the four causes of behavior

7/29/2019 The Four Causes of Behavior

1/5

The Four Causes ofBehavior

Peter R.

Department ofPsychology, Arizona State University, Tempe, Arizona

Judging whether is bet-

ter explained as an associative or

computational process requires

that we clarify the key terms. This

essay provides a framework for

discussing explanation, association,

and computation; it leaves learning

as an unexamined primitive.

Aristotle (trans. 1929) described

four kinds of explanation. Because

of mistranslation and misinterpre-

tation by "learned (San-

tayana, 1957, p. his four "be-

causes [aitia]"were derogated as

an incoherent treatment of causal-

ity (Hocutt, 1974). Although an-

cient, Aristotle's four

provide an invaluable framework

for modem scientific explanation,

and in particular for resolution

the current debate about learning.

In Aristotle's framework,are triggers, events that

bring about an "effect."This is the

contemporary meaning of

Philosophers such as Hume, Mill,

and Mackie have clarified the crite-

ria for identifying various efficient

causal relations necessity, suf-

ficiency, insufficient but necessary


2/5

it for?" and "Why does itcall for functional (final)

of the fittest,

theory, and

s in general provide

ers . Most of mo der n---

known as physics can be written in terms offunctions that optimize certain

rays following paths

as errant formal,

nt causes. A

them browse high foliage; this final

cause does not displace formal

(variation and natural selection)

and material (genetic) explanations;

nor is it an efficient causeSuch equations descri But non e of those othe r

course of change from one s causal explanations make sensein concert with initial c without specification of the final

cause. Biologists reintroduced final

causes under the euphemism "ulti-

mate mechanisms," referring to the

models are, they are not efficient and material causes of a

Two systems that share similar

in insects, birds, and bats-provide

ary pressures and varieties of strate-

gies adequat e for that function),even though the wings are not

mologues are not evolved fromthe same organ in an ancient for-

bear). Analogical-functional analy-

ses fall victim to "the analogical fal-

lacy"only when it is assumed that

similarity of function entails simi-larity of (evolutionary his-(physiological)

causes. Such confounds can be pre-

vented by accounting for each typeof cause--

Efficient then, are theI

initial conditions for a change ofstate; final are the ierminal

conditions; formal causes are mod-

els of transition between the initial

and terminal conditions; material

causes are the substrate on which

these other causes act.

Skinner (1950) railed against for-

mal (" theor iz ing" ) , ma te r i a land final

("purposive") causes, andefficient causes as "the vari-

ables of which behavior is a func-

tion." He wa s concerned tha t

complementary causes woul d beused in lieu of, rather than along

with, his functional analysis. But of

all behavioral phenomena, condi-

tioning is the one least able to becomprehended without reference

to all four causes: The ability to be

conditioned has evolved because

of the advan tage it confers in ex-ploiting efficient causal relations.

Final Causes

shapes behavioral

trajectories into shortest pa ths to

(Killeen,When a stimulus predicts a biologi-

cally significant event (an uncondi-tioned stimulus, US), animals im-prove their fitness by "learning

associations" among ex ternal

Copyright 2001 American Psychological Society


3/5

events, and between those events

and app ropr iate actions. Stable

niches-those inhabited by most

plants, animals, and fungi-neither

require nor support learning:

pisms, taxes, and simple reflexes

regularities of light, t ide, and sea-----

son. However, when the environ-ment changes, it is the role of learn-ing to rewire the machinery t o

exploit the new contingencies. Bet-

ter exploiters are better repre-

sented in the next generation. This

is the final-ultimate, in the biolo-

gists' of condition-ing. Understanding learning re-quires knowing what the learned

responses may have accomplished

in the environments that selected

for them.

