Eliminativism about consciousness
(with Liz Irvine)
2020 Oxford Handbook of the Philosophy of Consciousness (edited by Uriah Kriegel), Oxford University Press: Oxford, pp. 348–370
Last updated 9 July 2020
- 1 Introduction
- 2 Eliminativism
- 3 Consciousness
- 4 Entity eliminativism about consciousness
- 5 Discourse eliminativism about consciousness
- 6 Conclusion
In this chapter, we examine a radical philosophical position about consciousness: eliminativism. Eliminativists claim that consciousness does not exist and/or that talk of consciousness should be eliminated from science. These are strong positions to take, and require serious defence. To evaluate these positions, the chapter is structured as follows. In Section 2 we introduce the difference between entity eliminativism and discourse eliminativism and outline the typical strategies used to support each. Section 3 provides a brief overview of the kinds of consciousness we refer to throughout the chapter. Section 4 focuses on entity eliminativist arguments about consciousness: Dennett’s classic eliminativist argument (4.1); a rebooted version of Dennett’s argument (4.2); and recent arguments for ‘illusionism’ (4.3). In Section 5, we examine discourse eliminativist arguments about consciousness: methodological arguments from scientific behaviourism (5.1); arguments based on the empirical accessibility of phenomenal consciousness (5.2); and a stronger version of discourse eliminativism aimed at both phenomenal and access consciousness (5.3). In Section 6, we offer a brief conclusion.
If you meet an eliminativist, the first question to ask is, ‘What do you want to eliminate: entities or talk about entities?’ For any given \(X\), an eliminativist might say either or both of:
- \(X\)s do not exist
- We should stop engaging in \(X\)-talk, using \(X\)-concepts, or other practices ostensibly associated with \(X\) in science
We will call (1) entity eliminativism and (2) discourse eliminativism. Entity eliminativists claim that we should expel a specific entity from the catalogue of entities assumed to exist. This may be a matter of removing a particular individual from our ontology (e.g. Zeus), but it may also involve removing a property (e.g. being phlogisticated), an event (e.g. spontaneous generation), a kind (e.g. ghosts), or a process (e.g. extrasensory perception). In contrast, a discourse eliminativist seeks to rid science of certain ways of talking, thinking, and acting (e.g. talk about, and practices that attempt to investigate, gods, being phlogisticated, spontaneous generation, ghosts, or extrasensory perception).1
Entity and discourse eliminativism are distinct but obviously not unrelated positions. What we think regarding an entity’s existence does not, and should not, float free from what we say, think, and do in science. However, the relationship between the two is not so tight that one form of eliminativism can be inferred from the other as a matter of course.2 Someone might endorse one form of eliminativism but not the other. An entity eliminativist might do away with some entity (e.g. atoms) but decide to preserve talk, thought, and practices associated with that entity in science. For example, Mach claimed that atoms do not exist but he argued that physicists should continue to engage in atomic talk, thought, and action for their predictive and heuristic benefits: the atom ‘exists only in our understanding, and has for us only the value of a memoria technica or formula’ (Mach 1911 p. 49). Conversely, a discourse eliminativist might root out ways of talking, thinking, and acting from science but say that the entity underlying this rejected discourse nonetheless exists. In Section 5, we will see an example of this position with regard to scientific behaviourism’s treatment of conscious experience.
Let us examine entity eliminativism and discourse eliminativism more closely.
Entity eliminativism is defined by how it diverges from realism and agnosticism. A realist says, ‘\(X\)s exist’, an agnostic says, ‘We are not in a position to know whether \(X\)s exist or not’, and an eliminativist says, ‘\(X\)s do not exist’. In order for an entity eliminativist to defend her position, she needs to have a genuine, not merely a verbal, disagreement with the realist and agnostic. To this end, the eliminativist needs to make assumptions about what \(X\)s are and those assumptions need to be shared with the realist and the agnostic. The realist, agnostic, and eliminativist should agree on what \(X\)s would be like if they were to exist. What they disagree about is then whether \(X\)s do exist. Consequently, an argument for entity eliminativism generally involves two ingredients. The first is some way to identify the subject matter under dispute that is acceptable to all sides (realist, agnostic, and eliminativist). This is often done by providing a description of the essential properties of the entity, but, as we will see in Section 4.2, that is not the only way to do it. The second ingredient is an argument to show that no such entity exists. If the entity is identified by description, the second step may involve showing that no entity satisfies this description. According to Mallon et al. (2009), this kind of argument was used to defend eliminativism about beliefs (and other propositional attitudes): first, claim that in order for something to be a belief, it must satisfy a certain description D (given in this case by folk psychology); second, argue that nothing satisfies description D (because folk psychology is false); third, since nothing satisfies D, conclude that beliefs do not exist.
In contrast to entity eliminativism, discourse eliminativism targets talk, thought, and behaviour in science. Let us say that the concept ‘ding dong’ refers to nanoscale, spherical, tentacled lifeforms. A discourse eliminativist about ding dongs says that scientists should stop talking, thinking, and pursuing research programmes about ding dongs. One motivation for this may be the conviction that there are no ding dongs (i.e. one is an entity eliminativist about them). But being an entity eliminativist is neither necessary nor sufficient for being a discourse eliminativist about them. One might think that ding dongs exist (or be agnostic about them) but argue that scientists should avoid ‘ding dong’ talk because it is unproductive, misleading, or otherwise unhelpful. Conversely, one might think that ding dongs do not exist but argue that ‘ding dong’ talk, thought, and practice is useful to science and should be preserved: perhaps the ding dong concept is a useful way to group lifeforms or encourages useful practices (looking for entities at certain spatial scales).
Arguments for discourse eliminativism typically consist in defending a negative and a positive thesis. The negative thesis aims to establish that the talk, concepts, and practices targeted for elimination are somehow unhelpful, damaging, misleading, or otherwise problematic. In the case of conscious experience, a discourse eliminativist might argue that discourse about conscious experience is too subjective, hard to verify, does not generalise well, does not pick out a natural kind, produces intractable disagreements, does not cohere with other scientific talk, or otherwise leads to a degenerative scientific research programme. However, even if these points land, they rarely suffice to motivate discourse eliminativism. Science seldom switches course unless a better alternative is available. The positive part of a discourse eliminativist’s argument aims to show that an alternative way of talking, thinking, and acting is available. The discourse eliminativist argues that this proposed alternative discourse is, on balance, better for achieving our scientific goals than the one targeted for elimination. Diverse virtues may weigh in this decision, including purely epistemic virtues (e.g. telling the truth, not positing things that do not exist) but also predictive, pragmatic, theoretical, and cognitive virtues.
