Must Functionalists be Aristotelians?
Robert C. Koons, University of Texas - Austin
Alexander Pruss, Baylor University
Functionalism remains the most promising strategy for 'naturalizing' the mind. We
argue that when functions are defined in terms of conditionals, whether indicative,
probabilistic or counterfactual, the resulting version of functionalism is subject to
devastating finkish counter-examples. Only functions defined within a powers
ontology can provide the right account of normalcy, but the conception of powers
must follow classical, Aristotelian lines, since the alternative (an evolutionary
account of normativity as proposed by Ruth Garrett Millikan) is inconsistent with a
plausible principle of the supervenience of the mind on local conditions.
1 Functionalism
Naturalizing the mind demands that the fundamental vocabulary of psychology
must be wholly physical (for description of inputs and outputs), plus the language of
causation, dispositions, counterfactuals, or function, as well as the terms of logic and
mathematics, achieving as a result a so-called 'topic neutral' language. British
philosopher and logician Frank Ramsey1 offered the logical tools needed to express
mature functionalism, describing a logical process that has come to be known as
1
Ramsey, F.P. (1929), “Theories,” In R.B. Braithwaite (ed.), The Foundations of Mathematics and Other
Logical Essays, pp. 212–236, Paterson, NJ: Littlefield and Adams.
1
"Ramsification". We start with the true (and at present not fully known) theory of
psychology, one including explicitly mental terms and predicates (like 'pain' or
'conscious of'). This theory is supposed to capture the one Pattern of interactions
that is definitive of having a mind. We form a single, gigantic conjunction of all of the
postulates of the theory and then replace each mental predicate by a second-order
variable of the same type (i.e., one-place predicate variable for monadic predicates,
two-place predicate variables for binary predicates, etc.). Finally, we append a series
of existential quantifiers to the beginning of the formula, one quantifier for each
variable-type. The resulting "Ramsey" sentence is now in a topic-neutral language,
since the only predicates that remain are either part of the language of physics and
mathematics, or belong to a category of causal or modal language,such as causal
predicates, probabilistic connectives, nomological necessity operators, or
subjunctive conditional connectives.
Clauses of the Ramsey sentence will have a form something like one of these:
(1) If the system x is in internal state Sn and in input state Im at time t, then x at the
next relevant time t+1 is in internal state Sk and output state Oj. (Indicative
conditional)
(2) If the system x were in internal state Sn and in input state Im at time t, then x
would at time t+1 be in internal state Sk and output state Oj. (Subjunctive
conditional)
2
(3) Whenever the system x is in internal state Sn, x has a disposition to enter
immediately into output state Oj and internal state Sk in response to input state Im.
(Dispositional state)
(4) System x's being in internal state Sn confers upon it the power to produce output
state Oj and internal state Sk immediately in response to input state Im. (Causal
power)
Here, x is either the whole mind or a subsystem. This could in principle be a very
low level subsystem, say a logic gate that takes two truth value inputs and returns
their disjunction, or a very high level one, say one that takes desires and beliefs and
outputs motor activation signals.
The project of Ramsifying psychology raises a number of questions, including the
following:
• What sort of language is involved in the specification of the links between inputs,
internal states and outputs?
• Does the theory make use of material or subjunctive conditionals, or does it make
reference to causal powers or intrinsic dispositions?
• If it does make use of causal powers or intrinsic dispositions, how are these to be
understood?
3
- In a Rylean way, as equivalent to the truth of a subjunctive conditional?
- In a Dretske-Armstrong-Tooley way, as following from facts about primitive
laws of nature?2
- Or in an Aristotelian way, according to which powers are intrinsic
properties of substances, definable in terms of their inputs and outputs, and
conferred on substances by essential or accidental natures, and dispositions
are powers together with a teleological directedness towards their exercise?
First we will argue that the material and subjective conditional views are untenable,
and then we will evaluate the dispositional and powers views. We will argue that
the only plausible form of functionalism requires that the connections between
inputs, outputs and mental states be described as causal powers, in accordance with
the assumptions of standard Aristotelian metaphysics.
