I. Introduction
Multiple Realization (MR) is the claim that functional states (including mental states) can
be brought about by physically distinct realizers. For the last few decades the argument has been
though to refute reductive physicalism, the thesis that mental states and brain states are identical.
If a mental state can be brought about by two physical states that are not themselves identical,
then the mental state cannot be identified or reduced to those physical states. If this argument is
sound, reductive physicalism (at least in its simpler formulations) is false. 1
MR was first advanced as an empirical hypothesis: In the actual world, we can observe
organisms with very different neuro-physical constitutions that have the very same mental state.2
Moreover, MR is considered by many to be a ubiquitous phenomenon in the natural world,
occurring intra-individual and intra-species, as well as between members of distantly related
biological taxa.3 In this paper I call these claims into question.
In §2 I give a precise formulation of MR. I argue that MR is actually the conjunction of
two separate claims, both of which must be empirically demonstrated. I arrive at the following
formulation.
(MR) Two organisms T1 and T2 multiply realize the same mental state F iff
(MRi) the same Ramsey sentence is true of both T1 and T2 when
describing F, and
(MRii) T1 realizes F in virtue of instantiating causal powers P, T2
realizes F in virtue of instantiating causal powers Q, and P ≠Q
with respect to the relevance criterion
I argue further that the standards for each claim are higher than we might expect. While most of
the debate over MR has focused on (MRii), I contend that both claims are equally important, and
The English word ‘octopus’ derives from a Greek 3 rd declension masculine noun, not a Latin 2nd declension
masculine noun. The correct plural is ‘octopodes’, not ‘octopi’, and not ‘octopuses’ (for the same reason we do not
say ‘datums’, ‘stimuluses’, ‘phenomenons’ or ‘thesises’).
2
The classic expositions are in Putnam (1967), Block & Fodor (1972), and Fodor (1974)
3
In addition to the sources cited in n. 1, see
1
that the standards for each claim are higher than we might expect. (MRi) makes explicit that the
very same mental state must be realized in both cases, which requires causal homogeneity at the
functional level. (MRii) calls for functionally relevant heterogeneity, which I argue is not entailed
by compositional differences. In the next two sections I argue that neither (MR i) nor (MRii) is
satisfied with respect to pain in humans and octopodes, the example Putnam used to begin the
debate over MR. In §3 I show that the pain states of humans are functionally distinct from the
pain states of octopodes, and are therefore different mental states. I also advocate describing
mental states at a specific level of abstraction, which does not beg the question against either
side. In §4 I claim that human pain and octopus pain are realized by homologous, and a fortiori
causally similar, neural systems, and show that various objections to this claim misunderstand
MR and undercut the arguments for it. In §5 I show how the empirical evidence against MR can
be applied to other mental states between these two organisms, and to mental states between
other organisms.
Throughout this paper I will be focusing only on the hypothesis that mental states are
multiply realized across biological taxa.
This is but one of several empirically-minded
arguments for MR, but there are several reasons for focusing here.
For one, Shapiro
(unpublished) and Polger (2004) have adequately addressed the shortcomings of the neural
plasticity argument, which will cover intra-individual and intra-species cases.
Similarly,
Zangwill (1992) has shown that there is no real evidence of extraterrestrials or artificial
intelligence with the same mental states we have, and the mere possibility of such cases does not
count as empirical support. So if we are concerned with finding empirical support for MR,
diverse organisms seem like the best place to look.
2. Realization and Multiple Realization
MR is an inter-level relationship between realizer physical systems and realized
functional states, as depicted in Figure 1 below.
(Figure 1)
As the diagram shows, MR requires different realizations of the very same realized state. This
means that MR is actually the conjunction of two separate claims4: (i) at least two organisms
realize the very same mental state, and (ii) those organisms realize that mental state via distinct
causally relevant powers. What I call the empirical argument for MR is simply that both these
claims have empirical support. The two claims for MR each require some fleshing out. We’ll
start with (ii).
Two camps have dominated recent discussion of realization: the ‘Flat’ view ascribed to
Kim, Shoemaker, and Putnam and later defended by Polger and Shapiro, and the ‘Dimensioned’
view attributed to Block and Fodor and defended by Aizawa and Gillett.5 Despite their
disagreements, both sides accept the claim that only functionally relevant differences, i.e.,
differences that directly pertain to how a certain functional role is filled, contribute to MR.
Gillett calls this claim the relevance criterion.6 The disagreement centers on how broadly we
should understand the relevance criterion. Shapiro (2000, 2004) argues that many characteristics
of an object do not influence how that object realizes its functional state, and are therefore
functionally irrelevant. His example is the color of a corkscrew: red corkscrews remove corks the
4
Couch (2004) makes a similar point.
See Aizawa & Gillett (2009), Gillett (2002, 2003), Polger (2008, 2009), Shapiro (2000, 2008), Polger & Shapiro
(2009)
6
Gillett outlines the relevance criterion as follows: “Instances TI, instantiated in sl, and T2, instantiated in s2, are
distinct realizations of a property T only if there is a difference between the properties/relations whose contribution
of powers are those in virtue of which sl and s2 have the powers individuative of T” (2003:596).
5
same way that yellow corkscrews do. Shapiro claims that many compositional differences are
also irrelevant: steel corkscrews remove corks the same way aluminum ones do. Gillett (2003)
disagrees, arguing that the composition is relevant, because aluminum molecules and steel
molecules are chemically bonded in different ways, and these bonds directly contribute to the
metals’ respective rigidity.