In this chapter, we make use of Block’s distinction between access consciousness and phenomenal consciousness (Block 1990, 2007). ‘Access consciousness’ refers to the aspects of consciousness associated with information processing: storage of information in working memory, planning, reporting, control of action, decision making, and so on. ‘Phenomenal consciousness’ refers to the subjective feelings and experiences that conscious agents enjoy: the feel of silk, the taste of raspberries, the sounds of birds singing, and so on. The latter is the ‘feel-y’, subjective, qualitative, what-it-is-like-ness, ‘from the inside’ aspect of consciousness. We use the term ‘qualia’ to refer to this subjective feeling.3 Following Frankish (2016a), we define ‘experience’ in a purely functional way: mental states that are the direct output of the sensory system. This means that we do not assume that experience necessarily involves phenomenal consciousness.4
This chapter focuses on eliminativism about phenomenal consciousness. Access consciousness will make an appearance in the final section (5.3). For the purposes of this chapter, we do not presuppose anything about the relationship between access and phenomenal consciousness, although some of the eliminativist arguments discussed below do take a stand on this.
One of the striking features of entity eliminativism about consciousness is that it is perceived as a philosophical position that is self-evidently wrong. Critics say that questioning the existence of phenomenal consciousness is impossible. Each of us knows, by introspection, that we have phenomenal consciousness – what is more, we know this in a way that is not open to rational doubt. Not even Descartes doubted the existence of his subjective experience. Yet the eliminativist does. Assessments of eliminativist claims have been correspondingly harsh. Frances writes, ‘I assume that eliminativism about feelings really is crazy’ (Frances 2008 p. 241). Searle, ‘Surely no sane person could deny the existence of feelings’ (Searle 1997). Strawson says that eliminativists ‘seem to be out of their minds’, their position is ‘crazy, in a distinctively philosophical way’ (Strawson 2010 p. 101). Chalmers, ‘This is the sort of thing that can only be done by a philosopher, or by someone else tying themselves in intellectual knots!’ (Chalmers 1996 p. 188).5 How can anyone deny such a self-evident truth about our mental life? (As we will see in Section 4, entity eliminativism is typically combined with an attack against the reliability of our introspective access.)
The discourse eliminativist about phenomenal consciousness faces a similar, though perhaps not quite so daunting, challenge. Discourse eliminativists identify problems with a scientific discourse and seek to offer a better alternative. The challenge faced by a discourse eliminativist about phenomenal consciousness is that phenomenal consciousness appears to be an overwhelmingly important part of our mental life. Humans care about their phenomenal feelings: about the feelings that accompany eating their favourite dish, scoring a winning goal, being punched in the kidneys, or having their toes tickled. These feelings play a valuable, although hard to specify, role in our cognitive economy. For this reason, it seems that some reference to phenomenal consciousness should be made by any scientific psychology. A science that never talked about phenomenal consciousness would be incomplete. Even if phenomenal feelings do not exist (as an entity eliminativist says), science should still talk about phenomenal consciousness in order to explain why we (falsely, according to the entity eliminativist) take ourselves to be motivated by such feelings. Eliminating talk of phenomenal consciousness appears to ignore a significant aspect of human mental life and amounts to a failure of ambition for scientific psychology.
4 Entity eliminativism about consciousness
In this section, we examine three entity eliminativist arguments about phenomenal consciousness. The first is Dennett’s ‘Quining qualia’ argument (1988). The second is a rebooted version of Dennett’s argument that aims to avoid the standard objection to that argument (namely, that Dennett mis-characterises phenomenal consciousness). The third is the recent research project of ‘illusionism’, which is related to Dennett’s ‘Quining qualia’ argument but motivated on somewhat different grounds.
4.1 Dennett’s eliminativism about qualia
Dennett’s ‘Quining qualia’ paper looks, at least at first glance, like a classic entity eliminativist argument: it describes the essential properties of the alleged entity; shows that nothing satisfies this description; and concludes on this basis that no such entity exists. The description that Dennett gives of phenomenal consciousness (‘qualia’) is that it is ineffable (not describable in words), intrinsic (non-relational), private (no inter-personal comparisons are possible), and directly accessible (via direct acquaintance). The final property is related to the idea that we have privileged, incorrigible, or infallible access to qualia. Dennett argues that no entity satisfies this description. As a result, ‘Far better, tactically, to declare that there simply are no qualia at all’ (Dennett 1988 p. 44).
Dennett uses a number of ‘intuition pumps’ to get to this conclusion, which we summarise here.
First, it is plausible that how things subjectively feel is bound up with how you evaluate, or are able to categorise or discriminate between, your experiences. Someone’s first taste of a particular wine may be different to how it tastes to them after having become a wine aficionado. At first, my taste of wine was bound up with judgements of yukkiness and an inability to easily tell one wine from another. My current taste of wine is bound up with gustatory enjoyment and an ability to finely discriminate between different wines. This suggests that qualia are not intrinsic properties of experience: that the taste of this specific wine does not have a particular qualitative feeling (quale) for me independently of how I evaluate or categorise it. Instead, the way that wine (and other things) consciously tastes to me is at least partly determined by relational properties, such as whether I like it, and whether I can tell a Pinot Grigio from a Chardonnay.
This also puts pressure on qualia being directly accessible: it looks like I can’t tell very much about my qualia from introspection. Say that as you get older, you start liking strong red wine more. One possibility is that your sensory organs have changed, making strong reds taste different and more pleasurable compared to how they used to. On this scenario, you now have different qualitative feelings (more pleasant ones) on tasting strong reds than you had before. Another possibility is that your sensory organs have stayed the same but your likings for specific experiences have changed. You have roughly the same conscious feelings but now you like those feelings more. On the first scenario, your qualia change; on the second scenario, your qualia stay the same. Dennett puts it to us that we would not be able to tell, merely from introspection, which scenario we are in. Yet this should be easy to do if we really had direct (or infallible or incorrigible) access to our qualia.
