COMMENTARY

Increasing fluid intelligence is possible after all

Robert J. Sternberg*
Office of the Dean of the School of Arts and Sciences, Tufts University, Medford, MA 02155

A
lmost 40 years ago, Jensen (1) sure, to resolve the debate over Further Research
claimed that, when all is said whether f luid intelligence is, in at least There will be a need for follow-up on
and done, there is not much some meaningful measure, trainable. these interesting results because of sev-
one can do to raise people’s Given these positive results, a reader eral limitations of Jaeggi et al.’s study
IQs. Over the years, there have been might wonder why it has taken nearly 40 (10), none of them calling into question
various attempts to do so, which gener- years to show training effects. Why the the obtained results. Eight are of some
ally have yielded somewhat ambiguous wait? The reason is, in large measure, concern.
results (2). Even successful attempts (3) that recent cognitive-based theories of First, with regard to the main inde-
have typically involved training people intelligence have provided new insights
pendent variable, there was only one
on the same kinds of items on which into just what kind of training would be
training task in the study, so it is un-
they would be tested, so that it was not
clear to what extent the results can be
clear whether the training was generaliz-
generalized to other working-memory
able, rather than merely a result of prac-
tice effects on particular item types (4). Fluid intelligence is tasks. It would be important to show
Further complicating the picture have that the results are really about working
been studies showing that IQ tends to trainable to a significant memory rather than some peculiarity of
be fairly highly heritable, with most reli- the particular training task.
able estimates ranging from ⬇0.5 to 0.8 and meaningful degree. Second, with regard to the main de-
(5). More recently, heritability has been pendent variable, there was only one
found to vary both with age, with IQ kind of fluid-ability test, geometric ma-
becoming more highly heritable in later successful. It is only in fairly recent trix problems from various tests of the
years (6), and with social class, with IQ years, relative to the age of the field, kind found in the Raven Progressive
more highly heritable in higher social that so-called ‘‘working memory’’ has Matrices (15) and similar tests. It would
classes (7). Although heritability does come to be viewed as a key determiner be important to show that the results
not imply the fixedness of a trait (e.g., of fluid intelligence (11). Working mem- generalize to other fluid-ability tasks
height is highly heritable but also modi- ory is usually viewed as that part of rather than being peculiar to this kind
fiable), the mixed results of training long-term memory that is available for of task. Matrix problems are generally
studies have been taken to be consistent active information processing, including considered to be an excellent measure
with the notion that IQ is relatively placement of information into and re- of fluid intelligence (16), but they do
fixed. trieval of information out of storage. place a particularly great demand on
IQ may be viewed as a composite Tasks such as backward digit span and working memory. At the same time,
comprising multiple elements: In many the n-back task used in this study are fluid-ability tests tend to be highly cor-
theories of intelligence, a distinction is good measures of working memory (12). related with each other (17), so general-
made between fluid and crystallized in- So the basic idea in Jaeggi et al.’s study ization would appear likely. Whether
telligence (8). Fluid intelligence com- (10) is that one can use modern cogni- generalization extends beyond the ma-
prises the set of abilities involved in tive theory to serve as a basis for trix tests to other kinds of cognitive
coping with novel environments and es- training, which should then produce a
tests, such as of spatial, numerical, or
pecially in abstract reasoning; crystal- training regimen that will make a
other abilities, remains to be seen.
lized intelligence is the product of the meaningful difference. This idea proved
Third, it is important to remember
application of these processes. Fluid in- to be correct.
that the fluid-ability tests, although the
telligence is often measured by tests Jaeggi et al.’s article (10) is important
dependent variables in Jaeggi et al.’s
such as figural analogy, classification, to the field of intelligence because it
shows that training can improve fluid study (10), were originally intended to
and matrix problems, whereas crystal-
intelligence, can do so across intelli- be independent variables in predicting
lized intelligence is measured by tests of
gence levels, and can do so in a theory- meaningful behavior in the real world
vocabulary and general information (9).
based way. To the extent that past (18). Such behavior would include,
Increasing Fluid Intelligence investigators failed to obtain such robust among other things, educational and
In this issue of PNAS, Jaeggi et al. (10) results, it appears to be because they occupational achievement. Tests of intel-
have made an important contribution failed to use the cognitive–theoretical ligence have also been shown to have
to the literature by showing that (i) basis that served as the basis for train- implications for other kinds of success,
f luid intelligence is trainable to a sig- ing in Jaeggi et al.’s study. These results including even health-related behavior
nificant and meaningful degree; (ii) the have important educational-policy impli- (19). So it would be useful to show that
training is subject to dosage effects, cations, because they suggest that the the training transfers to success in
with more training leading to greater results of conventional tests of intellec- meaningful behaviors that extend be-
gains; (iii) the effect occurs across the tual abilities and aptitudes provide indi- yond the realm of psychometric testing.
spectrum of abilities, although it is ces that may be dynamic rather than
larger toward the lower end of the static and modifiable rather than fixed.
Author contributions: R.J.S. wrote the paper.
spectrum; and (iv) the effect can be Most researchers have viewed intelli-
obtained by training on problems that, gence as largely fixed (13), although The author declares no conflict of interest.
Downloaded by guest on December 31, 2021