Resolving this particular disagreement will bring out important considerations for MR
more generally. Aluminum is a highly reactive element, which forms electron bonds that are
difficult to break. Steel is an iron alloy, which is made by combining iron with carbon to form
bonds that are difficult to break. These bonds are, in both cases, formed by the interactions
between the electrons of each element. Though the differences are causally relevant, they are not
causally distinct; both steel and aluminum are rigid in the same way, and that rigidity allows
them to open corks the same way. But this does not entail that compositional differences are
never relevant. If steel were rigid in virtue of, say, its molecules being densely packed and
unable to move, while aluminum were rigid in virtue of its molecules forming strong electron
bonds, then their respective rigidity would contribute to MR for corkscrews. This shows us that
compositional differences are relevant to realization only when those differences contribute
causal powers in distinct ways.7
One may be tempted to respond here that steel and aluminum have to be causally distinct,
simply in virtue of being compositionally distinct. But there are problems to such an approach.
For one, this stipulation seems to beg the question in favor of MR. More importantly, the
7
This entails that, pace Polger (2008) and Shapiro (2000,2008), MR can be transitive. But the transitivity occurs
only in special prescribed circumstances, i.e., when specific lower-level causal powers are instantiated via different
mechanisms, which powers themselves contribute to the realization of some higher-level functional state. In other
words, lower-level differences are “screened off” only when those lower-level differences don’t result in higherlevel differences. Polger and Shapiro are correct, however, that MR is not always transitive; in many cases, lowerlevel differences do not result in causally-relevant higher-level differences (as with aluminum and steel corkscrews).
consequences of the view that all compositional differences entail functional differences are
untenable. Taking this suggestion seriously would commit us to holding that different samples of
water are realized in distinct ways, because they are compositionally distinct at the level of
quarks or strings, a la Funkhouser (2007).
This entails that water (and many other natural
kinds) are multiply realized. And if MR prevents the identification of a kind with its realizer
base, as MR proponents suggest, then water cannot be identified with H2O (because H2O cannot
be identified with a certain combination of hydrogen and oxygen, which cannot themselves be
identified with clusters of protons, neutrons, and electrons, and so on).8 This strikes me as a
reductio of the position. A further point concerns realizer-level and realized-level laws. If we
allow that multiply realized objects cannot be adequately described by the laws of their realizers
(Fodor 1974), then the laws of subatomic particles would not adequately describe water. But
microphysical laws do adequately describe water. Therefore either a) multiply realized objects
can be adequately described by the laws of their realizers (in which case MR is not a threat to
reductionism), or b) water is not multiply realized despite its being composed in different ways.
Proponents of MR are advised to choose the latter conclusion.
With these considerations in mind, we can formulate (ii) more precisely as follows.
(MRii) Two organisms T1 and T2 are distinct realizers of the mental state F only if
T1 realizes F in virtue of instantiating causal powers P, T2 realizes F in virtue of
instantiating causal powers Q, and P ≠Q with respect to the relevance criterion
Now on to claim (i). Before addressing it, I should note that far less attention has been paid to (i)
than to (ii) in most discussion of MR. One of the theses of this paper is that MR comprises two
distinct claims, each of which has higher standards of proof than is commonly regarded,
especially with respect to claims at the realized level. By precisely formulated exactly what (i)
8
It is no longer taken for granted that MR prevents reduction. See Bickle, 1998, 2003; Clapp, 2001; Keeley, 2001;
Kim, 1992; Kim, forthcoming, Polger, 2004; Richardson, 2008; Rosenberg, 2001; Shapiro, 2000, 2004; Sober,
1999b
states, we can see that MR advocates must do more than simply apply the same predicate to two
mental states.9 Or so I shall argue.
In what follows, I will use a functional analysis of mental states
10
There are several
reasons for doing so. Many proponents of MR are also functionalists, and indeed the earliest
expositions of MR were used to argue for functionalism in place of identity theory. 11 Second,
there is reason to think that realization is a special kind of metaphysical relationship which holds
between functional states and physical systems. If realization is not restricted, it is difficult to
distinguish realization from other metaphysical relationships like composition, supervenience,
and the like.12 It seems intuitive enough that realization is special – just compare the relationship
between the software and hardware of a computer to a heap of sand and its grains or the moral
properties of a world and its physical properties. Restricting realization to functional states helps
make sense of this special kind of relationship. Lastly, both sides of the debate over MR agree
that properties should be delineated causally, so that two properties with different causal profiles
are different properties.13 Since functional states are, in essence, causal states, talking about
mental states as functional states makes it easy to focus on the causal profiles of these states.
9
Cf. Heil 1999, 2003. Heil argues that we can make sense of MR if we think of second-order realized properties as
causally similar rather than identical. But as Shapiro (2008) argues, this way explicating MR does not do it justice.
If there is no single functional role or property to be realized, then it can’t be multiply realized. MR requires
realized and realizer states to be “same-different”, not “(slightly) different – different.” Even so, Heil’s distinction
between predicates and properties in this debate is significant. He is certainly right that language does not determine
ontology.
10
Though I will talk about mental states as functional states, my analysis does not require one to assume that
functionalism is true. Functionalism has its own problems (See Block 1978, Putnam 1988, among others). But a
functional analysis of mental states does not entail functionalism full-blown. If, for instance, mental states have
unique neural realizers, then identity theory might steel be true even if mental states can be described as functional
states.
11
Interestingly enough, Block & Fodor (1972) also discuss problems with functionalism, as does Block (1978).
