do the job rather well. Against the second, we ask why it is that in this case three familiar elements should be regarded as constituents of one unfamiliar entity. Gendler emphasizes the tightness of the causal connections that operate in Skywalk and other cases: despite all the reasoning one may go through, it’s hard to step onto the glass floor, and even harder to avoid feelings of anxiety while looking down. However – as Gendler herself stresses in discussing other cases – these connections are not unbreakable: habituation may reduce and even extinguish a person’s reluctance to step on the floor, and the anxiety that doing so causes. Moreover, the same behaviour, and the same affective response, can be prompted by quite other representations – e.g. the belief that standing on that floor would be impious. And other kinds of states may be very closely connected without there being much value in saying that they are constituents of a single state. When people believe they are being pursued by a bull, they almost invariably fear the bull, (occurently) desire not to be caught on its horns, and try to escape. Each of these states is rarely activated except in combination with the others. But we don’t give a better explanation of bull-avoiding behaviour by saying that the belief, the desire and the emotion are components of a single entity. Humeans say that it is always a belief–desire pair that motivates; they are not thereby committed to saying that action is produced by composite entities with beliefs and desires as their parts. It is one thing to say that some representational state has a certain typical effect on emotion and behaviour; quite another to wrap the representational cause together with its emotional and behavioural effects, and label them as a single entity.4
With these thoughts in place, it is tempting to extend their application. Take a pair of Gendler cases we will call ‘Poison’ and ‘Not Poison’. Paul Rozin and colleagues found that subjects were unwilling to take a drink from a bottle marked ‘Poison’, even though they knew that the content was harmless. Again, we agree to the salience of three elements in any plausible explanation of what is going on here: a state which represents the drink as poisoned, an affective state or states, a set of activated behavioural tendencies. We agree with Gendler that the representation is not a belief: the subjects are convinced the drink is harmless. But then what is it? One plausible answer is that it is an imagining. Seeing the label, a subject imagines the drink as poisoned, and this imagining has certain unpleasant emotional effects (as imaginings often do), which intensify the closer she gets to actually drinking the stuff; naturally, she is therefore disinclined to drink it. Of course, it is not essential that we represent the subjects as thinking through a practical inference of the form: ‘I don’t want to experience the emotion E; if I do Q, I will experience E; so, I won’t do Q’. Subjects might simply feel the unpleasant effects of the imagining when they move to touch the objects, prompting them to withdraw.
Gendler argues that in this case appeals to imagining are implausible because, as it turns out, people are also unwilling (to a degree) to drink from bottles labelled ‘Not poison’. Faced with the choice of drinking from bottles labelled ‘Not poison’ or from ones filled with very same liquid but labelled ‘Safe to consume’, most subjects opt for the latter. In Gendler’s view, it is not likely that seeing a ‘Not poison’ label induces one to imagine that one actually is drinking poison:
However, we don’t need to assume that the avoidance behaviour is the product of some past but mysteriously still-active imagining, or that part of a negative imagining takes on a causal life of its own. We say instead that in many situations it is entirely plausible that seeing a message of the form ‘Not-P’ provokes one to imagine P.5 When Donne says that no man is an island, a spontaneous, appropriate and perhaps intended response is that we should imagine men as islands, though doubtless we go on to do other things with the metaphor. If a threatening Mafioso puts a label ‘Not dead’ on your loved one, that may very well be because it so easily prompts the contrary imagining. Notably, in the ‘Not poison’ case, the label also bore a skull and crossbones image preceded by the word ‘Not’. An erotically charged image preceded by the word ‘Not’ is unlikely to provoke images of celibacy.
Is the reason for [the subject’s] hesitancy supposed to be that she had been imagining that the bottle contained cyanide, though now she is not – and that what she imagined in the past (though fails to imagine now) somehow explains her action at present? Or that her current imagining that the bottle does not contain cyanide somehow contains within it (in not-fully-aufgehoben form) the antithetical imagining that the bottle does contain cyanide? And that somehow this negated semi-imagined content – content that she has, throughout the entire process, been fully consciously aware of explicitly disbelieving – sneaks into the control center for her motor routines and causes her to hesitate in front of the Kool-Aid? (269–70).