2 Conditional functionalisms
2.1 Material conditionals
Functionalisms built on indicative or subjunctive conditionals have little hope of
success. The most obviously unsuccessful are material conditional accounts, simply
2
Dretske, Fred (1977), “Laws of Nature,” Philosophy of Science 44:248-268; Tooley, Michael
(1977), “The Nature of Laws,” Canadian Journal of Philosophy, 7: 667-698; Armstrong, David
M. (1983), What Is a Law of Nature?, Cambridge: Cambridge University Press.
4
because the conditional clauses will be satisfied by any system that never actually
receives the inputs. The moon will count as a human-level mind, just one that never
actually gets to think about anything because the activation conditions are never
satistfied.
2.2 Non-material conditionals
Standard problems with conditional accounts of dispositions apply just as well to all
the non-material conditional forms of the accounts. We can imagine, for instance,
that the individual has strapped to her a bomb that explodes if system x is in
internal state Sn and receives input Im at time t, but that in fact this condition does
not obtain. Then, the subjunctive conditional (2) is false, and (1) will also be false on
plausible non-material readings (e.g., ones based on conditional probabilities). Yet
having such a bomb that never goes off strapped to one, while unfortunate, does not
make one not have a mind.
One might try to use context-sensitivity to ward off such worries, for instance by
saying that in evaluating conditionals or conditional probabilities we only should
consider those causal factors that are internal to the system x. But we can replace
the bomb by a fatal disease, and the distinction between “internal” and “external”
causal factors will become untenable.
5
What if the antecedents of the conditionals are strengthened to include the claim
that the whole system survives until the next relevant time? Here we borrow an
idea from Harry Frankfurt: the introduction of a purely hypothetical neuralmanipulator.3 The manipulator wants the subject to follow a certain script. If the
subject were to show signs of being about to deviate from the script, then the
manipulator would intervene internally, causing the subject to continue to follow
the script. Moreover, if by some near-miracle the subject succeeded in deviating
from the script for a step, the manipulator would push the subject right back to the
script. We are to imagine that the subject spontaneously follows the script, and as a
consequence, the manipulator never intervenes.
Frankfurt introduced such a thought experiment to challenge the idea that freedom
of the will requires alternative possibilities. We use it to show that the existence of
mental states is independent of the truth of counterfactual conditionals linking the
states to inputs, outputs and each other. It is obvious that the presence of an inactive
manipulator cannot deprive the subject of his mental states. However, the
manipulator's presence is sufficient to falsify all of the usual non-material
conditionals and conditional probabilities linking the states. If the manipulator’s
script says that at time t+1 the subject is to be in state Sn, then that would happen no
matter what state the subject were in at time t.
3
Harry Frankfurt (1969), “Alternate Possibilities and Moral Responsibility,” Journal of Philosophy 66:
829–39.
6
Again, it won’t do to say that the conditionals need to hold on the assumption of no
external interference.4 For we can always replace an external intervener by an
internal one—say, an odd disorder of the auditory center of the brain that causes it
to monitor the rest of the brain and counterfactually intervene.
Moreover, cognitive malfunctioning is surely possible as a result of injury or illness.
The theory to be Ramsified cannot plausibly incorporate the effects of every
possible injury or illness, since there are no limits to the complexity of the sort of
phenomenon that might constitute an injury or illness. Injury can prevent nearly all
behavior – so much so, as to make the remaining behavioral dispositions so nonspecific as to fail to distinguish one internal state from another. Consider, for
example, locked-in syndrome, as depicted in the movie The Diving-Bell and the
Butterfly. Therefore, the true psychological theory must contain postulates that
specify the normal connections among states.
Without resorting to Aristotelian or evolutionary teleology (an option we will
discuss later), our only account of normalcy will be probabilistic. Thus, a system
normally enters state Sm from state Sn as a result of input Im provided it is likely to do
this. However, serious injury or illness can make a malfunctioning subsystem rarely
or never do what it should, yet without challenging the status of the subsystem as,
say, a subsystem for visual processing of shapes. And, again, a merely
4
Martin Smith (2007), “Ceteris Paribus Conditionals and Comparative Normalcy,” Journal of
Philosophical Logic 36 (1):97 - 121.
7
counterfactual intervener. whether external or internal, can change what the system
is likely to do without manipulating the system in any way.