Similarly, though Putnam (1967) used MR to argue for functionalism, Putnam (1988) repudiated the theory.
12
Cf. Polger (2004) and Polger & Shapiro (2009)
13
On the anti-MR side, see Polger (2004, 2008, 2009), Polger & Shapiro (2008), and Shapiro (2000, 2008). For the
pro-MR side, see Aizawa & Gillett (2009, unpublished), Gillett (2002, 2003).
A mental state, according to Lewis (1972) and others, can be described by a Ramsey
sentence in the following form:
∃x(x is caused by stimuli S1-n, & x tends to cause behaviors B1-m & x
tends to cause other mental states M1-k & ….)14
For our purposes, x refers to the realized mental state under investigation, while the S, B, and M
variables describe disjunctions of stimuli, behavior, and other mental states, respectively. Two
organisms realize the same mental state only if the same Ramsey sentence is true of both them.
Hence we can formulate (i) more precisely as follows:
(MRi) Two organisms T1 and T2 realize the same mental state F only if the same
Ramsey sentence is true of both T1 and T2 when describing F.
The same Ramsey sentence will be true of both T1 and T2 when the S, B, and M variables are
filled by the same stimuli, behavior, and mental states, more-or-less. I say “more-or-less”
because, technically speaking, the failure of any single part of the sentence to accurately describe
the world entails that the whole sentence is false. This would result in an incredibly low
probability of there ever being two instances of the same mental state. But as long as the
sentence describing each mental state has the S, B, and M variables filled in such that the two
sentences are (at least) more similar than different, we need not infer that the two mental states
are different.15 To use Lewis’ term, near realization of the Ramsey sentence is good enough.
If the same Ramsey sentence is true of two mental states, then they will generate similar
behavioral and mental effects from similar stimuli. But if we find a high degree of causal
heterogeneity (e.g., sets of stimuli or behaviors with more different members than members in
Strictly speaking this picture is overly simplistic, because the sentence I’ve given describes only one mental state,
and the Ramsey sentence method is typically used to define all mental terms at once. However, since we are talking
about only a few different mental states, using a Ramsey sentence to describe only one state at a time will help focus
and simplify the conversation. It may help to think of the specific Ramsey sentences I discuss below as parts of a
conjuncts of a larger sentence which contains all mental terms.
15
One way to think about this relatively lenient standard is the type-token distinction. If the Ramsey sentence
describes mental state types, then the sentence should sustain rather significant intra-type variation when applied to
token mental states.
14
common, stimuli that produce opposite reactions in different organisms, etc), then we have good
reason to think that the mental states in question are different functional states. And if we don’t
have the same functional state in both cases, we don’t have MR.
With both (i) and (ii) more clearly articulated, we can give more precise conditions for
MR.
(MR) Two organisms T1 and T2 multiply realize the same mental state F iff
(MRi) the same Ramsey sentence is true of both T1 and T2 when
describing F, and
(MRii) T1 realizes F in virtue of instantiating causal powers P, T2
realizes F in virtue of instantiating causal powers Q, and P ≠Q
with respect to the relevance criterion
(MRi) and (MRii) are each necessary for MR. They are jointly sufficient for MR.
In the next two sections I will argue that neither (MRi) nor (MRii) are empirically
supported. But before doing so, it will be helpful to briefly discuss what sorts of evidence we
will look for. When testing (MRi), we will look for causal homogeneity between two mental
states.
Of course, mental states are not themselves directly observable. But their causes and
effects are. By observing the stimuli and behavior that serve as the causes and effects of mental
states, we can make an inductive argument about those mental states. If we observe the same
stimuli causing the same behavior in two different organisms, we have reason to believe these
organisms share the same mental state. But if we see too much causal heterogeneity between the
stimuli and behavior of different organisms, we can infer that they do not share the same mental
state. When testing (MRii), we will look for causal heterogeneity in the neurology of the two
organisms. If the causal powers by which the brains of humans and octopodes realize pain states
are different, then we have evidence for MR. If these organisms realize pain states via the same
causal powers, we have evidence against MR.
3. Not of One Mind: Mental States in Humans and Octopodes
I’ve argued so far that MR is actually the conjunction of two claims, each of which
requires empirical support. In this section I show that observations of octopus behavior suggest
that they do not realize the same mental states that humans realize. Since there is not one mental
state realized between the two organisms, there can be no MR.
The Ramsey sentence which describes the mental states of octopodes takes the following
form:
∃x(x is caused by stimuli S1-n, & x tends to cause behaviors B1-m & x
tends to cause other mental states M1-k & ….)
I argue that the S, B, and M variables that make the sentences true for each organism are simply
too diverse.16 I should emphasize from the outset that this is not simply a matter of finding one
or two differences. Rather, the causal roles of human and octopus pain are radically dissimilar,
with far more differences than similarities. This diversity holds for both qualitative states and
propositional attitudes. Though the body of literature on octopus psychology is vast, I will focus
here only on pain.17
Hanlon & Messenger (1996) report that
[D]eliberately inflicted pain, in the form of a small electric shock (4-10 V, ac) has
been consistently and successfully employed to shape octopuses’ responses to
objects during discrimination training. In these circumstances octopuses react as a
human does when stung by a nettle.18
This suggests that octopodes and humans do share some behavioral responses, which gives
prima facie support to the claim that humans and octopodes occupy the very same ‘pain’ role.
