Testing a Cognitive Control Model of Human Intelligence

Abstract

The definition of human intelligence and its underlying psychological constructs have long been debated. Although previous studies have investigated the fundamental cognitive functions determining intellectual abilities, such as the broadly defined executive functions including working memory, the core process has yet to be identified. A potential candidate for such a role might be cognitive control, a psychological construct for the coordination of thoughts and actions under conditions of uncertainty. In this study, we tested a cognitive control model of intellectual ability by examining the association between cognitive control, measured by a perceptual decision-making task and by the attention network test, and general intelligence including components of fluid intelligence (Gf, concerning the ability to solve problems by abstraction) and crystalized intelligence (Gc, related to learning from prior knowledge and experience) measured by the Wechsler Adult Intelligence Scale. We also examined the potential role of cognitive control as a core process involved in another determinant of intellectual abilities, the working memory, measured by the N-back tasks and the working memory complex span tasks. The relationship among intelligence, cognitive control, and working memory was examined using structural equation modeling. Results showed that cognitive control shared a large amount of variance with working memory and both measures were strongly associated with Gf and Gc, with a stronger association with Gf than Gc. These findings suggest that cognitive control, serving as a core construct of executive functions, contributes substantially to general intellectual ability, especially fluid intelligence.

Figure 1

Schematic of the backward majority function task-masked (MFT-M). (a) An illustration of a 3:2 congruency condition in the MFT-M and the event sequence of a trial. Participants are requested to report the majority of arrow directions (left or right) among a set of imperative stimuli. The upper-right panel shows the three possible arrow ratios. (b) Timeline of the stimuli presentation under different exposure time (ET, in ms). Duration of each event is indicated by the length of each texture. Participants are required to make a response within a 2500 ms response window starting from the onset of the arrow set and the entire trial lasts 5000 ms.

Figure 2

Schematic of the revised attention network test (ANT-R). For each trial, a 100 ms cue (no box flashes under the no cue condition, a single box flashes under the valid and invalid cue conditions, and both boxes flash under the double cue condition) is followed by a variable fixation period (0, 400, 800 ms) and then by the imperative target. The target (the center arrow) is flanked on each side by two other arrows presented for 500 ms, followed by an inter-trial fixation period varying between 2000 and 12000 ms (mean = 4000 ms). Participants are required to report the direction of the target.

Figure 3

Schematic of the spatial and verbal N-back tasks. (a) Illustration of the spatial N-back task. Three conditions are involved in the task: 0-back, 1-back, and 2-back. Participants should make responses to the location of the yellow box each trial (1 represents up, 2 represents left, 3 represents right, and 4 represents down). The arrow under each stimulus indicates the correct response key. (b) Illustration of the verbal N-back task. Four conditions are involved in the task: 0-back, 1-back, 2-back, and 3-back. For the 0-back condition, participants should click ‘left’ botton when the letter X is presented, and click ‘right’ botton when other letters are presented. For the other conditions, participants should click ‘left’ button when the current letter matches the letter 1/2/3 times ago, otherwise click ‘right’ button.

Figure 4

Schematics of the working memory span tasks. (a) An illustrative trial of the operational span task (OSpan). A math problem to be solved is followed by a letter. Subsequently, another math problem to be solved is followed by another letter. At the end of the trial, all letters should be recalled by checking corresponding boxes sequentially, followed by a feedback. (b) An illustrative trial of the rotation span task (RotSpan). A rotated letter to be judged is followed by an arrow. Subsequently, another rotated letter to be judged is followed by another arrow. At the end of the trial, all arrows should be recalled (both size and direction) by checking corresponding arrowheads sequentially, followed by a feedback. (c) An illustrative trial of the symmetrical span task (SymSpan). A picture to be judged as symmetrical or not is followed by a square at one location of the 4 × 4 grid. Subsequently, another picture to be judged is followed by another square at a different location. At the end of the trial, all squares (locations) should be recalled by checking corresponding location boxes sequentially, followed by a feedback.

Figure 5

Structural equation model with general intelligence (IQ), cognitive control (CC), and working memory (WM) as latent variables. Double-headed arrows connecting second-order latent variables (ellipses) represent the correlations between constructs. Single-headed arrows from the second-order latent variables to the first-order latent variables (ellipses), and from the latent variables to the manifest variables (rectangles) represent the loadings on the constructs. Single-heading arrows with numbers pointing to the manifest variables represent the error variance of each task. Double-headed lines connecting the error variances represent the correlations between measures. Standardized coefficients of correlations and loadings are presented next to the arrows. Significant paths are displayed by the solid lines (ps < 0.05), while nonsignificant paths are displayed by the dashed lines.

Figure 6

Structural equation models with cognitive control and working memory as latent variables. (a) SEM with CC, Gf, and Gc as latent variables. (b) SEM with WM, Gf, and Gc as latent variables. Significant paths are indicated by solid lines (ps < 0.05), while nonsignificant paths are indicated by dashed lines.

Figure 7

Structural equation model with WM and CC as latent variables. Significant paths are indicated by solid lines (ps < 0.05), while nonsignificant paths are indicated by dashed lines.