With respect to the putative ineffability and privacy of qualia, Dennett refers to Wittgensteinian arguments that render entirely private and incommunicable states senseless. Dennett goes on to argue that there is a way in which experiences are practically ineffable and private, just not in the ‘special’ way intended by qualiaphiles. Imagine two AI systems that learn about their environment in a fairly unsupervised manner and so go on to develop different internal systems of categorising colour (adapted from Sloman & Chrisley 2003). These two systems will end up with some states that are (at least to some degree) private and ineffable. One system’s ‘blue’ states will be somewhat different to the other system’s ‘blue’ states just in virtue of the internal differences in the systems (e.g. the ‘blue’ states of each system will be triggered by a slightly different range of hues). In the same way, humans can be in distinct (practically) ineffable and private states because of idiosyncrasies in their perceptual and cognitive processing. Some of these differences one may discover empirically (e.g. that you and I disagree about whether a particular paint chip is blue), and so we can make our experience more ‘effable’, and less private. Dennett’s point is that ineffability and privacy of experience only amounts to this: practical and graded difficulties in assessing which internal state we are in, not essential properties of our experience.
In light of Dennett’s considerations, it looks like our supposedly phenomenal experiences do not satisfy the description associated with them: there is nothing ineffable, intrinsic, private, and directly accessible to an experiencer that determines the way that things (phenomenally) seem to them. Whatever produces our judgements and reports about phenomenal experience, it is not an entity of the hypothesised kind. Qualia, as characterised by Dennett’s description, do not exist.
A popular response to Dennett is to say that qualia were not successfully characterised by his description. This effectively undermines his argument at the first step. Perhaps partly for this reason, qualiaphiles nowadays tend to favour a minimalistic characterisation of qualia. Qualia need only have a phenomenal character (a ‘what-is-it-like-ness’ or subjective feel) (Carruthers 2000; Kind 2001; Levine 2001; Tye 2002). They need not be intrinsic, private, ineffable, or directly accessible properties of experience. Frankish (2012) describes this as the move from ‘classic qualia’ to ‘diet qualia’.6 Classic qualia are controversial entities; diet qualia are not:
Philosophers often use the term ‘qualia’ to refer to the introspectively accessible properties of experiences that characterize what it is like to have them. In this standard, broad sense of the term, it is very difficult to deny that there are qualia. There is another, more restricted use of the term ‘qualia’, under which qualia are intrinsic, introspectively accessible, nonrepresentational qualities of experiences. In my view, there are no qualia, conceived of in this way. They are a philosophical myth. (Tye 2002 p. 447)
Dennett’s intention is to eliminate both classic qualia and diet qualia.7 His argument, however, does not seem to engage with diet qualia: his description fails to pick out what qualiaphiles have in mind here.8 In the next section, we rework Dennett’s eliminativist argument to explicitly target diet qualia.
4.2 ‘Quining qualia’ rebooted
Dennett’s ‘Quining qualia’ argument attempted to identify qualia by description. Once we switch to diet qualia, appeal to description appears to be of questionable use as nothing identifies qualia apart from their (contested) phenomenal feel. So, rather than attempt to identify the target for elimination (or realism) by description, we should identify it in some other way.
A common strategy is to identify qualia by a kind of ostension: ask one to consider specific examples of qualia and then generalise to the kind they have in common.9 One identifies the subject matter at issue by asking one’s interlocutor to consider those of her experiences that allegedly have qualia (consider the feel of silk, …), drawing her attention to the supposedly felt aspects of these experiences, and saying more aspects of mental life of this kind. A set of examples, and how they are relevantly similar, are thus intended to fix the meaning of ‘qualia’. The realist and eliminativist may agree on this strategy: they may agree on the set of examples and how we take them to be similar (for example, they may have no difficulty in generalising to new cases). What the realist and eliminativist disagree about is whether the phenomenal feelings that appear to be present in these cases really are present. The realist claims that the experiences have, or instantiate, a property, ‘what-it-is-like-ness’, which should be added to our ontology. The ‘what-it-is-like-ness’ or phenomenal character is a real property of experience – as real as anything. The realist says that explaining what this phenomenal property is, how it comes about, and how it relates to physical and neural properties is the job of a theory of consciousness. In contrast, the entity eliminativist says that no such property exists (or is instantiated in the relevant cases). According to her, the examples are, in a sense, deceptive: they appear to show instantiation of a property, but that appearance is wrong. There is no such property of experience.
An entity eliminativist denies the existence of qualia, but she does not deny the existence of many of our judgements, beliefs, and desires about qualia. This allows her to agree with much of what a realist says about experience. She can agree that we believe that our experience has qualia, that it is hard for us to doubt that our experience has qualia, and that our beliefs and judgements motivate us to act in appropriate ways. Nevertheless, the eliminativist says, these beliefs and judgements are false. They are comparable to the beliefs and judgements that the ancient Greeks held about Zeus: deeply held and capable of motivating action, but fundamentally mistaken. We should no more take on the project of explaining what qualia are, how they arise, and how they relate to physical properties than we should for Zeus.
We divide the rebooted version of Dennett’s argument into two steps. The first step aims to defend a sceptical claim: that we do not know which (diet) quale our current experience instantiates. This claim goes beyond a mere failure of infallibility or incorrigibility. The claim is that we lack any knowledge at all about which quale our current experience instantiates. The second step leverages this epistemic claim to argue for qualia eliminativism. If the instantiation of one quale rather than another is unknowable, then instantiation of a quale is a difference that would make no difference to the world; on that basis, qualia should be eliminated.
As a starting point, notice that it is sometimes hard to tell which quale your current experience instantiates. Slow, subtle changes in experience may leave one uncertain about which subjective feeling you have – is the quale you have on looking at an Yves Klein blue painting now the same the one as you had a minute ago? Examples like this may present us with epistemically ‘bad’ cases of qualia knowledge: situations where for some reason we are unsure about which quale we have. Showing that there are some ‘bad’ cases, however, does not show that we can never know which quale we have.10
Focus instead on the apparently ‘good’ cases of qualia knowledge: cases where we appear to know whether our qualia have changed or are the same. Such cases often involve sudden or dramatic changes in one’s experience. If the Yves Klein painting facing you were suddenly exchanged for a bright yellow painting, you would know, not only that the painting had changed, but also that your qualia had changed (perhaps you would regard the latter as evidence for the former). Dramatic changes in experience seem to provide good cases of qualia knowledge. However, even in these ‘good’ cases, there are reasons to think that one lacks knowledge about which quale one has.