But closer inspection reveals that this is an oversimplification. Humans and octopodes are
16
Since this is a discussion of empirical arguments for MR, we must restrict ourselves
I should also note that I will be ascribing mental states to octopodes rather liberally. Our awareness of the
sophistication of animal cognition is steadily increasing, so the idea of, say, invertebrates with complex mental states
is not as controversial as it once was. In any case, if we can’t ascribe the same (complex) mental states to octopodes
as we do to humans, MR never gets off the ground.
18
Hanlon & Messenger 1996, p. 21
17
significantly different with regard to both the causal stimuli of their pain states and the
behavioral effects.
Clearly, pain-inducing stimuli for humans and octopodes differ significantly. Humans
feel pain from a whole range of stimuli that do not affect octopodes. There are, of course, the
psychological and emotional sources of pain, like betrayal, injustice, envy, or unrequited love.
But there are also many stimuli that both organisms are capable of processing, but which would
only serve as a pain stimulus for one organism. For instance, many octopodes live in cold water,
and are most active at night.19 These conditions would certainly be a stimulus with a high
probability of inducing a pain-state transition for humans (at least until hypothermia set in), but it
is not a pain-state input for octopodes. Similarly, eating large quantities of uncooked fish and
mollusks would likely be an input to cause humans to occupy a pain state, but this is not the case
for octopodes.
We could tell similar stories about stimuli that would cause octopodes to transition to a
pain state, but not humans (e.g., being out of the water for several hours). The same causal
differences occur in behavior. When an octopus occupies the ‘pain’ role, its behavioral outputs
might include swimming away, retracting its tentacles, or changing its color pattern.20 When a
human occupies the ‘pain’ role, her outputs might include uttering “Ouch”, crying, or imitating a
super-Spartan.21 Obviously, these two classes of behaviors have far more different members than
members in common. As before, the S, B, and M variables that make the Ramsey sentence for
pain true of octopodes make the same sentence false for humans. Hence we can infer that there
are two functional states, not one.
19
Nixon & Young (2003)
Smith (199)
21
See Putnam (1967)
20
This is not to say that humans and octopodes have no stimuli or behavior in common.
There are some areas of overlap between human and octopus psychology: for instance, both
sometimes flinch when poked by something sharp. To be sure, the simple fact that a stimulus
elicits one behavior in me and another in you is not enough to show we do not realize the same
mental state. But a few similarities are not enough to constitute the very same functional state.
Given the multitude of causal differences between the mental states of humans and octopodes,
we must conclude that their mental states are functionally distinct. Since there are two mental
states realized, not one, we do not have a case of MR.
One plausible response to my argument is to attempt to use a coarser grain to describe
mental states. As many authors have noted, the level of abstraction at which one describes
possible cases of MR makes a substantial difference to the analysis of such cases. 22 It may be
that I’ve just chosen the wrong level.
We might fill in the S variable of the Ramsey sentence
with “a noxious stimulus”, the B variable with “avoidance behavior”, and the M variables with
“anxiety, distress, etc”, and come up with a sentence that is true for the pain states of both
humans and octopodes. This method might generate true Ramsey sentences for many of the
mental states of both organisms. But there are several problems with this approach.
For one, there is always the worry that one is looking for a level of abstraction that makes
the results come out right. If we assume that mental states are multiply realized, we would look
for a grain of description that bears out this assumption. But clearly this approach begs the
question when we are trying to determine whether or not mental states are multiply realized in
the first place. Furthermore, the method I’ve used to describe mental states is sympathetic to
how proponents of MR describe it. In fact, by insisting only on near realization, I’ve presented a
22
Bechtel and McCauley, 1999; Bechtel and Mundale, 1999; Couch, forthcoming; Polger, 2004; Shapiro, 2008;
Sober, 1999; Sullivan, 2008
more charitable conception of functional similarity than may be needed.23
If principled,
functionalist-inspired reasons for determining the level of specificity at which to describe
functional states are not fair, I do not know what is. Moreover, the grain of description I’ve used
is also the grain used by scientists who study octopus psychology.
Yet even if one could motivate pursuing another level of abstraction, there are still
problems. Suppose I poke my Roomba with a pointed stick and it moves to the other side of the
room. It has exhibited avoidance behavior in response to a noxious stimulus; does that mean my
Roomba realized a pain state, and moreover the same pain state I realize?24 This level of
description commits us to being too liberal with our mental talk, ascribing mental states to beings
that lack them (Block 1978).25 In order to avoid this problem, we need to be more specific in our
descriptions of functional states. I contend that in doing so we will be unable to avoid arriving at
different functional states for humans and octopodes. At any but the vaguest level of description,
it is clear that the stimuli which cause human mental states are different from the causes of
octopus mental states, as are the behaviors caused by that state.
This precludes us from
describing the mental states of these organisms with the same Ramsey sentence, without which
we cannot satisfactorily demonstrate MR.
Another worry about using a coarser grain to describe functional states is that the
approach cuts both ways. If we can use a coarse grain to describe mental states, why can we not
use an equally coarse grain to describe brain states? After all, “Whatever neural activity occurs
when the organism encounters a noxious stimulus” is at the same level of abstraction as
23
Strictly speaking, a functionalist analysis of mental states seems committed to saying that any difference between
entails a difference in functional states. This is because functional states are defined holistically, i.e., with reference
to the whole machine table or Ramsey sentence of which they are a part (see, for instance, Kim (2006), p. 137-142).
But this suggests that no two mental states are in fact the same functional state (or at least that the probability of this
happening is very low). If no two mental states are functionally identical, then MR never gets off the ground.
