Cognitive basis of general fluid intelligence Adam Chuderski Institute of Psychology Jagiellonian University, Cracow, Poland.

Cognitive basis of general fluid intelligence

Adam ChuderskiInstitute of PsychologyJagiellonian University,Cracow, Poland

General fluid intelligence

The phenomenon of general intelligence: intraindividual unity and interindividual diversity of intellectual abilities.

Intelligence is a strongest single predictor of life success (30%),

predictor of income (~180$/1IQ),

& strong determiner of cognitive processing.

Simple task - Dual tasking

Task1: Press „<„ if both digits oddTask2: Press „>” if both letters same

B: 3A U5 D9 7EB8 1K 2D E3

R: <> <> <> <>

Two groups (N~75): SOA = 0 or 500 ms

Ind. differences in dual tasking

Complex task – insight problems

Insight - AHA! phenomenon, sudden appearance of non-obvious solution.

„How to throw a ping-pong ball that it reverses and comes back to you without bouncing any surface or being tied up to anything?”

Differences in insight problem solving

Factor studies of intelligence structure

Spearman – one general factor (g)

Cattell – fluid (Gf) vs. crystallized (Gc)

Thurstone, Guilford – many specific factors

Caroll – hierarchy of factors

Gf~g or even Gf=g

Biological basis of intelligence Brain size Number of neurons Number of synapses Myelination Size of prefrontal cortex Latency of evoked potentials Amplitude of evoked potentials Brain activity

Low correlations, complex interactions, difficult interpretation.

Cognitive basis of intelligence processing speed coping with perceptual complexity efficiency of attention short term memory capacity transfer to/from long term memory coping with novelty cognitive control metacognitive abilities using apt strategies knowledge acquisition working memory capacity (r=1.1)

Working memory is…

…a theoretical construct that refers to mechanisms underlying the maintenance and processing of task-relevant information during the performance of a cognitive task.

…a center of cognition linking perception, long-term memory, decision-making, and response generation and evaluation.

Working memory capacity measures

Task has to measure how much information is maintained in the context (during) information processing. Not just a short-term memory task.

Span tasks (reading/operation/digit span) Updating tasks (running memory, n-back) Spatial content short-term memory tasks

Structural Equation Modeling

WMC and Gf are usually computed as latent variables - indirect measures (instead of direct scores in memory tasks and intelligence tests; manifest variables) reflecting common variance in manifest variables of the same construct.

Latent variables are believed to measure „pure” WMC or Gf cleaned from task-specific noise

Conway et al., (2002)

RSPAN

OSPAN

CSPAN

WM gF

RAVEN

CATTELL

.98

Colom et al., (2005)

WM g

.96

Questions

Egg or chicken? Does Gf is a basis for WMC or it is WMC that determines Gf?

Is the link really so strong? Ackerman metaanalysis – 25% of common

variance Oberauer metaanalysis – 75% of common variance

Working memory or short term memory According to Ackerman – no difference According to Engle – big difference

What process underlies WMC?

WMC is a statistical construct, difficult for cognitive interpretation.

Why one has high WMC. Because of:

• high processing speed (Jensen)

• efficient attentional control (Engle)

• capatious adjustable focus of attention (Cowan)

Attentional (cognitive) control

Switching attention to proper content/processes, inhibiting interfering content, and monitoring, updating, and maintaining info in WM.

Significant differences in attentional control between high- and low-WMC subjects, e.g.:

-weak antisaccades of low-WMC Ss

-no coctail party effect of high-WMC Ss

Evidence for Gf-control link (Paulewicz et al., 2007)

Evidence for adjustable attention hypothesis (n-back task)

The need for computational models of individual differences in cognition

… the study of WMC-IQ correlations should be accompanied by computational modeling of the underlying cognitive processes, something that has been virtually completely absent to date. Of course, this call is tantamount to a call for the development of a complete computational process model of not just memory but also intelligence itself—clearly a big task whose completion time will be measured in decades, not years. Lest one think that this goal is too ambitious to be even considered at this moment in scientific history (Lewandowsky & Heit, 2006).

Mathematical models

Oberauer and Kliegl’s (2001) model of WM impairment in aging.

Probability of retrieval = 1/(1+exp(-A–t)/s)

A – item activation dependent on:

resource, decay rate, maximum number of chunks, cross-talk, interference. Last one wins.

Types of computational models of individual differences in cognition

symbolic vs. subsymbolic vs hybrid

of groups vs. of individuals

qualitative (structural) vs. quantitative (parametric)

Examples of computational models of individual differences in cognition

Quantitative & group – FAIR/BETTERRAVEN model (3CAPS; Carpenter, Just & Shell, 1990)

Qualitative & group – CC Reader model (Just & Carpenter, 1992), clinical disorders

Qualitative & individual – model of digit span (ACT-R; Daily, Lovett, & Reder, 2001)

Cross task modeling of Control -> Gf – a new project

ACT-R models of individual differences in 4 executive tasks (span task, updating, inhibition, switching). Fitting different parametrizations to non-obvious patterns of experimental data. The relevance of different parameters for generating differences can be evaluated Individual parameters are copied to the model of analogy making (Gf test) The 0-parameter cross-task prediction (executive tasks->analogy) is tested

0-parameter cross-task prediction

WM TaskGf Test

(Analogy)correlation

Model WM Task

fitting(χ2)

Model of Gf Test

copying the parameters

prediction(χ2)

individual parameters

optyma-lization

Conlusions

Not much is known yet on the direction and nature of relation between parameters of human cognitive architecture and intelligence.

Computational modeling may help identify the most promising candidates of architectural features for further investigation.

The role of working memory and control mechanisms of cognitive processes during ‘intelligent’ processing may be the key issue.