Suppose that while Lara is asleep a neurosurgeon operates on her brain. On waking, she finds that the world looks different: objects that before looked blue now look yellow. No one else notices the change, so Lara concludes that something must have happened to her. Prima facie, Lara appears to have justification to think that her qualia have changed. However, the change she notices is compatible with two hypotheses:
- Lara’s qualia have changed from those she had yesterday.
- Lara’s qualia remain the same but her memories of her past experiences have changed.
Lara’s post-surgery experiences are consistent with either hypothesis Q or R. On the basis of introspection, Lara knows that things ‘look different’, but she cannot tell what is responsible: a change in her memory, a change in her qualia, or some combination of both. The problem is more serious than a lack of certainty or failure of infallibility. Q and R appear to be equally well supported by Lara’s introspective evidence. Introspection alone appears to give her no knowledge at all about whether her qualia have changed.
Lara, however, has access to sources of knowledge other than introspection. What if she were to look at changes in her brain? For the sake of argument, put Lara into the strongest possible epistemic position with regard to the physical state of her brain. Suppose she has perfect neuroscientific knowledge and full scans of her brain before and after surgery. Furthermore, suppose (unrealistically, but helpfully for Lara) that there is a clear separation in the neural basis of Lara’s sensory systems and the neural basis of her memory systems, and that this is known to her. She can then reason as follows:
If the brain scans reveal that the neural change affected only my sensory systems and left my memory systems intact, I have reason to favour Q over R because only the systems that support my current experience, and not those that support my memories of past experiences, have been affected. Conversely, if the scans reveal that the neural change affected only my memory systems and left my sensory systems intact, then I have reason to favour R over Q because only the systems that support my memories, and not those that support my current experience, have been affected.
Thus it seems that empirical evidence can do for Lara what introspection alone cannot: it can give her reason to favour Q over R. (Of course, it is possible that both Lara’s sensory and memory systems have been affected by the surgery, but we will ignore this possibility as it would not help her.)
The problem is that the brain scans only provide Lara with information about changes to her neural events. This can help her in deciding between Q and R only if she knows where in the causal chain of those neural events her qualia experience is instantiated. Lara, however, does not know this, and by hypothesis it is not part of the (thin, minimal) concept of diet qualia. Consider two competing hypotheses about where her qualia experience occurs in the causal order of the neural events:
- Sensation \(\rightarrow\) qualia experience \(\rightarrow\) memory access
- Sensation \(\rightarrow\) memory access \(\rightarrow\) qualia experience
Lara’s reasoning assumed something like (1) is true: her qualia are instantiated after sensation but before memory access. Memory access is causally downstream from the qualia experience. This suggests that any surgery-induced change to the neural basis of her memory system should have a different effect on her qualia experience to any surgery-induced change to the neural basis of her sensory system: one affects something after her qualia experience, the other affects something before. If Lara were to discover that the neural change exclusively targeted her memory system, that suggests her current qualia experience has been unaffected, because the causal antecedents of that experience would be identical to what they were yesterday. However, if she discovers a neural change to her sensory system (e.g. that sensory channels that carry colour information have been swapped), and no change to the neural basis of her memory system, that suggests that a change to her sensation and hence to her current experience.
Unfortunately, this justification vanishes if (2) were true. On (2), Lara’s qualia experience is causally downstream from both her sensation and her memory access. A surgery-induced change to the neural basis of either system could then potentially affect her current qualia experience. A change to the neural basis of her memory could produce a change in her current qualia or a change in her memory of past experiences, or both. A change to the neural basis of her sensory system could produce a change in her current qualia or a change in the outputs of her memory systems, or both. The two factors about which Lara can detect change (the neural bases of her memory and sensation) both lie causally upstream from her qualia experience, and so are confounded in any causal inference about that experience.
No one knows whether (1), (2), or any number of other proposals about the location of qualia in the causal order is correct. This is not a limitation of the scientific resources we supply Lara – a better scanner or more neuroscientific data would not help. Nor is it something with which the concept of diet qualia can help: that concept is silent about where qualia are instantiated in neural events.
One might attempt to remedy this by correlating Lara’s introspective reports of qualia with her neural events to find out where among those events her qualia experience falls. However, we have already seen that there is no reason to trust Lara’s introspective reports about her qualia (when they occur or when they change) in this context.11 Thus, neither introspection, nor empirical knowledge, nor some combination of the two tells Lara which quale her current experience instantiates. Even in an apparently ‘good’ case, there is no reason to favour Q or R. Given that our own epistemic position is usually worse than Lara’s, our own qualia may, for all we know, be changing without us noticing.12
This accounts for Step 1 of the argument. Step 2 says that if changes in one’s current qualia are unknowable (either by introspection or by methods available to empirical science), then we should eliminate qualia from our ontology. The thought behind this is that Step 1 has shown that qualia are an extra ‘wheel’ that do not turn anything. Qualia have no discernible or characteristic effects on the world – for if they did, Lara could exploit those effects to detect changes in her qualia. A quale’s effect on us is always confounded with that of other factors (such as memory). Therefore, affirming the existence of qualia as independent entities/properties in our ontology seems unmotivated. Like Wittgenstein’s ‘beetle’,
[This] thing in the box has no place in the language-game at all; not even as a something; for the box might even be empty – No, one can ‘divide through’ by the thing in the box; it cancels out, whatever it is. (Wittgenstein 1958 sec. 293)
It is open to a qualia realist to insist that qualia should still be included in our ontology irrespective of our inability to independently track them (similarly, a beetle realist could insist that there really is a beetle in the box). But the realist’s position now begins to look unmotivated. The eliminativist has grounds, in contrast, to deny that qualia exist. Qualia do not earn their ontological keep. As inessential cruft, they should be eliminated.
Unlike Dennett’s original argument, the rebooted argument does not rely on the assumption that qualia are intrinsic, private, ineffable, or directly accessible. Before closing, we wish to flag two problems with the rebooted argument.
