Must Functionalists be Aristotelians? - Robert C. Koons

Must Functionalists be Aristotelians?
1 Functionalism
Functionalism remains the most promising strategy for 'naturalizing' the mind. It
resulted from synthesizing of logical behaviorism and brain/mind identity theory.
Like behaviorism, functionalism defines mental states as dispositions, intervening
between public, measurable inputs and outputs (behaviors). Unlike classical
behaviorism, which defined each mental state-­‐type directly in terms of the behavior
it produces in response to each stimulus, functionalism defines mental types
recursively. Each mental type is potentially both input and output for other types. A
key advantage of functionalism is that, like brain/mind identity theory, it provides
(on the assumption that only physical states are causally efficacious) a natural
explanation of the strong supervenience of mental properties on physical ones, a
critical commitment of naturalism. In fact, functionalism could taken to be either a
'second-­‐order' identity theory (as in Shoemaker) or first-­‐order identity theory (as in
David Lewis's version).
In the early 20 th century, the motivation behind naturalism was that of
"operationalizing" psychology, eliminating the need for things like introspection,
interpretation, or empathy. Such naturalism led to the demand 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 Ramsey offered the logical tools needed to
express mature functionalism, describing a logical process that has come to be
known as "Ramseyfication"or "Ramsification". We start with the true 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 language: either causal predicates, or
operators or logical connectives of nomological necessity or subjunctive
conditionality.
Mental properties can now be identified either with second-­‐order physical
properties (e.g., mental property n is identical to the second-­‐order property of
1

having a physical property that plays the role of variable n in the actual witness to
the truth of the Ramsey sentence) or with first-­‐order physical properties (e.g.,
mental property n is identical, in the class of human beings, to that physical
property that in fact plays the role of variable n in the actual witness to the truth of
the Ramsey sentence in the population of human beings).
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 is at
time t+1 be 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)
(3) Whenever the system x is in internal state Sn, x has the 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 one 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.
(a) How do we specify inputs and outputs, especially gross behavioral outputs? Can
this be done without including an element of interpretation? Can it really be done
entirely in the vocabulary of fundamental physics?
(b) What are the 'states' – the names of which are replaced by variables when
Ramsifying? States of what? Each bearer of the properties that realize the Ramsey
sentence must be capable of persisting through time, since the Ramsey sentence
(and the psychological theory from which it originates) makes reference to causal
connections among the property-­‐bearer's internal states, and these causal
connections require the passage of time. Thus, interpreting the Ramsey sentence
presupposes a solution to the problem of diachronic identity for mereologically
incontinent composite things – the diachronic identity of organisms, machines,
organs, neurons, etc. Is diachronic persistence of the property-­‐bearer built into the
functional specification? That is, are the biological and organic processes of material
intake and output specified in the psychological theory? If so, this leads to biological
parochialism: it entails that non-­‐organic machines and other entities cannot have
2

minds. If not, can the inputs and outputs be scattered arbitrarily far apart? If so, how
can we ensure that a scattered set of distinct persons and machines don't constitute
a single mind? (Searle's Chinese room counter-­‐example.) Moreover, how “natural”
must the states be? If we are sufficiently broad about what counts as a state, we will
have the problem of too many minds. For instance, in addition to my mind, there is
the danger that there will be another—“much bigger”—mind whose states are
conjunctions of my neuronal states together with the spiral nature of our galaxy
(which is causally irrelevant to my neuronal states and unchanging on an individual
human timescale).
(c) 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?
We will focus here on the third set of questions. 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 (including the notions of substance, accident,
essence, formal and final causes).
2 Conditional functionalisms
2.1 Material conditioanls
Functionalisms built on indicative or subjunctive conditionals have little hope of
success. The most obviously unsuccessful are material conditional accounts, simply
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.
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., conditional probabilities). Yet having such a
bomb strapped to one, while unfortunate, does not make one not have a mind, if the
bomb never goes off.
3