24
Roombas are small household robots that clean floors autonomously. I don’t really have one (which, incidentally,
causes me to realize a pain state), and if I did I would not poke it with a pointed stick.
25
See Block (1978)
“Whatever is going on in the organism’s head when it exhibits avoidance behavior in response to
a noxious stimulus.” If we can describe the mental state at a high level of abstraction, why can
we not describe the brain state in the same manner? The result, of course, would be mental states
are not multiply realized after all, because the very broadly understood functional state turns out
to be realized by a single, broadly understood neurological state. Trying to salvage MR by using
a broader level of abstraction might help at the functional level, but it will enervate the case for
MR at the realizer level.
I’ve attempted to show that making a case for MR is more difficult than it might appear,
especially at the realized level. The empirical argument for MR must provide evidence to satisfy
(MRi). But behavioral observations show us that (MRi) is not satisfied with respect to humans
and octopodes. The Ramsey sentences which are true of octopus pain are false when applied to
human pain, because the stimuli, behavior, and other mental states associated with the mental
state for each organism are too diverse. The same predicates may be applied to the mental states
of both organisms, but this is not sufficient to show that the organisms realize the very same
mental states.26 When the pain states of humans and octopodes are examined with sufficient care
and precision, it is clear that they are not the same state. Without evidence that the same mental
state is realized in both cases, (MRi) is not satisfied, and hence the empirical argument for MR is
unsubstantiated.
4. Man Pain, Mollusc Pain
MR requires both that the same functional state be realized by two physical systems and
that the physical systems realize that functional state in causally distinct ways. I formulated this
claim as follows.
26
Cf. Heil 1999, 2003, Shapiro 2008
(MRii) Two organisms T1 and T2 are distinct realizers of the mental state F only if
T1 realizes F in virtue of instantiating causal powers P, T2 realizes F in virtue of
instantiating causal powers Q, and P ≠Q with respect to the relevance criterion.
I this section I argue that this claim is not empirically supported. As before, I’ll focus on pain.
I argued in §II that not all differences in the realizers of a functional state result in MR.
Shapiro (2000, 2004) argues that two corkscrews that differ in color are not ipso facto multiply
realized, because color does not contribute to the exercise of a corkscrew’s functional state.
Similarly, aluminum and steel corkscrews do not count as cases of MR simply in virtue of their
differing molecular composition, because the property through which the corkscrews open
bottles (rigidity) is instantiated in the same way in both cases (chemical bonding between
electrons). In short, not all compositional differences satisfy the relevance criterion because
compositional differences do not entail functionally relevant causal differences.
When examining the neural realizers of pain in humans and octopodes, we must keep this
lesson in mind. Not all differences in the brains of these two organisms contribute to MR. For
instance, the octopus brain surrounds the esophagus, while the human brain does not. But the
property “surrounding the esophagus” has nothing to do with realizing the functional state
“pain”. It does not meet the relevance criterion, and therefore does not count toward MR. What
MR requires is for the brains of humans and octopodes to realize pain in causally distinct ways.
It turns out that just the opposite is true. First we’ll examine the realizers of pain states in
octopodes. Not only are the realizers of pain for humans and octopodes causally similar; they’re
the very same physical system. The data gathered by Stefano et al (2002) speaks for itself:
Past studies on neuronal mechanisms in insects and mollusks have revealed
remarkable structural, functional, and biochemical parallelisms with those in
vertebrates…. Numerous studies, in a variety of invertebrate species, have
identified neuropeptides in parts of the neuroendocrine and nervous system
apparatus, which is analogous to the hypothalamic-hypophysical system of
vertebrates. The list includes substances closely resembling the following
vertebrate
neuropeptides
and
hormones:
oxycotin,
vasopressin,
adrenocorticotropin (ACTH), α melanocyts stimulating hormose (MSH),
somatosostatin, substance P, neurotensin, hypothalamic growth hormone releasing
favor, insulin, glucagon, gastrin/cholecystokinin, vasoactive intestinal peptide
(VIP), pancreatic polypeptide (PP), secretin, luteinizing hormone releasing favor
(LHRF), and several endogenous opioids.27
Even the scientifically uninitiated will recognize some of the substances common to the nervous
systems of both humans and octopodes, like oxycotin and insulin. But these similarities are not
limited only to the chemicals used by invertebrate and vertebrates brains; they also have the
same structure and function. Stefano and his colleagues conclude
these commonalities of signal molecules, activities, and regulatory mechanisms,
must be viewed as demonstrating a continuity of information during the
development of various response systems throughout evolution, rather than the
appearance of “chance” similarities. One is therefore only left to conclude that
the invertebrate perturbation-stress systems developed many of the strategies for
mammalian stress phenomena.28
In other words, the conclusion is not simply that the organisms converged on similar physical
solutions, but rather that the reactions to pain between distantly related taxa are in fact instances
of the same process, modified by selection pressures over time but nevertheless essentially the
same.
It is important to remember here that the level of similarity we are looking for here is
functionally relevant causal similarity, not identity. The pain response systems of humans and
octopodes are not identical, nor are their memory consolidation systems. But they do not need to
be. Aizawa claims that “It is not enough that ordinary language, or even ordinary scientific
language, says that two structures are the same. Even if one can say that two structures are the
same, it does not follow that they are physically or chemically the same.”29 But this point goes
both ways. Saying that two structures are different does not entail that they are evidence of
27
Stefano et al (2002), p. 86, emphasis added
Ibid, p. 92, emphasis added
29
Aizawa (2007), p. 82
28
MR.30 For MR we need differences that satisfy the relevance criterion. We need evidence that
the same function is brought about in causally distinct ways.