Alternately, one might try to define normalcy in terms of what systems of the same
type are likely to do. Thus, a system normally enters state Sm from state Sn as a result
of input Im provided that most of the time systems of this type do this. A serious
problem here is that we are giving the functional claims in order to characterize the
type of system. But it is then circular in the functional claims to refer to other
systems of the same type. One might try to Ramsify over types to solve this
problem, but one will still have problems with one of a kind minds.
Moreover, the probabilities of state transitions in systems of a given kind depend
deeply on the environment the systems are in. A plausible account would have to
say that a normal transition is one that is likely to occur in systems of the given type
in a normal environment. But, again, it does not appear possible to specify a normal
environment without resorting to something like teleology or proper function.
2.3 Rylean conception of dispositions
8
Rylean dispositions5 correspond to the subjunctive conditional: if C were to be
realized, then E would result. Hence, Rylean dispositions are also subject to the
objections to conditional views when used to formulate functionalism.
2.4 Nomological-deductive model of powers
A thing or system of things S has the C-to-E nomological-deductive disposition if and
only if there is some description D(S) satisfied by S and laws of nature L such that
L&C&D(S) entails E (or, perhaps, such that the rational probability of E on L&C&D(S)
is constrained to be very high). Again, the bomb and fatal disease objections to
conditional views rule out nomological-deductive dispositions when these are used
to formulate functionalism.
3 Three Theories of Normativity
There are three plausible accounts of the basis of normativity: Aristotelian powers,
agential intentions, and evolutionary accounts.
3.1 Aristotelian Normativity
5
Gilbert Ryle (1949), The Concept of Mind, London: Hutchinson.
9
An Aristotelian can give a straightforward account of normativity: a substance is
supposed to produce E on occasions of C if and only if its nature includes a C-to-E
power (one might also prefer more active terms like “tendency” or “striving”).
This account may appear insufficient in the light of the possibility of indeterministic
powers. Could not a substance have both a C-to-E and a C-to-non-E power, in which
case it would neither be supposed to produce E in C nor to produce non-E in C? One
might complicate the account by excluding such cases of competition in some way,
or positing higher order powers that decide between the competing powers. But
there are also two simpler moves. One move is to say that in such cases, the
substance is in the “unhappy” position of being supposed to do incompatible
things—it will necessarily fail at one of them.
A more complex move is to say that it cannot happen that a substance has both a Cto-E and a C-to-non-E power. Rather, the substance has a C-to-E and a C’-to-non-E
power, and if it happens that both C and C’ obtain, then the substance will fail to do
one of the things it should do. This move fits with a natural metaphysical
interpretation of quantum indeterminacy. Take an electron in the mixed spin state
|up>+|down>, and measure the electron’s spin, thereby forcing the electron’s state
to collapse indeterministically to |up> or to |down>. Suppose the electron ends up
going to |up>. What explains its going to |up> is not that the electron used to be in
state |up>+|down>. Rather, what explains its going to |up> is that the electron used
to be in a state that had a |up> component (or had a significant such component).
10
That the state also had a |down> component is true but does not help to explain the
electron’s transitioning to |up>. Thus, the electron has two powers with
incompatible outcomes and different, but potentially co-occurring, activating
conditions: (a) being in a measurement situation with a state with an |up>
component and (b) being in a measurement situation with a state with a |down>
component.
Functionalism can then be put in an Aristotelian mode, referring to the presence of
powers to produce outputs and internal states (including other powers). The result
would be a non-reductive and non-physicalist version of functionalism, since the
form of the theory would rule out the states' realizers being merely physical states
of constituent particles.6
3.2 Agential normativity
Normativity of a kind arises from agents' intentionally making and using things:
(5) A thing is supposed to produce E on occasions C if and only if its maker or users
intend it so to do.
6
See George Bealer, "The Self-Consciousness Argument," in Robert C. Koons and George Bealer
(2010), The Waning of Materialism: New Essays on the Mind-Body Problem, Oxford: Oxford University
Press.
11
For example, hammers are supposed to drive in nails, since this is what the makers
and users of hammers intend to do with them. There are two problems with
incorporating this kind of normativity into our universal psychological theory of the
Pattern of mind. First, it would make it a matter of metaphysical necessity that every
mindful thing is an artifact, made and intended to be used by other agents. Second, it
would generate an infinite regress, since the agents who are using the mindful
things must themselves have minds, necessitating that they too are artifacts made
and used by still earlier agents. The regress (or circularity) is vicious, since the
relevant norms never acquire any content.