First, one might wonder why, even if the argument is correct, it nevertheless still seems to us that qualia exist. This ‘seeming’ does not go away even if one accepts the eliminativist’s conclusion. On this basis, one might press for a residual role for qualia that provides more than an eliminativist would allow: more than merely being associated with a set of dispositions to make judgements, or with having a set of beliefs about qualia (both compatible with those judgements and beliefs being false). Our relationship to qualia appears to be more primitive than this. It seems to us that our experiences have qualia and this ‘seeming’ is the evidence for our beliefs about qualia. How can this impression, this pre-doxastic ‘seeming’, be produced? One might tell a mechanistic and adaptationist story about how humans arrive at their false beliefs about qualia (Dennett 1991; N. Humphrey 1992). But what mechanistic story can be told that explains the production of seemings that generate and appear to confirm these beliefs? This is the ‘illusion problem’, discussed in the next section, and it remains an unsolved challenge for eliminativists.
Second, one might object to the qualia scepticism of Step 1. Step 1 relies on questioning the reliability of memory-based comparisons. A realist, however, might concede that Lara does not know whether a quale she has today is the same as one she had yesterday (perhaps because of confounds with memory), but deny that she lacks any knowledge at all of which quale she currently has. Imagine looking out on a mountain scene with green grass, grey rock, and blue sky. Multiple qualia are instantiated simultaneously in your current experience: what-it-is-like to see green, what-it-is-like to see grey, what-it-is-like to see blue, and so on. You can tell the difference between these qualia (you can make similarity judgements, detect that there are many qualia instantiated versus a few, distinguish between your visual, auditory, and proprioceptive qualia, and so on). None of these judgements appear to rely on memory comparisons. Within the domain of current experience, therefore, you appear to have some knowledge about which qualia your experience instantiates. But then, why think that qualia are wheels that turn nothing or are always confounded with memory in their effect on you?
4.3 The illusionist movement
Recently, interest in eliminativist approaches to phenomenal aspects of consciousness has been rekindled by Frankish, in particular in a special issue of the Journal of Consciousness Studies. Frankish outlines ‘illusionism’ as the view that experiences have no phenomenal properties and that our phenomenal feelings are ‘illusory’. We think we have experiences with phenomenal properties, but in fact we do not. Illusionism is a form of entity eliminativism about phenomenal consciousness even if the label ‘eliminativism’ is avoided for rhetorical reasons. It is motivated somewhat differently to Dennett’s entity eliminativism, and has a slightly different focus, so is worth discussion in its own right.13
First, illusionism is partly motivated by taking seriously the idea that phenomenal properties, and phenomenal consciousness, cannot be accounted for scientifically. Illusionism is seen as a way out of this problem. Second (and relatedly), the reasons for favouring illusionism are mainly rather general, theoretical reasons. The theoretical virtue of simplicity, or conservativism, suggests that the fewer entities/properties the better. Since illusionism gets rid of the metaphysically and epistemically problematic phenomenal properties, illusionism is better than alternative realist positions. Third, illusionism is often argued to be a research programme rather than a set of worked-out claims, and that this research programme is worth pursuing more than its alternatives. As we will see, illusionism comes with a range of difficult open questions.
Illusionism follows a slightly different tack to the typical argument for entity eliminativism described in Section 2. The first step is supposed to be to identify the contested entity/property in a way that can be generally accepted. This is not straightforward for phenomenal properties (see discussion in both this section and the preceding one). Second, the arguments motivating illusionism are not direct arguments to the effect that phenomenal properties, as described, do not exist; the position is largely motivated on other grounds (e.g. theoretical simplicity). The third step of the classic argument is to conclude that phenomenal properties do not exist. This is also concluded by some proponents of illusionism, but one could arguably also treat illusionism as a promising research programme without committing to this conclusion in advance.
Challenges to illusionism come in roughly three forms (the first two roughly track two of the steps above).
First, one might argue that it is neither obvious nor universally accepted what phenomenal consciousness is, or what phenomenal properties are, such that a proposal to eliminate them is comprehensible. Mandik (2016) states that ‘phenomenal’ is a technical (not folk) term, but one that is not clearly defined. As such, both eliminativist and realist talk about ‘phenomenality’ is unwarranted; in neither case is there a clear target to be eliminativist or realist about. Schwitzgebel (2016) tries to provide a minimal ‘definition by example’ that is not committed to any particular (troublesome) metaphysical or epistemic commitments, but as Frankish (2016b) points out, this is not substantive enough to sway the debate one way or the other.
Second, one might reject some of the main theoretical motivations for thinking that illusionism is the best or most reasonable philosophical position available. For example, Balog (2016) defends the phenomenal concept strategy, which preserves realism about phenomenal properties but concedes the existence of an explanatory gap. Prinz (2016) also defends a realist account of phenomenal properties, but one that tries to close the explanatory gap by providing neuroscientific explanation of at least some aspects of phenomenal consciousness. More generally, unless one is convinced that the theoretical virtues of illusionism (ontological parsimony, fit with existing non-phenomenal science, avoidance of the hard problem of consciousness) are superior to rival positions on consciousness, one is unlikely to be persuaded of illusionism.
Third, a cluster of worries arise around the ‘illusion problem’. This concerns how to account for the alleged illusion of phenomenality. How can one have experiences that appear to have phenomenal properties without any phenomenal properties existing? Frankish (2016a) labels those physical properties (perhaps highly disjunctive and gerrymandered) that typically cause us to misrepresent ourselves as having phenomenal qualities, ‘quasi-phenomenal properties’. Quasi-phenomenal redness is, for example, the physical property that typically causes (false) representations of phenomenal redness in introspection. According to Frankish, it is the tokening of these false introspective representations that is responsible for the illusion of phenomenal consciousness. He likens their effect on us to that of other resilient, mistaken perceptual representations such as those of impossible figures like the Penrose triangle (N. Humphrey 2011) or of colours as ‘out there’ in the world (Pereboom 2011).
