Adam Chuderski

The Contribution of Working Memory to Fluid Reasoning: Capacity, Control, or Both?

Fluid reasoning shares a large part of its variance with working memory capacity (WMC). The literature on working memory (WM) suggests that the capacity of the focus of attention responsible for simultaneous maintenance and integration of information within WM, as well as the effectiveness of executive control exerted over WM, determines individual variation in both WMC and reasoning. In 6 experiments, we used a modified n-back task to test the amount of variance in reasoning that is accounted for by each of these 2 theoretical constructs. The capacity of the focus accounted for up to 62% of variance in fluid reasoning, while the recognition of stimuli encoded outside of the focus was not related to reasoning ability. Executive control, measured as the ability to reject distractors identical to targets but presented in improper contexts, accounted for up to 13% of reasoning variance. Multiple analyses indicated that capacity and control predicted non-overlapping amounts of variance in reasoning.

The relational integration task explains fluid reasoning above and beyond other working memory tasks

This study aimed to evaluate how well fluid reasoning can be predicted by a task that involves the monitoring of patterns of stimuli. This task is believed to measure the effectiveness of relational integration—the process that binds mental representations into more complex relational structures. In Experiments 1 and 2, the task was indeed validated as a proper measure of relational integration, since participants’ performance depended on the number of bindings that had to be constructed in the diverse conditions of the task, whereas neither the number of objects to be bound nor the amount of elicited interference could affect this performance. In Experiment 3, by means of structural equation modeling and variance partitioning, the relation integration task was found to be the strongest predictor of fluid reasoning, explaining variance above and beyond the amounts accounted for by four other kinds of well-established working memory tasks.

Intelligence and Cognitive Control

The concept of “intelligence” has evolved in order to account for two facts, namely, intraindividual stability and interindividual variability of human intellectual performance. On one hand, people who outperform others in one class of tasks that involve reasoning, abstracting, or learning, will most probably excel in any other class of such tasks. On the other hand, within any class of cognitive tasks, one can find people who perform in an outstanding way as well as ones who fail. Early studies on the structure of intelligence examined if there is one general ability factor that manifests itself in all cognitive activities (g factor; Spearman, 1927) or maybe more domain-specific factors (e.g., linguistic, mathematical, etc., Thurstone, 1938) exist. These studies have converged to the widely accepted proposal (Caroll, 1993) of the three-layer hierarchy of factors, with general ability (g) on its highest level, several subordinate group factors (loading groups of tasks like mnemonic or perceptual tasks) on the middle level, and many low-level specific factors for particular tasks (e.g., one for perceptual speed). The most important middle-level factor is general fluid intelligence (Gf; Cattell, 1971), which represents human ability to adapt to the novelty and complexity by means of discovery of abstract relations in the environment and by their efficient goal-oriented application. In this respect, Gf differs from crystallized intelligence (Gc factor; ibidem), which consists in the application of one’s existing knowledge to the requirements of a situation.

An oscillatory model of individual differences in working memory capacity and relational integration

We present a novel computational model of the active buffer of working memory (WM). The model uses synchronous oscillations in order to bind an item and its corresponding context into one representation, while asynchronous oscillations are used to separate the representations. Due to the bindings, the model can ascribe proper meanings to items, as demonstrated by the replication of the effective rejection of distractors. The model predicts the inherent limitation of WM capacity in range of 1 to around six items that arises from the trade-off between the number and stability of separate oscillations. This trade-off depends on the strength of lateral inhibition exerted. The systematic variation in inhibition led to the exact replication of capacity distribution observed in a large sample, as well as to the prediction of a few novel capacity-related experimental effects. Finally, we showed that the differences in capacity can underlie the differences in a more complex ability of detecting relations governing a pattern of stimuli, called relational integration, which is known to be strongly related to the effectiveness of higher-order cognitive processing.

Bridge Over Troubled Water: Commenting on Kovacs and Conway's Process Overlap Theory