In several works Bickle has argued that memory-consolidation is uniquely realized across
the animal kingdom, in virtue of the fact that distantly related organisms, including mice, fruit
flies, snails, all use the same physio-chemical system for memory.31 Aizawa counters that the
memory-consolidation systems used by these organisms are not compositionally identical, and
therefore provide evidence for MR instead. For example, the cAMP response element binding
(CREB) proteins for fruit flies and mice share only 95% of the same amino acids.32 Even if the
same basic chemical are used by the same basic structures in the same basic way, they are not
completely similar, and therefore still count toward MR.
We can construct an analogous
argument with regard to the neurological realizers of pain in humans and octopodes. Now, this
argument is partly based on the assumption that any compositional difference is sufficient for
MR, which I have argued is false.
But even if we only consider functionally relevant
differences, the argument does not go through. The differences between the proteins in the
brains of humans and octopodes is analogous to the differences between the molecules of steel
and aluminum. Even though they are compositionally distinct, they are not functionally distinct.
Both steel and aluminum instantiate the property of rigidity in virtue of the bonding of their
constituent electrons. Similarly, the amino acid sequences in octopus and human brains transmit
information through nerves via chemical interactions. Though there are some differences in the
amino acid sequences in the various proteins shared between humans and octopodes, the amino
30
Cf. Kim (1992)
Bickle (2003, 2006, 2008). It turns out that octopodes may be added to this list. I’ll return to this point in the next
section.
32
Aizawa (2007), p. 71
31
acids still function the same way, and therefore contribute to the realization of a functional state
in the same way.
But even if we insist that these small differences between the brains of humans and
octopodes must somehow be causally relevant, they still do not make the case for MR. As many
have pointed out, if two realized states share the same basic set of realizer properties, then it is
unclear why we cannot reduce the realized property to that set of realizer properties. 33 For
instance, if pain in octopodes is realized by properties {A, B, C, D, E} and pain in humans is
realized by properties {A, B, C, F, G} then it seems perfectly reasonable to reduce pain to {A, B,
C, X}, where X is a finite and well-behaved disjunction of similar but not identical properties.
Moreover, given that the mental states are not identical in humans and octopodes, we arrive at a
situation where the functional similarities between humans and octopodes are the result of their
sharing A, B, and C, and the differences are the result of possessing different members of the
disjunct X. As Shapiro (2008) notes, this recipe – “Sort of the same but sort of different” – does
not work for MR. And lastly, as I argued in the last section, describing realizer states at a high
level of specificity invites us to do the same to realized states. If, as Aizawa suggests, 5%
difference is enough to count as distinct at the realizer level, then by parity of reasoning 5%
difference should count as distinct at the functional level. The result is that (MRii) is satisfied at
the expense of (MRi).
Despite the physiological dissimilarities between humans and octopodes, the brains of the
two organisms are not different in the way MR requires. Rather, it appears that the painresponse systems of humans and octopodes are in fact the very same physical systems. These
systems are not, of course, identical, but they are functionally similar in a way that precludes
MR. Even the small differences between them are not enough to support the case for MR,
33
See, for example, Kim (1992), Heil (2003)
because the remaining similarities permit us to reduce the mental states in question to the
conjunction of their shared properties and the members of an orderly disjunct of other similar
properties, and because level of distinction at the realizer level undercuts the case for MR at the
realized level. Hence (MRii) is not satisfied, and the empirical argument for MR is unsupported.
V. Discussion
I argued above that MR is the conjunction of two separate claims, which I formulated as
follows:
(MR) Two organisms T1 and T2 multiply realize the same mental state F iff
(MRi) the same Ramsey sentence is true of both T1 and T2 when
describing F, and
(MRii) T1 realizes F in virtue of instantiating causal powers P, T2
realizes F in virtue of instantiating causal powers Q, and P ≠Q
with respect to the relevance criterion
The empirical argument for MR claims that both (MRi) and (MRii) are empirically supported. I
have attempted to show that this is not the case. Behavioral data show us that the mental states
of humans and octopodes are too causally diverse to be counted as the same mental state, as
evidenced by the fact that the same Ramsey sentence cannot accurately describe both organisms.
Neurological evidence suggests that humans and octopodes realize a pain state by means of the
same neuro-physical system and therefore the same causally relevant properties. Hence the
evidence tells against both claims of MR, which leads us to conclude that the empirical argument
for MR fails.
One might justifiably wonder what the significance of this conclusion is. After all, one
bit of empirical evidence is not particularly compelling on its own. Even if pain is not multiply
realized in humans and octopodes, it might be that other mental states are. Or, we might think
that pain is multiply realized in other creatures. Alternatively, we might think that pain is such a
basic mental state that it would find common expression throughout the animal kingdom, but that
other, more complex mental states might be more sporadic and therefore more likely to be
multiply realized. In short, failure to find evidence of MR in one case does not entail that MR is
unsubstantiated tout court. As the dictum goes, absence of evidence is not evidence of absence.
This is a reasonable response to make. Something needs to be said about how to
generalize from a particular case to MR more generally. My response will come in stages. The
first reply is that pain is not the only mental state for which empirical arguments for MR are not
empirically supported. The second reply is that evidence against MR for these particular mental
states are representative of other mental states. The third reply is that looking at other organisms
is unlikely to provide different results. My response here cannot, of course, prove that there are
no cases of MR in the natural world. But it does put the burden of proof on the MR advocate to
supply evidence to corroborate his claims.