The worry is how exactly this is supposed to work. It is not clear how a false representation caused by non-phenomenal properties could produce an appearance or ‘seeming’ of phenomenality. And as Prinz puts it, ‘what is it about beliefs in experience that causes an illusion of experience?’ (2016 p. 194). How is it that these representations cause illusions of subjective experience when other sorts of false representations do not? Related to this is a worry about how such false introspective representations get their content (Balog 2016). Representations of phenomenal feelings are not like other empty or non-referring representations (‘unicorn’, ‘the largest prime’), which seem to get their content by being semantic composites from representations that do refer (‘horse’, ‘horned’, ‘largest’, ‘prime’). Representations of phenomenal experience do not seem to decompose into representations of non-phenomenal properties at all.
Illusionism promises to get us away from the hard problem. It effectively eliminates the ‘data’ the hard problem asks us to explain – phenomenal feelings. Prinz (2016) argues that the illusion problem and the hard problem in fact face similar difficulties. In both cases, we need to identify what phenomenal properties are. In the hard problem, we need to explain how phenomenal properties come out of ‘mere matter’: how feelings arise in an apparently non-phenomenal system. In the illusion problem, we need to explain how (vivid!) illusions of phenomenality come about in entirely non-phenomenal systems. The challenge is to explain how an illusion of phenomenality (worthy of that name) arises in a non-phenomenal system. In both cases then, one needs to explain how something suitably like phenomenality arises from ‘mere matter’. By the time one has done this, it might be just as easy to be a realist as an illusionist.
Frankish (2016a) briefly discusses the relationship between illusionism and discourse eliminativism: ‘Do illusionists then recommend eliminating talk of phenomenal properties and phenomenal consciousness? Not necessarily’ (p. 21). We agree. However, Frankish goes on to suggest that a commitment to discourse eliminativism can only be avoided by an illusionist if the phenomenal terms in science are redefined to refer to quasi-phenomenal properties. This seems to us neither necessary nor likely.
First, as Frankish says, it would depart from what these terms usually mean in other contexts, and so would invite confusion. Second, although we agree with Frankish that an illusionist scientific psychology is likely to need to talk about quasi-phenomenal properties, this could most naturally be done with a response-dependent characterisation of those properties: refer to the physical properties that typically give rise to specific (false) phenomenal representations. Keeping track of quasi-phenomenal properties does not require changing the meaning of phenomenal terms in science. Third, as Frankish (2016b) says, it is no part of illusionism to say that the illusion of conscious experience is not important or useful to the experiencer. Graziano (2016) and Dennett (1991) argue that phenomenal consciousness plays an important and evolutionarily explicable role in our mental lives. It is reasonable to expect that scientific psychology would therefore want to study it. This study could be done while bracketing questions about the existence of phenomenal properties.14 In a similar way, a scientific psychology that studied childhood dreams might talk about the role of representations of Santa Claus and unicorns in a child’s cognitive economy without attempting to redefine those terms to refer to the disjunctive collection of physical properties that typically cause the child to token those false representations. Talk of phenomenal feels can remain in science, albeit with the disclaimer that the entities that allegedly stand behind this talk do not exist.
5 Discourse eliminativism about consciousness
We now turn to discourse eliminativism. Discourse eliminativism seeks to rid science of talk, concepts, and practices associated with phenomenal consciousness. In this section, we look at three discourse eliminativist arguments. The first is based on concerns raised by psychologists at the start of the twentieth century. The second is based on more contemporary concerns about how to study phenomenal consciousness independently of access consciousness and the mechanisms of reportability. The third is based on the worry that the concept of consciousness fails to pick out a scientifically usable category of phenomena.
5.1 Scientific behaviourism
One of the goals of scientific psychology in the first half of the twentieth century was to redefine psychology, not as the study of the mind, but as the study of observable behaviour. Scientific behaviourists argued that scientific psychology should avoid talk of internal mental states, and in particular, talk of conscious states (Hull 1943; Skinner 1953; Watson 1913).
The rise of behaviourism in science was at least partly due to the perceived failure of an earlier attempt to pursue scientific psychology via use of introspection (Titchener 1899). The debate on the nature of imageless thought was held up as an example of how unproductive that research programme was. One side in the debate appealed to introspection to argue that all thoughts were analysable into images; the other used similar evidence to argue for the opposite conclusion. The disagreement was widely seen as impossible to resolve because the evidence from the two sides could not be compared in an unbiased way. By the mid-twentieth century, introspective methods were discredited and study of conscious experience in science largely abandoned (G. Humphrey 1951).
Scientific behaviourists sought to reform psychology in such a way as to avoid these methodological difficulties. The subject matter of scientific psychology should be publicly observable, publicly verifiable, or independently experimentally controllable events. Scientific psychology should eliminate talk of conscious experience and use of introspective methods. However, this did not mean that behaviourists thought that mental states, including states of phenomenal consciousness, did not exist: ‘The objection to inner states is not that they do not exist, but that they are not relevant in a functional analysis’ (Skinner 1953 p. 35).15 Scientific behaviourists proposed an alternative way of talking, thinking, and acting that they argued was superior (in predictive, explanatory, and methodological terms) to a scientific psychology than one that appealed to, or attempted to study, conscious experience. Phenomenal consciousness, notwithstanding its ontological status, should be excluded from the realm of scientific psychology.
Positivistically-inclined philosophers argued, based on related considerations about verification and public accessibility, for various ontological and/or semantic lessons about conscious experience (Ryle 1949; and less clearly, Wittgenstein 1958). This prompted them to redefine mental state language in terms of behavioural dispositions and/or to eliminate qualitative conscious feelings from ontology. However, connecting these two lines of thought – one about utility to scientific practice and the other about ontology/semantics of ordinary language – requires accepting auxiliary claims about verifiability, the role of science, and the scope of our knowledge. Such links are widely questioned today. Many scientific behaviourists did not perceive such links at the time either and they argued for the elimination of talk of conscious experience from science based on pragmatic rather than ontological/semantic concerns.
5.2 Eliminativism via independent access
A different methodologically motivated form of discourse eliminativism about phenomenal consciousness is found among some consciousness researchers today. It stems from problems involved in trying to operationalise consciousness, or in finding ways to experimentally probe it.
One way of operationalising consciousness is via some kind of reportability: a subject is conscious of a stimulus if and only if they report it or respond to it in some way. This sounds fairly straightforward, but there are problems with using reportability as a marker for the presence of phenomenal consciousness, rather than as a marker for the cognitive capacities associated with consciousness. These problems can motivate a position of discourse eliminativism about phenomenal consciousness.