at least superficially, do not resemble others have argued strongly for its modi- See companion article on page 6829.
those on the f luid-ability tests. Their fiability (14). The latter are shown to be *E-mail: [email protected].
study therefore seems, in some mea- justified in their beliefs. © 2008 by The National Academy of Sciences of the USA

www.pnas.org兾cgi兾doi兾10.1073兾pnas.0803396105 PNAS 兩 May 13, 2008 兩 vol. 105 兩 no. 19 兩 6791– 6792
 Fourth, Jaeggi et al. (10) mention that, which can lead readers to query whether The subjects were all recruited from
with training, the value of fluid intelli- a placebo treatment in an additional the University of Bern community,
gence for predicting performance on control group might have led to a stron- which is likely to be a rather selective
other tasks can change. It would be im- ger comparison. In future work, one sample of individuals not typical of the
portant to show that the predictive power would want to include a training alter- population either of Switzerland or,
of the fluid-ability tests, after training, is native that teaches something expected more generally, of developed countries
at least as high as or possibly even higher not to be relevant to performance on or certainly the world. It would be par-
than it was before training took place. the fluid-ability tests. ticularly important to test elderly peo-
Fifth, as Jaeggi et al. (10) recognize,
Seventh, because there is only one ple, who are at risk for loss of f luid
their study does not address whether the
study, there is a great need for replica- ability. The sample was also relatively
training is durable over extended
periods of time. Too often increases of tion. In single studies, one can never be small (n ⫽ 70).
intelligence obtained through training sure whether there are aspects of the None of these criticisms detracts from
programs have proved to be fleeting design or procedure that lead to results the central importance of the results of
(20). Future work would need to assess that later prove not to be replicable. Jaeggi et al.’s study (10). On the con-
durability over varying periods of ex- Finally, the effects need to be exam- trary, they suggest that their study
tended time. ined on a much wider range of ability should and probably will be the first in a
Sixth, the control group in Jaeggi et levels and, in general, of types of par- long series instigated by this pioneering
al.’s study (10) had no alternative task, ticipants than were tested in this study. research.

1. Jensen AR (1969) How much can we boost IQ and itability of IQ in young children. Psychol Sci 14:623– 628. 14. Feuerstein R (1980) Instrumental Enrichment: An Inter-
scholastic achievement? Harvard Edu Rev 39:1–123. 8. Carroll JB (1993) Human Cognitive Abilities: A Survey vention Program for Cognitive Modifiability (Univer-
2. Detterman DK, Sternberg RJ, eds (1982) How and How of Factor-Analytic Studies (Cambridge Univ Press, New sity Park Press, Baltimore, MD).
Much Can Intelligence Be Increased? (Erlbaum, York). 15. Raven J, Raven JC, Court JH (2003) Manual for Raven’s
Mahwah, NJ). 9. Mackintosh N (1998) IQ and Human Intelligence (Ox- Progressive Matrices and Vocabulary Scales. Section 1: Gen-
3. Kramer AF, Willis SL (2002) Enhancing the cognitive ford Univ Press, New York). eral Overview (Harcourt Assessment, San Antonio, TX).
vitality of older adults. Curr Directions Psychol Sci 10. Jaeggi SM, Buschkuehl M, Jonides J, Perrig WJ (2008) 16. Anastasi A, Urbina S (1997) Psychological Testing (Pren-
tice–Hall, Englewood Cliffs, NJ), 7th Ed.
11:173–177. Improving fluid intelligence with training on working
17. Cattell RB (1971) Abilities: Their Structure, Growth, and
4. Detterman DK, Sternberg RJ, eds (1993) Transfer on memory. Proc Natl Acad Sci USA 105:6829 – 6833.
Action (Houghton Mifflin, Boston).
Trial: Intelligence, Cognition, and Instruction (Ablex, 11. Engle RW, Tuholski SW, Laughlin JE, Conway ARA
18. Binet A, Simon T (1916) The Development of Intelli-
Norwood, NJ). (1999) Working memory, short-term memory, and gen-
gence in Children, trans Kite ES (Williams & Wilkins,
5. Plomin R (2004) Nature and Nurture: An Introduction eral fluid intelligence: A latent variable approach. J Exp Baltimore, MD).
to Human Behavioral Genetics (Wadsworth, Thousand Psychol General 128:309 –331. 19. Gottfredson LS, Deary IJ (2004) Intelligence predicts
Oaks, CA). 12. Kane MJ, Engle RW (2002) The role of prefrontal cortex health and longevity, but why? Curr Directions Psychol
6. Plomin R, DeFries JC, McClearn GE, McGuffin P (2008) in working-memory capacity, executive attention, and Sci 13:1– 4.
Behavioral Genetics (Worth, New York), 5th Ed. general fluid intelligence: An individual differences 20. Spitz HH (1992) Does the Carolina Abecedarian Early
7. Turkheimer E, Haley A, D’Onofrio B, Waldron M, perspective. Psychonom Bull Rev 9:637– 671. Intervention Project prevent sociocultural mental re-
Gottesman II (2003) Socioeconomic status modifies her- 13. Jensen AR (1998) The G Factor (Praeger, Westport, CT). tardation? Intelligence 16:225–237.
Downloaded by guest on December 31, 2021

6792 兩 www.pnas.org兾cgi兾doi兾10.1073兾pnas.0803396105 Sternberg