Scientific theories build bridges connecting available evidence in novel ways. They are “the glue that holds scientific observations together, for they summarize the chains of cause and effect (that help) to understand how the world works” (Hunt, 2011, p. 65). Scientific theories consider measurable variables; are objective; account for data; and, crucially, are never confirmed. We admire theories still playing the scientific game because they have not been refuted. But this must be for good reasons, not for bad ones. Let us say from the outset that we applaud Kristof Kovacs and Andrew Conway’s (this issue) effort for introducing a theoretical proposal presented under the rubric process overlap theory (POT). As properly underscored by Johnson (2013), conceptual tools are essential for making sense of both the huge amounts of already available data and the new findings derived from neuroscience and molecular genetics. POT is aimed at connecting evidence derived from psychometrics, cognitive psychology, and neuroscience in an attempt to explain one of the most replicated findings in science, namely, the positive manifold: When individuals picked at random from the general population complete varied mental tests, those with better scores in a given test tend to have better scores in the remaining tests (and vice versa). The statistical analysis of a correlation matrix comprising these mental tests produces a general factor (g). No matter how this information is analyzed, at the end of the day we will distill g if there is a truly positive manifold in the data. The interest about the nature of g dates back to the researcher who discovered this empirical fact, Charles Spearman (1904, 1927). The comprehensive book by Jensen (1998) reviews the major and minor topics related to g and discusses extensively the positive manifold at the psychometric, cognitive, biological, and genetic levels. In this regard, several points can be highlighted after his review: (a) g results from the common source of individual differences observed after resolving a variety of mental tests; (b) some tests are better measures of g than others, but the superficial characteristics of the former do not help to characterize g, (c) individual differences in cognitive abilities are remarkably greater than ability differences within a given individual; (d) psychometric g cannot be interpreted as a cognitive process or a brain feature; and (e) g might be compared with a computer’s CPU. One question of paramount relevance relates to the unitary nature of g. In this regard, Kranzler and Jensen (1991) analyzed a large battery of intelligence tests and elementary cognitive tasks failing to support the proposal that a unitary process underlies the general factor (g). However, Carroll (1991) reanalyzed their data arriving at the conclusion that “it seems parsimonious to assume that g is unitary and represents a single entity ... that influences a great variety of behaviors and performances, including speed and efficiency of information processing” (p. 434). If g is not unitary, statistical analyses should reveal several high-order factors, but this is hardly the case. Bridges must be solid enough to resist earthquakes. As we show here, POT is a suggestive and courageous bridge built over troubled water. Let’s see two examples before moving forward. According to the Spearman Law of Diminishing Returns, g explains more variance at lower levels than at higher levels of cognitive ability. Following Kovacs and Conway (this issue), this factor differentiation shows that “g is far from being a constant .... The domain-generality of the positive manifold varies across ability level” (p. 155). However, there is no Spearman Law of Diminishing Returns effect for the relationship between fluid intelligence (Gf) and working memory capacity (WMC; Gignac & Weiss, 2015; Kroczek, Ociepka, & Chuderski, 2016). Furthermore, Abad, Colom, Juan-Espinosa, and Garc ıa (2003) observed that strong differentiation effects are found when crystallized batteries are analyzed (Detterman & Daniels, 1989; Lynn, 1992). However, the analysis of fluid batteries reveals meager or null differentiation effects (Deary et al., 1996; Fogarty & Stankov, 1995). Their own study, analyzing 4,253 individuals, compared Gf and crystallized intelligence (Gc) batteries, finding a very weak effect for the former and a remarkable effect for the latter. They suggested that the higher correlations observed in the bottom half of the intelligence distribution might be a by-product of educational differences separating the low and high IQ bands, not a genuine intelligence effect. The second example refers to “goal neglect.” This cognitive function is thought to reflect a limit in WMC. Individuals must

An Integrated Utility-Based Model of Conflict Evaluation and Resolution in the Stroop Task

Cognitive control allows humans to direct and coordinate their thoughts and actions in a flexible way, in order to reach internal goals regardless of interference and distraction. The hallmark test used to examine cognitive control is the Stroop task, which elicits both the weakly learned but goal-relevant and the strongly learned but goal-irrelevant response tendencies, and requires people to follow the former while ignoring the latter. After reviewing the existing computational models of cognitive control in the Stroop task, its novel, integrated utility-based model is proposed. The model uses 3 crucial control mechanisms: response utility reinforcement learning, utility-based conflict evaluation using the Festinger formula for assessing the conflict level, and top-down adaptation of response utility in service of conflict resolution. Their complex, dynamic interaction led to replication of 18 experimental effects, being the largest data set explained to date by 1 Stroop model. The simulations cover the basic congruency effects (including the response latency distributions), performance dynamics and adaptation (including EEG indices of conflict), as well as the effects resulting from manipulations applied to stimulation and responding, which are yielded by the extant Stroop literature.

From neural oscillations to reasoning ability: Simulating the effect of the theta-to-gamma cycle length ratio on individual scores in a figural analogy test.