First, consider the distinction between phenomenal and access consciousness. Phenomenal consciousness refers to felt conscious experiences, (diet) qualia, raw feels, and so on. Access consciousness refers to the aspects of consciousness that are associated with, or that can be used in, cognitive capacities like reasoning, action, verbal report, and so on. If we somehow knew that access and phenomenal consciousness were always bound together (no cognitive access without phenomenal consciousness and vice versa), then scientific ways of probing access consciousness would also function as scientific ways of probing phenomenal consciousness. That is, if phenomenal consciousness and access consciousness always go together, then probing access consciousness just is to probe phenomenal consciousness. In this case (absent any other problems), it would be perfectly legitimate for the term ‘phenomenal consciousness’ to figure in scientific discourse, because the phenomenon it picks out is scientifically accessible.
The problem is that it is not obvious whether the aspects of consciousness picked out by access and phenomenal consciousness are always co-present. According to Block (1995), there may be instantiations of phenomenal consciousness (raw feels) without any related cognitive access (ability to respond to or report about these raw feels). Block has outlined a number of examples where this might happen, including situations where subjects may have highly detailed and specific phenomenal experiences, but be unable to report the details of them (Sperling paradigm); cases of phenomenal consciousness of unattended items; and possibly cases of hemi-spatial neglect, where subjects do not appear to have access to phenomenal experiences from some part of their visual field (see Block 2007, 2011, 2014; Irvine 2011; Phillips 2011 for discussion). In most of these cases, there is evidence that subjects are at least processing sensory information that they are unable to report about. Block’s claim is that there is a layer of untapped and unaccessed phenomenal consciousness present in these cases, in addition to whatever can be overtly reported or measured.
The lack of a way to probe the phenomenal aspect of consciousness independently of the accessibility aspect makes it difficult (or impossible) to scientifically assess these claims. It looks as though any way of probing phenomenal consciousness requires that the experience have some measurable effect on the subject, possibly such that she can report it in some way. That is, accessing phenomenal consciousness relies on it being associated with some kind of cognitive function or capacity, therefore accessing phenomenal consciousness relies on it being associated with access consciousness. So, if an instance of phenomenal consciousness is not associated with access consciousness, then it looks like we cannot tell if it is present or not. As Dehaene et al. (2006) note, whether participants in an experimental situation ‘actually had a conscious phenomenal experience but no possibility of reporting it, does not seem to be, at this stage, a scientifically addressable question’ (p. 209).
Partly in response to this worry, Block, Lamme, and colleagues have argued for the possibility of indirectly investigating these purported instances of phenomenal consciousness without accessibility (Block 2011, 2014; Lamme 2006; Sligte et al. 2010). The idea here is to find some reasonable and measurable marker for the presence of consciousness in cases where phenomenal (and access) consciousness is clearly present (call this marker, \(M\)). The marker could be a particular neurophysiological signature (e.g. evidence of strong feed-forward processing), or a behavioural marker (e.g. ability to complete a particular type of task based on a set of visual stimuli). One then argues that if the special marker \(M\) is present in a subject, then regardless of whether the subject appears to be conscious of the test stimulus according to other standard measures of (access) consciousness, the subject is phenomenally conscious of that stimulus. That is, marker \(M\)’s presence guarantees that a subject is phenomenally conscious of the test stimulus, even if they do not report seeing it, or cannot perform a range of actions that we usually associate with being conscious of a stimulus. The subject is phenomenally conscious of the stimulus without having cognitive access to that experience.
However, problems of interpretation abound here. Such behavioural and neurophysiological evidence could be taken as indirect evidence of phenomenal consciousness without access consciousness, but it could also be interpreted as evidence of unconscious processing (i.e. that we got the special marker \(M\) wrong), or of graded cognitive access and phenomenal consciousness of the stimulus (see replies to Block 2007). There are no direct scientific grounds on which to choose between these interpretations, because there is no direct way to assess whether marker \(M\) has anything to do with phenomenal consciousness.
One response to these discussions is to advocate discourse eliminativism about phenomenal consciousness. This is based on accepting that there is no direct way to probe phenomenal consciousness independently of cognitive access, and that there are no straightforward empirical ways of testing the claim that phenomenal consciousness can be present independently of cognitive access. In this case, the only aspect of consciousness that can definitely be probed scientifically is cognitive access, that is, access consciousness. In terms of scientific practice, the safest methodological route is to drop talk of phenomenal consciousness. Something like this position appears to be taken by a number of consciousness researchers (possibly including Dehaene).
This position is compatible with a range of claims about the ontology of phenomenal consciousness. One might say that phenomenal consciousness can (possibly or probably) exist without cognitive access, or be agnostic about this possibility. Alternatively, one might argue, with Cohen & Dennett (2011), that if a phenomenally conscious state is not accessible to scientific enquiry or to the subject having it (e.g. via some kind of report), then it is (evolutionarily, cognitively) implausible to call it a state of consciousness at all. In this case, if phenomenal consciousness exists, it always co-occurs with cognitive access.
5.3 Eliminativism via identity crisis
The argument for discourse eliminativism about phenomenal consciousness outlined above is based on a problem with accessing the phenomenon in question. Another kind of discourse eliminativism is based on the problem of identifying the phenomenon in question. For the sake of argument, ignore the problem of access raised in the previous section. Assume that phenomenal consciousness always co-occurs with access consciousness (perhaps for the reasons suggested by Cohen and Dennett), so that we can (for the minute) work just with the term ‘consciousness’ which will pick out both. Even with the problem of access out of the way, it is still questionable whether the concept of consciousness picks out a clear category of phenomena that is scientifically useful. If it does not, this provides a new motivation for discourse eliminativism about consciousness, and (by assumption) discourse eliminativism about phenomenal consciousness.