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 cancer, and the distinction between “internal” and “external” causal
factors will become untenable.
What if the conditionals are strengthened to include the claim that the whole system
survives until x+1? Here we borrow an idea from Harry Frankfurt: the introduction
of a purely hypothetical neural-­‐manipulator. The manipulator wants the subject to
follow a certain script. If the subject were to shown signs of being about to deviate
from the script, then the manipulator would intervene internally, causing the
subject to continue to follow the script. We are to imagine that the subject
spontaneously follows the script, and as a consequence, the manipulator never
intervenes.
Frankfurt introduced the thought experiment to challenge the idea that freedom of
the will requires alternative possibilities. We introduce 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. It is false that if the
subject were not in state Sn, then he would not immediately afterward be in state
Sm, since the presence of the manipulator makes it impossible for the subject not to
be in state Sm, if that is part of the script.
Moreover, cognitive malfunctioning is surely possible: e.g., as a result of injury or
illness. The theory to be Ramseyfied 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 non-­‐
specific 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.
But without resorting to Aristotelian teleology, 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.
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
4

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 Ramseyfy 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.
3 Three rival conceptions of function, power or
disposition
3.1 The Aristotelian conception of powers
The Aristotelian notion of power is characterized by four features:
1. Powers are possessed by substances (things with a per se unity).
2. Powers are intrinsic to those substances.
3. Powers are defined in terms of an input/output function.
4. Powers are conferred on substances by their essential or accidental natures.
An Aristotelian C-­‐to-­‐E power can be defined as: under circumstances C to produce a
characteristic effect E (either in oneself or in something standing in the appropriate
relation R(C) to oneself).
An Aristotelian power can be defeated or thwarted, and its effects can be distorted
by interfering factors. There could be higher-­‐order powers, whose characteristic
effect is the blocking or distorting of the exercise of lower-­‐order powers, as an
antidote neutralizes the characteristic effect of a poison.
Aristotelian powers are "conferred" on things by the natures of substances and
accidents. This connection too is defeasible. A thing can have a nature such as to
have power P and yet lack P. This too can be explained in terms of the operation of
higher-­‐order powers: e.g., the power to deprive something with nature N of one of
its natural powers.
3.2 Rylean conception of dispositions
5

Rylean dispositions correspond to the subjunctive conditional: if C were to be
realized, then E would result, and hence are subject to the objections to conditional
views when used to formulate functionalism.
3.3 Nomological-deductive model of powers
A thing or system of things S has the C-­‐to-­‐E nomological-­‐deductive disposition iff
there is some description D such that S has D, and there are laws of nature L such
that L&C&D entails E (or, perhaps, such that the rational probability of E on L&C&D
is constrained to be very high). Again, the bomb and cancer objections to
conditional views rule out ND dispositions when these are used to formulate
functionalism.
4 An Aristotelian Functionalism?
Can a plurality of substances bear jointly something analogous to a causal power?
Could they possess jointly some sort of reducible causal power? No – the only
alternative to Aristotelian power is a disposition, grounded in a subjunctive
conditional or in laws of nature. Powers are possessed by things by virtue (or as an
aspect of) their natures, what it is to be such a thing. Mere heaps of things have no
nature.
An Aristotelian is a kind of ontological emergentist: the whole organism, with its
associated powers, is a metaphysically fundamental entity, not merely a roundabout
way of referring to its material atoms and their relations.
Aristotelian powers of a substance provide an objective basis for normativity: if a
substance has the Aristotelian C-­‐to-­‐E power, then it is supposed to produce effect E
whenever in circumstances C (except, perhaps, when the C-­‐to-­‐E connection is
overridden by one of the substance's own essential powers).
Functionalism could 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. See George Bealer, "The Self-­‐Consciousness Argument," in
Koons & Bealer 2010: the Aristotelian can make use of impredicative, type-­‐free
higher order theory, allowing the second-­‐order, functionally specified properties
(powers) to be among the properties in the domain of the Ramsey quantifiers.
5 Last-ditch defenses
6

