Cognitive Control
Cognitive Control
Cognitive Control
COGNITIVE CONTROL
KENNETH R. HAMMONDs AND DAVID A. SUMMERS
University of Colorado University of Kansas
The theory developed here proposes that performance in cognitive tasks in-
volves two distinct processes: acquisition of knowledge and cognitive control
over knowledge already acquired. A conceptual and analytic framework is
presented which allows for the disentanglement of knowledge and control,
and for the quantification of each. Evidence from studies of multiple-cue
probability learning, clinical judgment, and interpersonal conflict supports
the theoretical usefulness of this framework and indicates that poor perform-
ance in cognitive tasks can often be attributed to incomplete cognitive control,
rather than incomplete knowledge. The importance of cognitive feedback—•
as opposed to traditional outcome feedback—for the development of cognitive
control is illustrated.
Although learning theorists have long habits"), and that a good share of the "use of
emphasized the distinction between learn- habits" depends on current motivation [p. 287].
ing and performance, little attention has Other theorists, also stressing the import-
been given to skill in the application of ance of distinguishing between learning and
knowledge in tasks which do not involve performance, have suggested additional
motor performance. Rather, there is an determinants of the application of knowl-
implicit assumption that once knowledge edge ; for example, fatigue (Hilgard, 1956)
has been acquired, the application of this and instructions (Mandler, 1967). Indeed,
knowledge is largely dependent on certain Mandler goes somewhat further and sug-
experimental circumstances. gests that a distinction be made between
For example, Deese and Hulse (1967) "the conditions under which a subject will
point out that discover a particular rule [and] his abilities
sometimes we have good reason to believe that an to use ... it [p. 22; italics added]."
organism is simply not demonstrating what it has Although the explanations for the inap-
learned because we have not chosen the proper condi- propriate application of knowledge men-
tions which will assure the overt display of the ap- tioned above are by no means identical,
propriate behavior [p. 62].
they are (with the exception of Mandler)
Leeper (1970) makes a similar point, but quite similar in one important respect.
draws attention specifically to motivational Specifically, they all assume that the po-
factors: tential for full application of knowledge
One of the distinctive points of cognitive learning
accompanies full acquisition of knowledge—
theory—a principle developed by Lashley (1929), except in psychomotor tasks.
Tolman (1932), and others—is that a clear distinc- The position taken here, however, is that
tion must be made between learning (or the "acquisi- acquisition and application are independent
tion of habits") and performance (or the "use of components of learning in cognitive tasks
1 as well as psychomotor tasks. Conse-
This paper is based on research undertaken at quently, we argue that the development of
the Institute of Behavioral Science, University of
Colorado, and is Publication 140 of the Institute. skill—or control—in the application of
The research was supported by National Institute knowledge already acquired is a matter to
of Mental Health Grant MH-16437-02. The be investigated. The purpose of this paper,
authors wish to thank Thomas Stewart for his therefore, is to introduce the concept of
assistance. We are particularly indebted to Berndt
Brehmer of the University of Umea for his valuable cognitive control, to indicate its theoretical
contributions made while he was a visiting Re- and methodological context, and to illus-
search Associate at the University of Colorado. trate its empirical significance in studies
2
Requests for reprints should be sent to Kenneth
R. Hammond, Institute of Behavioral Science, Uni- of human learning, judgment, and inter-
versity of Colorado, Boulder, Colorado 80302. personal behavior. In addition, traditional
COGNITIVE CONTROL 59
outcome feedback is identified as an im-
pediment to the development of cognitive
control.
THEORY
Fundamental to our treatment of cogni-
tive control is a methodological and con- S'l
ceptual framework developed for studying CRITER10N JUDGMENT
/ X ^*— — — •—•x s
//1 ^' ' A
s' \\
>•
/ /
/ "*'
•
1 .6
'
'
^v ''
^-'^^-~S
S
2
tZsv
\:
i I
11
11
t t p8 II
. . G
•i
4 5 6 8 9 3 4 5 6
BLOCK BLOCK
FIG. 2. Indexes of achievement (ra), knowledge (G), and control (J?,) in two multiple-cue
probability learning tasks. (In each task condition, n = 20.)
accurate performance; that is, subjects (R3) was far below optimal for subjects in
were told how each cue was related to the the nonlinear task, even after 200 trials.
criterion dimension and how each cue For subjects in the linear task condition,
should be weighted. Moreover, this infor- control (Ra) was hardly distinguishable
mation was provided to the subjects in two from knowledge (G).
different ways; verbally or pictorially. These findings indicate that knowledge
Following each learning trial, the subject and control can be disentangled empirically,
was given the usual outcome feedback; as well as statistically, and demonstrate
that is, the subject was informed of the that even when knowledge is complete,
correct answer. imperfect cognitive control can prevent
The results of this experiment indicated high achievement.
that poor performance in complex inference It is important to note that the findings
tasks (e.g., those involving nonlinear rela- summarized above are not peculiar to tasks
tions) can be attributed to difficulties in involving probabilistic cue-criterion rela-
cognitive control, as well as to difficulties tions ; that is, irreducible error. For ex-
in acquiring knowledge about the task. ample, Brehmer (1969) studied perform-
At the conclusion of training (200 trials), ance in four different multiple-cue tasks—
the level of predictive accuracy (ra) all permitting perfect accuracy (Re = 1.00)
achieved by subjects in the nonlinear task —and found considerable differences ac-
condition was significantly lower than that cording to the type of task relation the sub-
achieved by subjects in the linear task con- jects were required to learn and use. When
dition (.58 and .84, respectively). Yet, as the criterion was a simple linear function of
can be seen in Figure 2, knowledge, as the cues (Ye = X\ + X^), the subjects had
measured by G, was essentially the same little difficulty in rapidly achieving a high
in these two tasks by the end of training. level of performance. When the task re-
In contrast, however, cognitive control quired the subjects to utilize a complex
62 KENNETH R. HAMMOND AND DAVID A. SUMMERS
UJ
o
1
cc
.6
o
li-
ce /
UJ
Q.
z .4
4 5 4 5
BLOCK BLOCK
FIG. 3. Indexes of achievement (ra], knowledge (G), and control (R,) in two multiple-cue
learning tasks in which perfect achievement is possible. (In each task condition, n = 10.)