Actually, Gendler does not deny that imagination plays a role in some of the puzzling cases she discusses. As with perception in Skywalk, she grants that imagination may be operative in some of those cases. But where it is, it is mediated by alief:
Notice the slide from ‘imagination is mediated by alief’ – which we deny – to ‘imagination is mediated by an innate or habitual propensity’ – which we recognize to be very likely true for many cases. Such a propensity is just one component of alief. If the mere presence of a propensity to behave counts as an alief, we have all long believed in aliefs. And believing in such propensities settles nothing about whether, as Gendler argues, we need representational states over and above perceptions, beliefs, desires and the rest of folk psychology.6
Imagination gives rise to behavior via alief. What happens in imagination may have (non-pretend) effects beyond imagination – but it does so when the process of imagining activates a subject’s innate or habitual propensity to respond to an apparent stimulus in a particular way (299).
In sum then we agree that Gendler cases generally are the upshot of a causal process prominently involving a representation, an affect, and a behavioural routine. But we don’t agree that the representational element is always or even usually something other than a folk-psychologically acknowledged entity. We say that it very often is a perception, an imagining or a belief. Take experiments of Henk Aarts and Ap Dijksterhuis: subjects exposed to a picture of a library or of an exclusive restaurant are asked to anticipate visiting them, and subsequently display behaviours appropriate to those environments – speaking more quietly, eating more politely (call these cases ‘Library’ and ‘Restaurant’).7 Why is it implausible to suppose that the relevant representations here are instances of imagining being in a library/restaurant? Why not explain the behaviour of the sports fan watching a televised rerun who loudly encourages her favourite player (call this ‘Sports Fan’), by saying that she vividly imagines being at the match? 8 In other cases, belief and/or desire may play a role. People who honestly endorse egalitarian values sometimes display racist reactions and behaviours, such as excluding well-qualified candidates with ethnically stereotypical names from short-lists for jobs (call this ‘Racist Hiring’).9 A rational response to such a discovery about your own behaviour would surely be a lowering of confidence in your freedom from racist beliefs or desires. The result wouldn’t need to be a wholesale abandonment of the conviction that you are not racist. You might better conclude that you haven’t yet achieved consistency in your overall belief state, or that you still have desires contrary to your values: circumstances all too common in the imperfect world of human thinking.
We don’t say that the representational element in Gendler cases is always a familiar, folksy state of belief, desire, imagining or perception. Like empirically oriented philosophers of mind generally, we are open to the suggestion that the folk do not have an exhaustive taxonomy of causally distinctive mental states. Take behaviours prompted by word priming – another kind of Gendler cases. When people walk more slowly after having been primed with elderly-related words (call this ‘Slow Walking’),10 the reason may not be that priming has caused them to imagine a stereotypically older person, nor that they have been infected with elderly-representing beliefs, desires or perceptions; most of these subjects turn out to be unaware of the age-related theme within the pattern of words that were displayed. Such cases may well involve the spreading activation of elderly-related representations in a network that operates sub-personally. Of course, imagination may play some role here: imaginings prompted by some of the priming-words (‘old’, ‘retired’…) may be part of what causes the activation of stereotypical representations of the elderly. The point is whether such stereotypical representations, which are the direct cause of slow walking behaviour, are themselves instances of imagination (or of some other familiar folk-psychological state). We agree with Gendler that they probably are not; a full explanation of Slow Walking requires us to recognize also representational states at some lower level. A more obvious case where non-folk psychological representations come into play is ‘Racist Irritability’: subjects primed subliminally with African American faces show greater unconscious hostility than they otherwise would to small inconveniences (278).
Our conclusion so far is that the representational components featuring in the explanations of Gendler cases are not of one kind – a special kind, proprietary to aliefs – but of many kinds, most, though not all, of which are familiar folk-psychological kinds.