5.1 Reliance on actual causal connections?
What if the functionalist reformulated the psychological theory, replacing any
reference to dispositions or powers with a reference to an actual causal connection
between tokens of the relevant types? Clauses of the resulting Ramsey sentence
would look something like this:
(4) There has been at some point in the past a system x in internal state Sm and
input state In such that x's being in states Sm and In (immediately) caused x to be in
internal state Sk and output state Oj.
Such a functionalism would make mental states extrinsic to their bearers. This
would give rise to extreme Swampman problems. What mental state I am in -­‐-­‐ and
even whether I am in any mental state at all -­‐-­‐ would depend on a large number of
extrinsic facts about the causal history of similarly constructed systems in the
remote past. My thoughts could be deprived hypothetically of some of their content,
and my experience of some of its constituents, by simply deleting certain remote
facts about actual causal connections among the states of entities similar to me in
constitution.
5.2 Merely possible causal connections?
Suppose the functionalist were to replace the dispositional connections with the
claim that it is metaphysically possible for items of the first kind to cause (or to be
part of the cause) of items of the second kind. The clauses of the Ramsey sentence
would look like this:
(5) It is metaphysically possible for a system x's being in states Sm and In to cause
immediately x to be in states Sk and Oj.
This requires some form of nomological or dynamic essentialism. If the causal
powers are only contingently associated with a property, then every physical object
would be conscious and rational, since its physical states could have the relevant
causal connections. So, at the least, the functionalist must make one small step
toward Aristotelianism.
This account would lead to massive indeterminacy of content, thanks to the
possibility of misfirings. For example, my concept of the number two would also be
a concept of the number three, since the first concept is capable of playing, on
particular occasions, the role demanded by the second concept. To tighten this up,
the account would have to introduce normativity: a state can fill a certain role if it is
metaphysically possible that it does so correctly.
7

Thus, any viable version of functionalism must incorporate standards of
normativity.
6 Accounts of normativity
6.1 Aristotelian account
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 the C-­‐to-­‐E
power. A similar account can be given of conditional norms: a substance is supposed
to produce E on occasions of C, given that it has accident A, if and only if the nature
of A includes the C-­‐to-­‐E power (and this power is not neutralized or overridden by
any of the substance's own natural powers).
6.2 Agential intentions
Normativity of a kind arises from agents' intentionally making and using things:
(6) A thing is supposed to produce E on occasions C if and only if its maker or users
intend it so to do.
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 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:
(7) 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.) This again results in a vicious
regress or circularity.
8

Could the circularity be turned into a recursive definition of normativity? No,
because there is no norm-­‐free base case for the specification of psychological states.
6.3 Evolutionary accounts
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.
7 Objections to evolutionary accounts of normativity
7.1 Backward-looking vs. forward-looking accounts
Evolutionary normativity could be defined in either a backward-­‐looking or forward-­‐
looking form, depending on whether an adaptation is defined in terms of a kind's
actual evolutionary history or in terms of its present propensities to survive and
reproduce itself.
(8) Backward-­‐looking version: a C-­‐to-­‐E disposition is an adaptation of kind F if and
only if:
(a) F is a historical reproductive family: present members of F exist because they are
copies of earlier members,
(b) many members of F have the C-­‐to-­‐E disposition, and
(c) each member of F has an ancestor (or, perhaps, a large number of ancestors)
whose successful self-­‐reproduction was caused by at least one exercise (or, perhaps,
many exercises) of its C-­‐to-­‐E disposition.
(9) Forward-­‐looking version: a C-­‐to-­‐E disposition is an adaptation of kind F if and
only if
(a) F is a potential reproductive family: many present members of F have a
significant propensity to make future members of F,
(b) all reproductively fit members of F have the C-­‐to-­‐E disposition, and
(c) the presence of the C-­‐to-­‐E disposition in the reproductively fit members of F
plays an indispensable role in the causal explanation of those members'
reproductive fitness.
The forward-­‐looking accounts fall down at exactly the crucial point: they require
prior specification of the kind's natural or normal environment. No kind of thing is
reproductively fit in all possible environments, and every kind of thing is
reproductively fit in some possible environment. For example, shoes are
9