Efficient Causes

These are the prototypical kinds

of causes, important enough for sur-vival that many animals haveevolved sensitivity to them. Parame-

ters that are indicators of efficient

in space andtime, regularity of

association, and similarity-affect

both judgments of causality by hu-

mans 1993) and speed of con-ditioning &Matute,

Material Causes

The substrate of learning is the

nervous system, which provides an

embarrassment of riches in mecha-

nisms. Development of formal and

efficient explanations of condition-

ing guide the search for opera-tive neural mechanisms. turn,

elucidation of that neural architec-ture can formal modeling,

as parallel connectionist mod-els-neural nets- that emul ate

various brain functions. Each of the

four causes is a resource for under-standing the others.

Formal Causes>/---~.------

Models are proper subsets of all

that can be said in a modeling lan-guage. Associationist and computa-

tional

languages of

and automata, respectively. Their

structures are sketched next.

Associative Models

Material implication, the suffi-cient relation (if C, then E; symbol-

ized as provides a simplisticmodel of both efficient causality

and conditioning. holds thatwhenever C, then also E; it

whenever C and E. When the pres-ence of--a-cue (C, the conditioned

stimulus, or CS) accurately predictsa reinforcer (E, the US), the strength

of the relation increases. The

conditional probability of the US

given the

this all-or-none relation to a proba-bility. Animals are to

the presence of in the ab-

sence of the CS, only ifthis probability is a cause

said to be necessary for the effect.

Unnecessary effects degrade condi-

tioning, just as unexpected events

make an observer question hisgrasp of a situation.

Good predictors of the strengthof learning are (a) thebetween-"-,"~..-----'--probabilities andu

rn

the diagnosfic-

iswhich the

uncertainty

occurrence of the effect (US). As isthe case for all probabilities, mea-

surement of these conditionals re-

quires a defining context. This maycomprise combinations of cues,physical surroundings , an d his-tory of reinforcement. Reinforce-

ment engenders an updating of theconditionals; speed of conditioning

depends on the implicit weight ofevidence vested in the prior condi-

tionals. The databases for some con-

ditionals-such as the probability

of becoming ill after experiencing a

particular taste-often start small,

so that one or pairings greatly

increase the conditional probability

and generate tas te aversions. Ear-

lier pairings of the taste and health,however, will give the prior condi-

tionals more inertia, causing theconditional probability to increase

more slowly, and possibly protect-

ing the individual from a taste

aversion caused by subsequent as-

sociation of the taste with illness.

More common stimuli, such as

may be slow to condition.--..--;--

because of---

that is not associated with illness.

Bayes's theorem provides a formal

model for this process of updating

This ex-

emplifies how subsets of probabil-ity theory can serve as a formal

model for association Asso-

ciative theories continue to evolve

in light of experiments manipulat-

ing contextual variables; Hall

(1991) provided an excellent his-

tory of the progressive constraint

of associative models by data.

Computational Models

Computers are machines that

associate addresses wi th contentsthey go to a file specified by

an address and retrieve either a da-

tum or an instruction). Not only do

computers associate, but associa-tions compute: "Every finite-state

machine is equivalent to, and can

b 'simulated' by, some neural net"

1967, p. 55). Computers

can instantiate all of the associative

models of conditioning, and their

inverses. For the computational

metaphor to become a model, itmust be restricted to a proper sub-set of what computers can do; oneway to accomplish this i s via the

theory of automata (H opki nsMoss, 1976). Automata theory is aformal characterization of compu-

tational architectur'es. A critical

distinction amo ng automata i s

Published by Publishers


4/5

memory: Finite automata can dis-tinguish only those inputs (histo-

ries of conditioning) tha t can be

represented in thei r finite internal

memory. Representation may be

incrementally extended with exter-nal memory in the form of push-down stores, f inite rewritable

disks, or infinite tapes. These am-plified architectures correspond toChomsky's

free grammars, context-sensitive

grammars, and universal Turing

machines, respectively. Turing ma-

chines are models of the architec-tu re of a gene ral-purpose

t h a t c a n c o m p u t e a l l

expressions that are computable by

any machine. The architecture of a

Turing machine is deceptively sim-

ple, given its it is

access to a potentially infinitememory "tape" that gives it this

power. Personal computers are in

principle Turing siliconuniversality

has displaced most of the brass in-struments of an earlier psychology.