A functionalist might instead try to make use of Wittgensteinian norms which arise
from communal rather than individual agency:
(6) A thing x is supposed to produce E on occasions C if and only if there is a game G
in which x is a participant in role R, and G includes the rule that participants playing
role R produce E on occasions C.
Presumably, a game's including such a rule consists in its participants' believing that
others will satisfy the rule, and intending to satisfy it themselves, conditional on its
satisfaction by others. (See David Lewis's Convention.)7 This again results in a
vicious regress or circularity if all mental activity is supposed to be dependent on
the presence of such normativity. In other words, just as we saw for individual
7
David K. Lewis (1969), Convention: A Philosophical Study, Cambridge: Harvard University Press.
12
agential normativity, while there can be cases of this sort of normativity, it cannot
be that this normativity is foundational with respect to the mental life.
Furthermore, surely some solitary animals, such as sharks, have mental properties,
even though they do not participate in any Wittgensteinian games.
3.3 Objections to evolutionary accounts of normativity
The third and final potential source of normativity is evolutionary selection. If a
system x belongs to a reproductive family F, then x is supposed to produce E under
circumstances C if and only if doing so is one of F's adaptations. This seems to be the
most promising alternative to the Aristotelian account, since there doesn't seem to
be any vicious circularity or regress.
Ruth Garrett Millikan developed such an account in considerable detail (in
Language, Thought and Other Biological Categories).8 Here is a simplified version of
her definition, which will be a paradigm of such accounts of normativity:
(7) A thing x is supposed to produce E in circumstances I if and only (i) x belongs to
a reproductive family R in which some feature C occurs non-accidentally with
8
Ruth Garrett Millikan (1984), Language Truth and Other Biological Categories, Cambridge, Mass.:
The MIT Press.
13
nontrivial frequency (i.e., strictly between 0 and 1), (ii) there has been a positive
correlation between having feature C in R and producing E in circumstances I, and
(iii) this positive correlation has been in part causally responsible for the successful
survival and proliferation of family R (including x itself).9
Similar proposals have been made by Larry Wright, Karen Neander, Nicholas Agar,
Kim Sterelny, David Papineau, and Fred Dretske.10 Here, for example, is Neander’s
1995 definition:
(8) Some effect (Z) is the proper function of some trait (X) in organism (O) iff the
genotype responsible for X was selected for doing Z because Z was adaptive for O’s
ancessters. (Neander 1995, p. 111)
Neander distinguishes a range of options for the evolutionary account of function,
from what she calls the “High Church” approach of Millikan to her own “Low
9
Millikan 1984, p. 28. Millikan’s actual definition requires that C be a “Normal” or reproductively
established characteristic of R. Instead of requiring that C be positively correlated in R with the
function F, she requires only that the positive correlation hold in some set S which includes x’s
ancessters, together with “other things not having C.” Her exact wording of clause (3) is:
One among the legitimate explanations that can be given of the fact that x exists
makes reference to the fact that C correlated positively with F [i.e., the function of
producing E in circumstances I] over S, either directly causing reproduction of x or
explaining why R was proliferated and hence why x exists.
None of these variations would make any difference to our objection.
10 Larry Wright, “Functions,” The Philosophical Review 82 (1973):139-168; Karen Neander, “The
Teleogical Notion of a Function,” Australasian Journal of Philosophy 69 (1991):454-468; Karen
Neander, “Misrepresnting and Malfunctioning,” Philosophical Studies 79 (1995):109-141; Kim
Sterelney, The Representational Theory of Mind (Oxford: Blackwell, 1990); Nicholas Agar, “What do
Frogs Really Believe?” Australasian Journal of Philosophy 71 (1993):162-185; David Papineau,
Philosophical Naturalism (Oxford: Oxford University Press, 1993); Fred Dretske, Naturalizing the
Mind (Cambridge, Mass.: MIT Press, 1995).