Showing that pain is not multiply realized in humans and octopodes does not prove that
all mental states are not realized in humans and octopodes. But evidence suggests that pain is the
rule, not the exception. A similar story can be told about the perceptual states and propositional
states of octopodes. One dramatic difference between octopus and human psychology is their
sensory modalities. Each possesses senses the other simply lacks. One of these, ‘infrasound’, is
an analogue of hearing in humans.34 Cephalopods lack the air-filled cavities and hard, vibrating
body parts necessary for hearing sound. But they are sensitive to low-frequency vibrations
undetectable by humans.35 A second sensory difference between humans and octopodes is the
latter’s possession of chemoreception, which roughly corresponds to taste and smell in the
former. Octopodes are sensitive to chemical stimuli, especially in the suckers located on their
34
35
Hanlon & Messenger 1996
Packard, Karlsen, & Sand (1990)
eight tentacles.36 This chemical sense can be used both tactically and at a distance, and it plays
an important role in the search for food and in navigation.37 The sensory diversity between
humans and octopodes also holds for the modalities that both organisms have in common. Much
has been made of the similarity in the eyes of octopodes and mammals.38 Yet despite the
structural similarities, there is one striking difference: octopodes are colorblind.39 Vision is a
dominant sense for octopodes, but they use differences in contrast rather than differences in hue
to make visual discriminations.40 Another difference with octopodes is that control and function
of their limbs are largely decentralized.41 Each arm contains about 43.75 million nerve cells (350
million total), while the brain contains roughly 150 million nerve cells. 42 The result is that many
of the limbs actions are “curiously divorced from” and “performed without reference to” the
brain.43 This would be like having legs that always got you where you need to go, even though
you don’t feel them moving. Both with respect to the senses unique to octopodes and to the
senses we have in common, octopus psychology and human psychology look quite different.
Clearly, if octopodes have different senses than humans then the mental states
corresponding to those senses will also be different. For instance, because octopus vision is
based on contrast rather than color, the visual-state stimuli for octopodes differ from human
visual stimuli. A red object in a field of green could cause a perceptual state in humans, but it
would not serve as such an input for an octopus. This point is even stronger when applied to
senses the two organisms do not share. Obviously chemical signals in the water are not stimuli
36
Rossi & Graziadei, 1958
Mather 1995; Hanlon & Messenger 1996; Nixon & Young 2003
38
Putnam 1967; Block & Fodor 1972; Messenger 1991; Mather 1998; Shapiro 2000; Gilett 2003; Gleadhall &
Shashar 2004
39
Hanlon & Messenger 1996
40
Messenger, 1977; Messenger, Wilson & Hedge 1973
41
Nixon & Young 2003
42
Young 1971
43
Hanlon & Messenger, 1996, p. 15
37
for human psychological states, nor are vibrations of air stimuli for octopodes. This point applies
not only to perceptual states themselves, but also to the propositional attitudes that are the
outputs of that perception. For instance, both humans and octopodes may hold the belief “There
is a fish nearby”. But the stimuli which can cause the belief for an octopus (e.g., chemicals in
the water, the low-frequency vibration of the water, etc) are not remotely close to the stimuli
which might cause the belief for humans (e.g., a bright red, fish-shapped sense datum, a sound
argument that entails “There’s a fish nearby”, etc.). As with pain, the Ramsey sentence that
accurately describes the perceptual states or propositional attitudes of an octopus is true only
when the S, B, and M variables are filled in such a way that makes the sentence false when
describing human mental states.44
At the realizer level, more neural similarity can be found with regard to memory and
learning.
Aizawa (2007) Aizawa & Gillett (2009) and Bickle (2006, 2008) discusses long-term
potentiation (LTP) as a possible “molecular mechanism” for memory consolidation using Nmenthly-D-aspartate (NMDA) receptors. Shamrat et al (2008) show that octopodes also use LTP
for memory consolidation.45 At the structural/functional level, they demonstrate that
the octopus, like vertebrates, shows a clear separation between the sites of shortand long-term storage…. Short- and long-term memory of nonreflexive behaviors
thus appear to have a universal organization principle in which short-term traces
are stored in the behavior-controlling circuitry separate from the site acquiring or
controlling the consolidation of long-term memory traces.46
At the molecular level, they suggest that the mechanisms are N-menthly-D-aspartate (NMDA)
dependent, much like the analogous mechanisms in the hippocampus of vertebrates (also see
Tsesarsaia et al, 2009). Llinás & Sugimori (1995) and Augustine et al (1995) discuss the cylic
44
A similar story can be told for the behavioral effects of these perceptual and propositional states. Holding the
mental state “There’s a fish nearby” will likely result in the octopus behavior “Grab with tentacle”, while it cause
me to exhibit “take a photograph for the vacation scrap-book” behavior.
45
See also Glanzman, 2008; Hochner et 2006; Hochner et al 2003
46
Shamrat et al (2008), p. 341, emphasis added
adenosine monophostphate (cAMP) and calcium binding proteins found in the synapses of the
cephalopod brain; these substances are also discussed at length by empirically-minded
philosophers.47
The similarities between the two organisms are so similar, in fact, that
researchers using research on cephalopods as a complement for the studying of diseases in
humans, going so far as to investigate the effects of using compounds based on chemicals
extracted from octopodes to treat human neurological conditions.48
The upshot of all this is that pain is not the only mental state which is not multiply
realized in humans and octopodes. The perceptual states and propositional attitudes of the two
organisms are also functionally distinct, while the memory-consolidation systems are causally
homogenous.