Several existing computational models of working memory (WM) have predicted a positive relationship (later confirmed empirically) between WM capacity and the individual ratio of theta to gamma oscillatory band lengths. These models assume that each gamma cycle represents one WM object (e.g., a binding of its features), whereas the theta cycle integrates such objects into the maintained list. As WM capacity strongly predicts reasoning, it might be expected that this ratio also predicts performance in reasoning tasks. However, no computational model has yet explained how the differences in the theta-to-gamma ratio found among adult individuals might contribute to their scores on a reasoning test. Here, we propose a novel model of how WM capacity constraints figural analogical reasoning, aimed at explaining inter-individual differences in reasoning scores in terms of the characteristics of oscillatory patterns in the brain. In the model, the gamma cycle encodes the bindings between objects/features and the roles they play in the relations processed. Asynchrony between consecutive gamma cycles results from lateral inhibition between oscillating bindings. Computer simulations showed that achieving the highest WM capacity required reaching the optimal level of inhibition. When too strong, this inhibition eliminated some bindings from WM, whereas, when inhibition was too weak, the bindings became unstable and fell apart or became improperly grouped. The model aptly replicated several empirical effects and the distribution of individual scores, as well as the patterns of correlations found in the 100-people sample attempting the same reasoning task. Most importantly, the models reasoning performance strongly depended on its theta-to-gamma ratio in same way as the performance of human participants depended on their WM capacity. The data suggest that proper regulation of oscillations in the theta and gamma bands may be crucial for both high WM capacity and effective complex cognition.

The quadratic relationship between difficulty of intelligence test items and their correlations with working memory

Fluid intelligence (Gf) is a crucial cognitive ability that involves abstract reasoning in order to solve novel problems. Recent research demonstrated that Gf strongly depends on the individual effectiveness of working memory (WM). We investigated a popular claim that if the storage capacity underlay the WM–Gf correlation, then such a correlation should increase with an increasing number of items or rules (load) in a Gf-test. As often no such link is observed, on that basis the storage-capacity account is rejected, and alternative accounts of Gf (e.g., related to executive control or processing speed) are proposed. Using both analytical inference and numerical simulations, we demonstrated that the load-dependent change in correlation is primarily a function of the amount of floor/ceiling effect for particular items. Thus, the item-wise WM correlation of a Gf-test depends on its overall difficulty, and the difficulty distribution across its items. When the early test items yield huge ceiling, but the late items do not approach floor, that correlation will increase throughout the test. If the early items locate themselves between ceiling and floor, but the late items approach floor, the respective correlation will decrease. For a hallmark Gf-test, the Raven-test, whose items span from ceiling to floor, the quadratic relationship is expected, and it was shown empirically using a large sample and two types of WMC tasks. In consequence, no changes in correlation due to varying WM/Gf load, or lack of them, can yield an argument for or against any theory of WM/Gf. Moreover, as the mathematical properties of the correlation formula make it relatively immune to ceiling/floor effects for overall moderate correlations, only minor changes (if any) in the WM–Gf correlation should be expected for many psychological tests.

High intelligence prevents the negative impact of anxiety on working memory

Using a large sample and the confirmatory factor analysis, the study investigated the relationships between anxiety, working memory (WM) and (fluid) intelligence. The study showed that the negative impact of anxiety on WM functioning diminishes with increasing intelligence, and that anxiety can significantly affect WM only in people below average intelligence. This effect could not be fully explained by the sheer differences in WM capacity (WMC), suggesting the importance of higher-level cognition in coping with anxiety. Although intelligence moderated the impact of anxiety on WM, it was only weakly related to anxiety. In contrast to previous studies, anxiety explained the substantial amount of WMC variance (17.8%) in less intelligent participants, but none of the variance in more intelligent ones. These results can be explained in terms of either increased motivation of intelligent but anxious people to cope with a WM task, or their ability to compensate decrements in WM.

Much Ado About Aha!: Insight Problem Solving Is Strongly Related to Working Memory Capacity and Reasoning Ability.

A battery comprising 4 fluid reasoning tests as well as 13 working memory (WM) tasks that involved storage, recall, updating, binding, and executive control, was applied to 318 adults in order to evaluate the true relationship of reasoning ability and WM capacity (WMC) to insight problem solving, measured using 40 verbal, spatial, math, matchstick, and remote associates problems (insight problems). WMC predicted 51.8% of variance in insight problem solving and virtually explained its almost isomorphic link to reasoning ability (84.6% of shared variance). The strong link between WMC and insight pertained generally to most WM tasks and insight problems, was identical for problems solved with and without reported insight, was linear throughout the ability levels, and was not mediated by age, motivation, anxiety, psychoticism, and openness to experience. In contrast to popular views on the sudden and holistic nature of insight, the solving of insight problems results primarily from typical operations carried out by the basic WM mechanisms that are responsible for the maintenance, retrieval, transformation, and control of information in the broad range of intellectual tasks (including fluid reasoning). Little above and beyond WM is unique about insight.