It was suggested above that there is a reasonably broad consensus that assessing the presence or absence of consciousness has something to do with reportability. Reportability can be realised in a number of ways, however, some of these are incompatible with each other (see Irvine 2013 for review). One ‘objective measure’ (taken from psychophysics) of consciousness relies on forced-choice tasks: for example, subjects are shown a masked stimulus for a short period of time and are ‘forced’ to choose between two response options (stimulus present/absent, stimulus was a square/circle). On the basis of their response, the subjects’ underlying ‘sensitivity’ to the stimuli is calculated. The resulting objective measure of consciousness is highly stable and not subject to biases, but it is liberal, and often attributes consciousness of stimuli to subjects who explicitly deny having any. As a result, it is sometimes criticised as merely being a measure of sensory information processing and not of consciousness (e.g. Lau 2008). Despite being acknowledged as problematic, objective measures tend to be used in studies of consciousness because of their desirable properties as scientific measures (they are stable, bias-free).
In contrast, ‘subjective measures’ of consciousness use free reports or similar responses generated by experimental participants. The experimental methodology may be based around emphasising careful use of introspection, assessing subject’s confidence in their reports (sometimes using wagering), or just recording simple, untutored responses. Subjective measures get closer to what the subjects themselves acknowledge about their conscious experience. However, the precise ways that subjective measures are generated can have a significant impact on whether consciousness is deemed to be present or absent (or somewhere in between) (Sandberg et al. 2010; Timmermans & Cleeremans 2015). As scientific measures, they are highly unstable and subject to bias. They also regularly conflict with objective measures (except under artificial training conditions), and they are generally thought to be conservative (they normally do not capture all instances of conscious experience).
These difficulties reappear in debates about the neural correlate(s) or mechanism(s) of consciousness. Behavioural measures of consciousness are key in identifying these correlates and mechanisms. Roughly speaking, one chooses a behavioural measure; identifies the neural activity that occurs when the measure says that consciousness is present; and treats this as ‘the’ correlate or mechanism of consciousness. However, using different behavioural measures (unsurprisingly) leads to the identification of different neural correlates. The latter span all the way from ‘early’ neural activity for some liberal measures of consciousness (which may capture early sensory processing), to ‘late’ and attention-based neural activity for conservative measures (which may capture later cognitive uptake of the conscious experience) (see Irvine 2013). Without agreement about what counts as the ‘right’ behavioural measure of consciousness, there can be no agreement about what the neural correlates and mechanisms of consciousness are.
The plethora of measures and mechanisms of consciousness is not necessarily problematic in itself, but Irvine (2012) argues that there is no methodologically viable way of resolving disagreements between them when they conflict. Each measure has its pros and cons, but none is both scientifically adequate (i.e., fairly stable over repeated measures and bias-free) and fits with pre-theoretic commitments about consciousness. To choose one measure would be to (operationally) define consciousness by fiat, which would undermine the motivations for engaging in ‘real’ consciousness science in the first place. Furthermore, the mechanisms that correlate with these varied measures do not form a well-demarcated scientific kind, or even a well-demarcated group of kinds. They have no more in common than any arbitrary group of mechanisms within perception and cognition. They range across sensory processing, attention, decision making, report, and meta-cognition.
This suggests a reason for eliminating talk of consciousness from science. There are a wide range of incompatible things that ‘consciousness’ could pick out, and no methodologically acceptable way of deciding between them. If a scientific concept is surrounded by such problems, then (if they are bad enough) that is motivation for eliminating the concept. These methodological problems are compounded by pragmatic ones. Given that it is unclear what ‘consciousness’ refers to, talk of consciousness generates unproductive debates and miscommunication; it blocks the generation of useful predictions and generalisations; and it promotes misapplications of research methodologies and heuristics. That is, there are negative practical consequences from continued use of the concept ‘consciousness’ in science.
There is also a better alternative. This alternative demands that researchers use terms that clearly demarcate the phenomena under study, potentially by referring to how they are experimentally operationalised. This could be done by splitting up phenomena previously grouped under the single heading ‘consciousness’ by how they are measured (e.g. forced-choice tasks, confidence ratings, or free report). Using these more specific terms avoids the problems above. By precisely specifying what the phenomena are and how they are measured, there is no ambiguity about which phenomenon is picked out. This would also make it possible to identify the neural mechanism that generates the phenomenon, make robust predictions and generalisations about the phenomenon, and avoid miscommunication.
As before, discourse eliminativism is not tied to entity eliminativism (for example, Irvine’s (2012) position does not entail entity eliminativism of any sort). Discourse eliminativism is about which representations, concepts, methods, and practices are appropriate and useful to science. Whatever consciousness (access or phenomenal) is may still be out there, even if the concept of ‘consciousness’ is not a useful one for science.
In this chapter, we have reviewed a variety of arguments for entity and discourse eliminativism. Entity eliminativists deny the existence of phenomenal consciousness; discourse eliminativists deny the utility of talking about phenomenal (and perhaps access) consciousness in science.
Entity eliminativism can be defended in a number of ways. A standard method is to describe the entity in question, then show that nothing satisfies that definition (4.1). This can be expanded to the method of using examples to fix the subject matter (4.2). A third approach, taken by illusionists (4.3), is to use a loose definition of the relevant entity/property, but argue that whatever this refers to, it is theoretically and metaphysically simpler and more productive to assume that the entity does not exist. A problem that faces entity eliminativists of all types is the ‘illusion problem’, a mirror image of the hard problem faced by realists, which requires an eliminativist to explain how something non-phenomenal can give rise to something that seems phenomenal.
Discourse eliminativism concerns the net benefit to science of various ways of talking, thinking, and acting. Classic scientific behaviourism focused on what could be measured in a public and ‘observable’ way, eradicating talk of mental states (5.1). More recent scientific work on consciousness has tended to move away from discussion of phenomenal consciousness on the basis that it is not clear whether scientific methodology can probe it independently of the cognitive abilities associated with access consciousness (5.2). An argument can also be made that the general concept of consciousness should be eliminated from scientific talk given the problems in clearly demarcating the phenomenon in question (5.3). Eliminating discourse about phenomenal consciousness from science might seem to remove a key concept in explaining human behaviour. However, this is not necessarily the case: specific reports and judgements about phenomenal consciousness can still function in explanations, and as explanatory targets in their own right.
We would like to thank Uriah Kriegel and Tim Bayne for helpful comments on an earlier draft of this chapter.
- 1 Introduction
- 2 Eliminativism
- 3 Consciousness
- 4 Entity eliminativism about consciousness
- 5 Discourse eliminativism about consciousness
- 6 Conclusion