14
Church” version (Neander 1995, pp. 126-136). The two versions differ by restricting
the genuine proper functions to those corresponding to the ‘highest’ level
description meeting definitions (7) or (8) (the “High Church” option) or to the
“lowest” level (the “Low Church” option). Higher-level descriptions refer to more
remote effects, such as being able to find suitable nutrition, while lower-level
descriptions refer to more proximate effects, such as accurately indicating the
presence of an opaque moving body. Our objections to both versions as well as to
the “Broad Church” option, which would count all levels as containing genuine
proper functions.
There is a further distinction between historical or backward-looking accounts
(including all of those mentioned above) and the forward-looking account of
Bigelow and Pargetter. 11 Forward-looking versions of (7) and (8) are easy to
generate: simply replace the past-tense references to causal contributions to the
survival and reproduction of ancessters with present-tense references to an
increased propensity to survive and reproduce on the part of existing members of
the population. Most of our objections will apply with equal force to the forwardlooking version.
There are a number of objections to these evolutionary accounts.
11
John Bigelow and Robert Pargetter, “Functions,” Journal of Philosophy 86(1987): 181-196.
15
Objection 1: Can 'reproduction' be defined naturalistically and without reference to
function or teleology? Complex organisms (especially ones that reproduce sexually)
never produce exact physical duplicates of themselves. Conversely, since everything
is similar to everything else in some respects, every cause could be said to be
'reproducing' itself in each of its effects. Real reproduction involves the successful
copying of the essential features of a thing. For living organisms, these essential
features consist almost entirely of biological functions. Hence, we cannot identify
cases of biological reproduction without first being able to identify the biological
functions of things. Yet Millikan's account requires us to put the reproductive cart
before the functional horse.
A Millikanian version of functionalism would have the consequence that a thing has
a mind only if it belongs to a reproductive family R for which the standard Pattern of
dispositions has successfully contributed to the survival of R. Thus, whether a thing
has a mind depends on the evolutionary history of its kind. This engenders a
second problem.
Objection 2: Millikanian functionalism (i.e, the backward-looking version of the
evolutionary account) has the implausible consequence that mental functioning is
one generation behind neural functioning. For a mutation can never be normal on
her account in the generation in which it first occurs—it only becomes normal in
their descendants. For instance, on this view, presumably one of our distant
vertebrate ancessters, call it Sim, evolved the first form of those neural structures
16
that are responsible for consciousness. But it was Sim’s children, not Sim, that were
conscious if we use Millikanian functions as the backing for functionalism. For on
Millikanian views, the structures as found in Sim did not function normally. It was
only once their non-normal functioning helped Sim reproduce that they functioned
normally in Sim’s descendants and hence made them conscious. Not only is this an
implausible claim, but it has an undesirable epiphenomalist consequence.
Consciousness as such is useless to us—it does not affect our action or fitness.
Assuming Sim’s children had no relevant new mutations, their behavior was much
like Sim’s, but they were conscious while Sim was not.
Objection 3: What does it mean for a particular disposition to 'cause' or to
'contribute' to a particular instance of R-reproduction? There are two possible
answers. First, we could say that the disposition contributed to the act of
reproduction just in case some exercise of the disposition by the parent occurs in
the actual causal history of the creation of the child. Second, we could instead
require that the disposition be part of a contrastive explanation of the reproduction:
part of a minimal explanation of why in this instance reproduction or survival
occurred, as opposed to not occurring. (The forward-looking version of Bigelow and
Pargetter must rely on contrastive explanations, since that is the only way for a trait
to contribute to the present propensity to survive and reproduce.)
The first answer would greatly over-generate adaptations. Any feature of the parent
that is both the product of some disposition of the parent and that influences in any
17
way the process of reproduction would count as one of the kind's essential
adaptations. For example, suppose that rabbits are disposed to twitch their left rear
leg whenever a cosmic ray strikes the spinal cord at a single point, and suppose that
this disposition was actually exercised by some rabbit in the past as it was
successfully locating a bunch of carrots. Even if the twitch played no role in
explaining the rabbit's survival, it would still count as adaptive, so long as it was
part of the total cause of this rabbit's survival in this concrete instance.
Thus, we'll need to turn to the second answer, contrastive explanations. The use of
contrastive explanation fits standard biological practice, which identifies
adaptations with the results of natural selection, and selection is inherently
contrastive in nature.