Rs. For when Goldberg produced perfect FIG. 4. Brunswik's lens model as applied to the
control by replacing the clinical judge with two-person case. (Note that characteristics of the
a perfectly consistent linear model (in judgment task are represented on the left, while
characteristics of each person's cognitive system are
effect setting R, = 1.0), performance (ra) represented on the right.)
improved for 86% of the judges. Similar
findings have been reported by Dawes arise under these conditions, and all have
(1971) in a study involving judgments found that the reduction of conflict occurs
made by graduate admissions committees.
slowly, if at all, even when subjects are
These results support Dudycha and given outcome feedback which clearly in-
Naylor's (1966) contention that dicates to each that his cognitive system is
humans tend to generate "correct" strategies but inadequate (Hammond & Brehmer, in
then, in turn, fail to use their own strategy with any press). A commonsense explanation sug-
great consistency . . . One is left with the con- gests that the slow reduction in conflict
clusion that humans may be used to generate infer-
ence strategies but that once the strategy is obtained merely reflects the reluctance of most
the human should be removed from the system and people to alter their beliefs (even in the
replaced by his own strategy [p. 127]. face of contradictory evidence). Examina-
tion of conflict behavior using the theoreti-
The results also confirm our views concern-
cal analysis proposed here, however, points
ing the theoretical distinction between
to a different interpretation.
knowledge and control.
Such analysis is enhanced by the fact
Cognitive Control and Interpersonal Conflict that not only can each subject's behavior be
examined in relation to an environmental
There is evidence that the framework task system, but each subject's behavior can
proposed here is significant not only for also be examined in relation to the response
individual performance, but for certain system of the other (Hammond, Wilkins,
kinds of two-person interactions as well. & Todd, 1966). As can be seen in Figure 4,
Of particular concern here are those inter- the theoretical framework outlined earlier
personal conflicts (quarrels) which arise —including the concept of cognitive con-
when two persons must exercise their judg- trol—is directly applicable to this type of
ments under conditions of uncertainty. interpersonal situation as well as to the
Such quarrels are studied in settings in individual learning and judgment tasks
which two or more individuals who have described above: r represents agreement
a
learned to utilize probabilistic cues in between the responses of the two subjects,
quite different ways (different cue weights G represents the convergence between the
and/or different function forms) are re- two systems, and RI and Rz represent the
quired to reach joint agreement in multiple- cognitive control exerted by Si and Sz,
cue probability learning tasks. respectively.
A large number of studies have investi- The results from interpersonal conflict
gated the conflict, or disagreement, which studies conducted in eight different nations
64 KENNETH R. HAMMOND AND DAVID A. SUMMERS
2 3 2 3 4 ' I 2 3
BLOCK BLOCK BLOCK
FIG. 5. Indexes of achievement (ra), knowledge (G), and control (R,) in a multiple-cue proba-
bility learning task under conditions of outcome, mixed, and cognitive feedback. (In each feed-
back condition, n = l2.)
The major innovation in this study is about the task. The G indexes for these
that one group of subjects received only two groups were quite high throughout,
cognitively oriented feedback during learn- and were near identical by the end of 200
ing (referred to as "lens model" feedback trials. On the other hand, control in the
by Todd and Hammond). Using an on- cognitive feedback condition was signi-
line computer arrangement, each subject ficantly (p < .01) higher than in the mixed
was periodically given information about feedback condition, even at the end of
task properties (cue validities), as well as training. (Note that neither knowledge
properties of his cognitive system (cue nor control improved substantially in the
utilization coefficients). A second group outcome feedback condition.)
was given only traditional outcome feed- In short, the cognitively oriented feed-
back ; that is, the correct answer after every back given to subjects in the cognitive and
trial. A third group was given both cogni- mixed feedback conditions facilitated the
tive and outcome feedback (referred to as acquisition of knowledge about task proper-
"mixed" feedback). ties. The addition of outcome feedback to
Using ra as a measure of performance, the mixed feedback condition, however,
Todd and Hammond found that the group served to decrease cognitive control and
which received only cognitive feedback per- thus contributed to the suboptimal per-
formed most accurately, followed by the formance achieved by these subjects.
mixed feedback group, and finally by the A second study employed the same non-
outcome feedback group. That the cogni- linear judgment task used by Deane et al.
tive feedback group performed better than (described earlier), but used an on-line
the outcome feedback group is theoretically computer graphics terminal as a means of
gratifying; but what is surprising is that providing feedback. This technique en-
subjects who received both the cognitive abled the subject to compare periodically
and outcome feedback ("mixed" feedback) his cue weights and function forms with the
performed less well than subjects who were properties of the task system in pictorial
given lens model feedback only. form. (Some practical implications of the
A reanalysis of the Todd and Hammond rapid learning associated with computer
data revealed that performance in the mixed graphics techniques are pointed out in
feedback conditions can be explicated in Hammond, 1971.) Even though a small
terms of cognitive control. As can be seen number of subjects was employed (N = 5),
in Figure 5, subjects in both the cognitive the experiment yielded clear results. Per-
and mixed feedback conditions were able formance (ra) reached the limit of achieve-
to achieve a high degree of knowledge ment after two or three comparisons. It
66 KENNETH R. HAMMOND AND DAVID A. SUMMERS