eproductively fit in the context of a shoe-­‐duplicating service. In fact, every possible
kind of thing is optimally adapted to some possible environment. Consequently,
without a restriction to normal environments, the forward-­‐looking version would
classify every possible feature as adaptive and, hence, as normative. This fact
renders the forward-­‐looking account useless to the functionalist, since normative
conditions are supposed to distinguish proper functioning from malfunctioning.
The backward-­‐looking account sidesteps this difficulty, since it identifies the
'normal' environment of a historically reproductive kind with the environmental
conditions in fact encountered by ancestral members of the kind. This may lead to
some over-­‐generation of normativity, since it seems plausible that living things
occasionally survive very abnormal conditions by some lucky accident. The
backward-­‐looking theorist must treat the features that enabled such serendipitous
success as real adaptations of current members of the kind, however unlikely it is
that the supposed 'adaptation' will ever be used again. However, this over-­‐
generation can be mitigated by requiring a certain number of past repetitions of
actual contributions to survival before a disposition counts as adaptive.
7.2 Millikan's account as a paradigm of the backward-looking
theory of adaptation
Ruth Garrett Millikan developed such an account in considerable detail (in
Language, Thought and Other Biological Categories). Here is a simplified version of
her definition:
(10) A thing x is supposed to produce E in circumstances C if and only x belongs to a
reproductive family F, members of F do regularly produce E in circumstances C, and
some past instances of successful reproduction of members of F were caused by
those members' producing E in circumstances C.
There is a serious problem with Millikan's account: 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.
7.3 The Great Grazing Ground objection
10

A Millikanian version of functionalism would have the consequence that a thing has
a mind only if it belongs to a reproductive family F for which the standard Pattern of
dispositions has successfully contributed to the survival of F. Thus, whether a thing
has a mind depends on the evolutionary history of its kind.
What does it mean for a particular disposition to 'cause' or to 'contribute' to a
particular instance of F-­‐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 is 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 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
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.
[Insert the Great Grazing Ground objection here]
We can also consider a Frankfurt-­‐like variation on GGG. Suppose that there is a
remote but vigilant bio-­‐manipulator who wants life on earth to follow a certain
script (for obscure numerological and theological reasons). If the script is followed,
the manipulator does nothing; at the first sign of incipient deviation, he annihilates
the earth. In fact, the script is spontaneously and coincidentally followed by the
actual course of events, so the manipulator does nothing. The manipulator's
presence falsifies all the usual subjunctive conditionals, resulting in no natural
selection, and, consequently, no adaptations, no evolutionary normativity, and no
mental states.
A second variation: in the case of deviation, the manipulator would produce a very
large number of copies of each member of a reproductive kind. In this way, each
thing's functioning as it does causes a reduction in its reproductive fitness. In this
scenario, every feature of living things would be maladaptive.
11

7.4 Non-Humean, Realist Accounts of Causation
The Millikanian might argue that the naturalized account of adaptation and
normativity can be salvaged from the GGG objection by shifting from a Humean
conception of contrastive causal explanation (defined in terms of subjunctive
conditionals) to an anti-­‐Humean one. One way to do this would be to appeal to
Aristotelian powers. A power of a biological substance is historically adaptive if
there have been cases in the history of the substance's kind in which an exercise
that power, in conjunction with the exercise of other powers of that substance and
of things in the substance's environment, had the substance's successful survival or
reproduction as one of its effects.
Is there a third way, in addition to the Humean and Aristotelian approaches? Once
again, we could turn to the nomological-­‐deductive theory of dispositions. Clearly the
laws in question must be defeasible or "oaken" laws, since there is no intrinsic
feature of organisms that determines nomologically that that creature will succeed
in surviving and reproducing itself.
What sorts of properties are related by causal laws? They must be properties that
occur in fundamental causal laws, or properties whose instances are linked by
primitive causal ties. If we accept naturalism as the constraint on an acceptable
functionalism, then it is physical properties, and more specifically, microphysical
properties (fundamental properties of fundamental particles) that appear in such
laws. Gross, macrophysical properties appear only in derived, non-­‐fundamental
laws. However, it is impossible to derive defeasible laws about the adaptiveness of
biological features without adding postulates about the normal environment of the
organism in question, the issue that the functionalist was hoping to avert through an
evolutionary account of normativity.
12