Gendler has several general arguments designed to fend off the suggestion that aliefs, or their representational components, can be replaced by more familiar states such as imaginings. We have space to deal with only one: the argument from hyperopacity. According to this argument, the representational component of alief is irreducibly different from imagination because it has the property of being hyperopaque, which imagination does not have:
Alief contexts are what we might call hyperopaque: they do not permit salva veritate substitution even of expressions that the subject explicitly recognizes to be co-referential. Even if I believe that the phrases ‘not poison’ and ‘safe to consume’ pick out co-extensive classes of substances, even if I focus on that belief and hold it vividly before my mind, even if the synonymy of these two terms is crucial to my views about some other matter, still the aliefs activated by the two expressions may be wildly dissimilar. Imagination, by contrast, is not hyperopaque in this way. If I explicitly recognize that P and Q are synonymous, and I imagine P while focusing explicitly on the co-referentiality of P and Q, then in imagining P I imagine Q. Alief just isn’t imagination (270; our emphasis).
There are two ambiguities in this argument. The first is due to an unacknowledged transition from ‘co-referentiality’ to ‘synonymy’. Since the example given in the quotation, on which we want to focus, concerns the pair ‘safe to consume’/‘not poisonous’, we will simply grant that this pair meets the condition of synonymy, and hence also the weaker condition of co-referentiality.11 The second ambiguity is more significant: the definition of ‘hyperopacity’ changes as we go from the case of alief to that of imagination – a crucial change, since Gendler’s claim is that there is a single property which alief has and imagination lacks. What is the change? Gendler specifies the hyperopacity that applies to alief as follows:
(1) Alief is hyperopaque insofar as, even if a subject vividly understands that two expressions P and Q are synonymous, the aliefs activated by P can be different from those activated by Q.
But the hyperopacity test applied to imagination is different:
(2) Imagination is not hyperopaque because, if a subject vividly understands that two expressions P and Q are synonymous, then her imagining that P will be the same as her imagining that Q.
Thus we have two distinct definitions:
As defined in H1, hyperopacity is a property that has to do with the causal power of expressions; in H2, it appears as a property concerning the identity conditions for mental states. These are quite different (and independent) properties: possession of the first, but not of the latter, is an empirical matter, depending on the contingent psychological effects of a given stimulus.
H1. A state-type S is hyperopaque if it is possible that, when a subject vividly understands that two expressions P and Q are synonymous, a token of S activated by P is different from a token of S activated by Q.
H2. A state-type S is hyperopaque if it is possible that, when a subject vividly understands that two expressions P and Q are synonymous, her S-ing that P is distinct from her S-ing that Q.
Should we understand Gendler as claiming that alief is hyperopaque in the sense of H1, while imagination is not? That claim would be false. There is every reason to think that imagination is hyperopaque in the sense of H1; indeed we don’t know of a single, conventionally recognized mental state that isn’t hyperopaque in this sense. Take the Not poison case. We grant that this case provides an instance of H1: even though the subject recognizes the synonymy of ‘not poison’ and ‘safe to consume’, these two expressions cause her to act in two different ways, because they activate in her mind two different representations. The question then is: what kind of representations? Our answer, already given, is that they are instances of imaginings. We have suggested that in the context of Rozin’s experiment the label ‘Not poison’ activates an imagining of poison, while the label ‘Safe to consume’ does not. Thus imagination is hyperopaque in the sense of H1. But so is, for example, belief. Suppose Rozin’s subjects are told that the experiment has been hijacked by a mad scientist who puts real poison in the bottle labelled ‘Not poison’; seeing that label will cause them to believe that the bottle is poisoned, while seeing the label ‘Safe to consume’ won’t. Context can always interfere with the effects an expression has on the mind. Gendler has not given any reason to think that hyperopacity of type H1 distinguishes alief from imagination, or from anything else. So there is now no barrier to thinking of imagination as the mental state that best explains Not Poison. And since Not Poison was brought on by Gendler as a way of casting doubt on the possibility that imagination explains Poison, neutralizing her argument about Not Poison allows us to explain Poison in terms of imagination as well.