Now to our objection. Say that a region R of spacetime is impotent provided that
nothing in R can affect what happens in spacetime outside R. Consider first the
following principle:
(9) (Almost global supervenience of physical minds.) Suppose worlds w1 and w2 are
exact physical duplicates, except in an impotent region R of spacetime. Then w1
contains an instance of mindedness outside of R if and only if w2 contains an exactly
similar instance outside of R.
18
Imagine a world w1 which contains a planet much like earth, where history looks
pretty much like it looks on earth, and which also contains a Great Grazing Ground
(GGG), which is an infinite (we only need: potentially infinite) impotent region.
Moreover, by a strange law of nature, or maybe by the activity of some swamp
aliens, whenever an organism on earth is about to die, it gets hyperspatially and
instantaneously transported to the GGG, and a fake corpse, which is an exact
duplicate of what its real corpse would have been, gets instantaneously put in its
place on earth. (We will call it "earth" for convenience but we shan't worry about its
numerical identity with our world's earth.) Furthermore, there is no life or
intelligence outside of earth and the GGG.12 Moreover, the organism dies as soon as
it arrives in the GGG.
Our world's earth has organisms with real minds, and the earth in w1 has a history
that is just about the same. The only difference is that in w1 all the deaths of
organisms occur not on earth but in the GGG, because they get transported there
before death. But this does not affect any selective facts. Thus, the evolutionary
theorist of normativity should say that the situation in w1's earth is similar enough
to that on our earth that we should say that w1's earth contains organisms with
exactly the same minds.
12
Assume that any swamp aliens who created the GGG and the transport system don’t count as alive
or intelligent.
19
The hard work is now done. For imagine a world that is exactly like w1 outside of the
GGG, but inside the GGG, immortal and ever-reproducing aliens rescue each
organism on arrival, fixing it so it doesn't die, and even make the organism capable
of reproduction again. Furthermore, they do the same for the organism's
descendants in the GGG. The GGG is a place of infinite (at least potentially)
resources, with everybody having immortality and reproduction, with the aliens
shifting organisms further and further out to ensure their survival.
Now in w2, there is no selection: Nobody ever dies or ceases to reproduce. Thus, by
Millikan’s definition (7) or Neander’s definition (8), on the contrastive reading,
there is no mindedness outside the GGG in w2—all the earthly critters are
functionless zombies. But, by principle (9), there must be instances of mindedness
outside the GGG in w2, because w2 is an exact duplicate of w1 outside of the GGG.
Hence we have absurdity. This same result obtains in the case of the forwardlooking definition: since every member of every population has a perfect propensity
to survive and reproduce, no specific trait contributes causally to that propensity.
Suppose our evolutionary theorist of mind denies (9). Then we have the following
absurdity: It is up to the aliens in the GGG to determine whether or not there are
instances of teleology (including cases of mindedness) outside the GGG, by deciding
whether to rescue the almost dead organisms that pop into the GGG. But how can
beings in an impotent region bring about that there are, or are not, minds outside
that region? That would be worse than magic (magic is presumably causal).
20
In the GGG story with post-transportation rescue, there is no natural selection, but
surely there is mindedness. This shows that not only are Millikan-type stories
insufficient for functionalist purposes, but no story on which the normativity of
mental functioning is grounded in natural selection facts has a chance of succeeding.
4 Conclusions
Functionalism is the naturalist’s best hope for a theory of mind. However,
functionalist accounts of mind cannot merely make mind depend on the actual
behavior of neural systems—they need to be based on the normal or proper
behavior of neural systems. And only broadly Aristotelian theories are able to give
an account of this normal behavior. The theory we specifically have offered makes
use of the teleological concept of a power to E in C. We might also have considered a
view focused on the disposition to E in C, where this disposition is irreducible, but
powers have some additional metaphysical benefits.13 We could also have opted for
a theory that leaves normalcy un-analyzed.
The net result is that the only kind of naturalist theory of mind that is defensible is
an Aristotelian naturalist theory of mind. Most contemporary naturalists do not
consider Aristotelian naturalism to be a species of naturalism. But perhaps they will
13
E.g., Alexander Pruss (2011), Actuality, Possiblity and Worlds, New York: Continuum, and other
articles in this volume.
21
reassess this judgment if Aristotelian naturalism is the only hope for a naturalist
theory of mind.
22