The Crucial Distinction

Memory is also what divides theassociative from the comput a -

tional approaches. Early reductionof memory to disposition requiresfewer memory states than late re-duction and permits faster-reflex-

ive-responses; late red uction is

more flexible and "intelligent."imals' behavior may reflect compu-

tation at any level up to, but not ex-

ceeding, their memory capacity.

Most human behaviors are simple

reflexes corresponding to finite au-tomata. Even the most complicated

repertoires can become "automa-tized" by practice, reduc ing anoriginally computation-intense re-sponse-a child's attempts to tie ashoe--to a mindless habit. The ad-aptation permitte d by learning

would come at too great a price if it

did not eventually lead t o auto-

matic and thus fast responsivity.

Consciousness of actio n permits

adaptation, unconsciousness per-

mits speed.In traditional associative theory,

information is reduced to a poten-

tial for action ("strength" of associ-

ation between the CS and US) and

stored on a real-time Such fi-

nite automata with limited memo-ries'-are ina dequa te as models ofconditioning because "thenature of

the representation can change-the

sort of information it holds can be

influenced by [various post op-erations]" (Hall, 1991, p . 67). Rats

have memorial access to more ofthe history the environment and

consequences than captured by

simple Bayesian updating of dispo-

sitions. Miller Blaisdell,

tol, Gunthe r, Miller, 1998; see

also this issue) provided one com-putational model that exemplified

such late reduction.

If traditional associaters are too

simple to be a viable model of con-ditioning, unrestricted computers(universal Turing machines) are

too smar t . Ou r f ini te memory

stores fall somewhere in between.

Automata theory provides a gram-mar for models that ran ge from

simple switches an d ref lexes,

through complex conditional asso-ciations, to adaptive systems thatmodify their software as theylearn. The increased memory this

requires is sometimes internal, and

sometimes external- found in

marks, memoranda, an d behavior

("gesturing facilitates the produc-

tion of fluent speech by affecting

the ease or difficulty of retr ieving

wo rd s from lexical memory,"Krauss, 1998, p. 58). Context is of-ten more than a cue for

it constitutes a detailed,addre ssabl e form of storage lo-cated where it is most likely to be

needed. Perhaps more often than

we realize, the medium is memory.The difference between

tionistic and computational models

reduces to which automata they

are with; and this is

correlated with early versus late re-duction of information to action.

The challenge now is to identify

the class an d capacity of automata

that are necessary to describe the

capacities of a species, and the ar-chitecture of associations within

such automata that suffice to de-

scribe the behavior of individualsas they progress through condi-

tioning.

Comprehending Explanation

Many scientificstem not so much from differences

in understanding a phenomenon as

from differences in understanding

explanation: expecting one type of

explanation to do the work of other

types, and objecting when otherscientists do the same. Exclusive

focus on final causes is derided as

teleological, on material causes asreductionistic, on efficient causesas mechanistic, and on formal

causes as "theorizing." But respect

for the importance of each type of

explanation, and the correct posi-

tioning of constructs within appro -

priate empirical domains, resolves

many controversies. For example,

formal constructs;

they are not located in the organ-

ism, but in our probability tables or

computers, and only emulate con-

nections formed in the brain, and

contingencies found in the inter-face of behavior a nd environment.

Final causes are not time-reversed

efficient causes. Only one t ype of

explanation is advanced when we

determine the part s of the brainthat are active during conditioning.

Provision of one explanation does

not reduce the nee d for the othertypes. Functional causes are not al-ternatives to efficient causes, but

ompletions of them.Formal analysis requires a lan-

g u ag e , an d mo d e l s mu s t b e a

proper subset of that language. The

signal issue in the formal analysis

of conditioning is not association

2001 American PsychologicalSociety


5/5

the four causes of behavior

Documents