This brings me to my second reply, that these specific mental states are
representative of other mental states we might consider. Qualitative states, perceptual states, and
propositional attitudes are all very broad categories of mental states.
They might not be
exhaustive, but they come close. Further investigation is likely to show that the patterns I’ve
discussed apply to other similar mental states. Even if we cannot conclude that no mental states
are multiply realized by humans and octopodes, we can conclude that the evidence of MR is not
yet forthcoming.
But what about MR in other organisms? Regarding MR for specific psychological states,
there are grounds for generalization here too. Pain is likely a very basic mental state, which is
likely to be found throughout the animal kingdom. So even if it is realized the same way in
humans and octopodes, it might be multiply realized elsewhere. But this response cuts both
ways. Because pain is so pervasive, there are more opportunities to multiply realize it. So the
probability of it being realized in the same way by such distantly related organisms with such
47
48
See Aizawa (2007) Aizawa & Gillett (2009) and Bickle (2006, 2008)
Budlemann et al 2008; D’este et al 2008; Boyle et al 1999
different selection pressures is all the more surprising.49 A similar worry holds for more complex
mental states. In virtue of being more complex, there are more places for the mental state to
differ, which lowers the probability of two different organisms realizing it. For instance, the
mental state “not surprised but still disappointed” is a rather nuanced mental state, the functional
identification of which is very close to other mental states like “stoically resigned” or “not
surprised but still annoyed”. Fine-grained mental states like these require subtle functional
differences to differentiate them, making it far less likely that the very same state is realized in
different ways.
Lastly, it is not at all clear that the prospects for MR would be improved by looking
elsewhere. I suspect one of the reasons that Putnam used octopodes in his discussion is that they
are such radically different creatures.
The octopus is intelligent enough for interesting
psychological investigation, yet in appearance it is physiologically alien. Moreover, the
evolutionary distance between humans and octopodes is vast; their last common ancestor,
Urbilateria, is hypothesized to have lived 5.5 million years ago (See figure 2). 50 Yet despite all
this, octopodes are far more intelligent than other invertebrates. All in all, the octopus seems an
ideal candidate for MR. So the question is “If not here, where?” If we cannot find MR in such
diverse organisms as humans and octopodes, where can we find it? If MR were true, we would
predict a much higher probability of finding genuine cases in the most distantly related taxa, and
lower probability in closely related organisms. We didn’t find MR where it was most likely to
be, so the odds of finding it elsewhere are slim. We can frame the issue as a dilemma: The less
49
It should also be noted that Stefano et al (2002) discuss the pain-response system in terms of vertebrates and
invertebrates, not specific species. This suggests that pain is realized the same way in all organisms that realize
pain.
50
Chen 2004, Erwin & Davidson, 2002 Fendokin & Wagner 1997
similar two organisms are, the less likely we are to ascribe the same mental states to them. But
the more similar they are, the more likely it is that they have similar realizers.51
(Figure 2) Image adapted from Nielson, 2001.
So it seems that the empirical evidence against pain being multiply realized in humans
and octopodes is less isolated than might first appear. Pain is not the only mental state for which
we have evidence against MR. Other kinds of mental states, like propositional attitude, are not
multiply realized either. Since the pattern is clearly against MR, some clear evidence must be
supplied before we take MR seriously as an empirical hypothesis.
Moreover, it is highly
unlikely that further investigation will come up with different results. The odds of finding other
mental states that satisfy the criteria for MR is low, as are the odds of finding other organisms
that stand a better chance of supporting MR. It is the nature of empirical evidence that one data
set is rarely conclusive. But the weight of the evidence is clearly against MR, and at the very
least its proponents must do more to defend it.
6. Conclusion
51
Polger (2002) raises a similar worry about MR is a slightly different context.
In this paper I have argued that MR is actually two separate claims:
(MR) Two organisms T1 and T2 multiply realize the same mental state F iff
(MRi) the same Ramsey sentence is true of both T1 and T2 when
describing F, and
(MRii) T1 realizes F in virtue of instantiating causal powers P, T2
realizes F in virtue of instantiating causal powers Q, and P ≠Q
with respect to the relevance criterion
The empirical argument for MR holds that both (MRi) and (MRii) enjoy evidential support. I
have argued that for pain in humans and octopodes, MR is false.
At the realized level,
behavioral observations suggest that pain is not functionally equivalent, and is therefore not the
same mental state in both organisms. At the realizer level, neurological evidence shows that the
pain-response systems of both organisms are incredibly similar, as it is a homologous trait shared
by both creatures. The same pattern holds for other mental states between the two organisms.
Further, the odds of finding genuine cases of MR between other organisms are low. The more
diverse the organisms, the less likely their mental states are functionally equivalent; the more
similar the organisms, the less likely their brains are functionally distinct. Instead of the obvious
and ubiquitous natural phenomenon suggested by Putnam (1967) or Block & Fodor (1974), we
find that the preponderance of evidence is against MR.
So where do we go from here? I’m inclined to think that the answer to this question
depends on how much one cares about contingency, possible worlds, and the like. If you’re
optimistic about the prospects of AI, then finding evidence of MR may be simply a matter of
time. If all you’re after is the logical possibility of MR, then the lack of evidence for it won’t
matter much. But if you’re a stickler for what really exists in the actually world, then the lack of
evidence for MR is telling. Without evidence for MR, it isn’t clear what the ramifications of the
hypothesis are. If I’m right, finding that evidence may be more difficult than we realize.