What about hyperopacity as defined in H2? We can grant Gendler’s claim that imagination is not hyperopaque in this sense, while alief is.12 After all, we have never argued that alief and imagination are identical. We say that aliefs do not exist, and even if they did, their distinctness from imagination is established by definition: alieving, we are told, is a four-place relation between a subject and a threefold content, imagining is a two-place relation between a subject and a single content. Our interest is in showing that imagination can account for cases like Not Poison, and hyperopacity in the sense of H2 is irrelevant to that question. What is relevant is H1, because the state involved in Not Poison needs to be hyperopaque in the sense of H1. And H1, far from being an exclusive property of alief, is a characteristic property of imagination, as well as of all other representational states we know of.
4. Taxonomies of mental states
Our arguments so far have been defensive; Gendler has argued that old-fashioned, alief-free ways of explaining cases like Skywalk, Poison and Not Poison won’t do, and that aliefs do a better explanatory job. We have tried to show that the old way is not so very feeble. Have we established that the old way is the right way? No. Establishing that this or that behaviour is caused by this or that mental state – believing, perceiving, imagining … – is in most cases impossible; we cannot look at images of brain activation and say with confidence: ‘That’s a belief lighting up, and not an alief after all’. Our mentalistic explanations, and the categories that underpin them, can be held to only as provisional. But they should be altered or abandoned only with good reason, and we have not been given a good reason to do that for the great majority of Gendler cases.
This is not a retreat to the philosopher’s comfort zone, populated only with reassuring entities recognized by folk-psychology. We have accepted that there are representational states of various kinds, operating at various levels within a human mental economy, and that some of these representations do not fit the folk categories very well. It is natural to explain cases like Slow Walking or Racist Irritability by appeal to some kind of representational state – but not to a folksy one. Perhaps it is here that we are closest to Gendler; looking for a name for such unfamiliar representational states – non-propositional representational states over which the subject has minimal or no control – it’s tempting to call them ‘aliefs’. But we’d be left with a dispute with Gendler about what aliefs are, and about their explanatory scope. Our aliefs would be thinner, less causally efficacious things: poor cousins to Gendler’s complex, powerful and unifying entities. They would be simple representations with typical causal roles that help to explain cases like Slow Walking and Racist Irritability; they wouldn’t explain Skywalk, Poison, Not Poison, Library, Restaurant, Sports Fan, Racist Hiring or most of other Gendler cases. We’d also have a quarrel with Gendler about the underlying similarities and differences between her various cases. She claims that, by calling on alief in all these cases, we see commonalities between them that would otherwise be invisible (282; 283; 287). For us those cases, beyond some structural analogies, remain stubbornly diverse.
Our poor relatives of alief may not be the only additional representational category we need. Notice that Gendler cases are by and large characterized by their conservativeness; they are cases where behaviour is guided by habit, often in ways which subvert or constrain the subject’s openness to possibilities. There is a class of human behaviours at the opposite extreme: cases where, without the mediation of conscious reasoning or exercise of will, we generate creative solutions to practical or theoretical problems. Creative processes involve unexpected, unconventional and fruitful associations between representations and between representations and actions – without the subjects generally having access to how that happens. We suggest that both creative and habitual processes are triggered by an initial representation – not always specifiable at the personal level and sometimes occurring beyond conscious will – that leads, through a chain of barely conscious associations, to various other states, the last of which can be either another representative state or an action. While in Gendler cases, the associations lead to familiar patterns of thought and behaviour, in creative processes they lead to unpredictable, novel outcomes. Paradigmatic instances of those outcomes – jazz improvisations, free dance/speech performances – are unlikely to derive from some kind of (creative) conceptual representation, since they exceed in speed and fineness of grain the resources of conceptual thought.13 Creativity is still a great mystery, but some of the mystery may be relieved if, encouraged by the study of Gendler cases, we postulate yet another class of representations, alief-like in their non-conceptual form and in their distance from consciousness and will, but utterly different in their effects on action and in their wildly promiscuous tendency to fleeting liaisons with other representations.
As well as having poor and dull cousins, (non-existent) aliefs may turn out to have exciting and disreputable relations as well.