This article provides an introduction to the Gf-CODE cognitive training approach that I am adopting with my IQ Mindware apps.
.
It is inspired by a number of brain training studies published in late 2018 and 2019. Two of these are meta-analyses looking at all relevant studies to date. Meta-analyses enable us to draw ‘state of the art’ conclusions.
.
Gf-CODE cognitive training is designed to target a proposed underlying neural ‘machine language’ of executive cognition and fluid reasoning for rapid, flexible learning and the ability to adapt this learning to new, unfamiliar situations in decision-making, problem solving, comprehension and skill acquisition.
.
In Part 1 I’ll explain what all brain training aims at – far transfer. Then I’ll explain what Gf-CODE training is, and how it is proposed to enable reliable far transfer. In Part 2 I’ll briefly review the 2018-2019 evidence for this framework.
.
PART 1
.
Far Transfer: The Point of Brain Training
What train training apps developed by scientists in labs are designed to do is result in transfer of the training on the game to cognitive skills that you can use in real life – for example memory, intelligence/aptitude tests, general problem solving, decision-making, attention control, learning efficiency, and so on.
.
This kind of transfer is called far transfer – and is contrasted to where your training has benefits only with closely related games, called near transfer.
.
Here is an illustration of the far transfer concept. In this case training on evidence-based apps leads to improved performance in an IQ/aptitude test and an oral presentation of complex information.
.
.
The Gf-CODE framework provides a unifying account of effective, far-transfer brain training for learning and intelligence (g).
.
Well-designed brain training apps target the the ‘atomic’cognitive functionsand processes within our brain’s ‘central processor’. These core functions are shown in the diagram below (don’t worry about the details for now!).
.
This system constitutes our fluid intelligence (Gf) – our capacity to grasp changing situations, learn rapidly and apply our learning in novel situations. All directed learning and intelligent cognition engages these functions and processes in varied ways. The more efficient and coordinated this system is, the higher our intelligence will be.
.
Working Memory and Executive Processes
.
Working memory (the light grey box) is our mental workspace – the information we can hold in mind at any given time to reason, solve problems, make decisions, comprehend and learn, whether cognitive (e.g. math) or motor skills (e.g. playing the piano or learning BJJ). Input gating is what information our attention selects for this workspace, focusing, operating and disengaging is what we do with the information in the workspace under attentional control, and output gating is what we make use of from this ‘thinking’ to guide our decisions and actions. ‘Mindware’ are the task-specific rules and strategies we make use of to guide performance.
.
Machine Code Analogy: Hence Gf-CODE
The Gf-CODE which coordinates information flow between the elementary functions shown in the diagram is like machine code in a computer. Niels Taatgen explains the concept in his seminal work The Nature And Transfer Of Cognitive Skills.
.
At the lowest level, a computer program is built out of machine code. Machine code consists of a fixed set of instructions, each of which performs primitive actions within the computational system. The function these instructions have to perform is to move around information between the different components in the computer. …With a small and finite set of instructions, computers are capable of implementing any algorithm.
.
As with a computer, cognition has several specialized modules for perception, motor responses, and memory which are coordinated by atomic “machine level” instructions – the Gf-CODE.
.
Neural Basis of the Gf-CODE
The neural basis of the Gf-CODE is in the coordinating functions of the core executive network hubs of the dorso-lateral prefrontal and parietal cortices: the frontoparietal network (FPN).
.
There are many networks in the brain, but while many of these cannot communicate with each other, this FPN hub network is designed because of its global connectivity to be able to flexibly transfer information, knowledge and skills effectively between all the networks for intelligent, adaptive thought and action throughout the brain. Cognitive neuroscientists call this ‘compositional coding’ (review).
.
Frontoparietal network: compositional coding
.
The Gf-CODE and Psychometric Intelligence (g)
Drawing on the research of Kovacs & Conway’s (2016) in their Process Overlap Theory (POT), I propose that the flexible hub FPN is the neural basis of fluid intelligence (Gf), which has functional overlap with the other broach abilities of IQ. Gf is highly correlated with full scale measures of IQ (g), and measures of the Gf processes (shown in the workspace diagram above) are related psychometrically to all the broad abilities of IQ shown below.
.
.
How Gf-CODE Training Works
At the highest level, attention coordinates each of the atomic Gf functions in the box and arrow ‘workspace’ diagram above, and each of the functions is associated with a simple type of cognitive skill or procedure.
.
Effective cognitive training doesn’t focus on just one of these functions but on all of them. At the minimum it focuses on the right combinations of them for particular types of cognitive skill. The trouble with standard dual n-back (DNB) training is that it only targets a limited number of these functions but not others. For instance, DNB targets attention shifting and updating, but not output gating or disengaging. And it turns out that output gating and disengaging are more closely linked to reasoning and problem solving – as measured by IQ tests. You need to target these functions in your brain training to improve your reasoning and problem solving.
.
Effective training can also open bottlenecks in the information flow of these functions – where you may have a blockage in the overall flow of processes – such as in disengaging or output gating. There is considerable evidence, as we will see, that it is largely these kinds of weak links that differentiate between higher and lower IQ individuals.
.
Large Overall Efficiency Gains from Small Efficiency Gains
Here is recent meta-analysis data for all relevant n-back studies to date showing small dual n-back training gains on working memory capacity (Gwm) equivalent to about 4 standardized points.
.
Effect sizes of working memory training
.
Note that the far transfer training gain here is very real! (As are the training benefits for fluid intelligence and cognitive/attention control). The claim there is no dual n-back benefit for intelligence is simply untrue.
But the benefits are clearly small – compared to e.g. the near transfer benefits for untrained n-back tasks.
But let’s think like an engineer.
.
From an efficiency point of view, it’s true that:
.
the overall efficiency of a system is equal to the product of efficiencies of the individual subsystems or processes (ref).
.
Most brain training studies look at the efficiency gains from training of just one or two functions (e.g. the ability to focus and maintain information in working memory as measured by working memory capacity (Gwm)).
.
We can interpret dual n-back here as training one subsystem of entire system underlying fluid intelligence performance.
.
Let’s imagine that DNB training only results in an efficiency gain of e.g. 5% for the ‘focus’ (maintaining items in the workspace) subsystem. But if we compare the overall efficiency gain for fluid intelligence of this kind of training with the overall efficiency gain if we improve each of 5 subsystems by 5 (including disengaging and output gating) the difference can be very striking – perhaps a 50% or greater performance improvement.
In this way, what some interpret as merely ‘fine-tuning’ in the efficiency of established processes such as working memory focus/maintenance, by collectively, result in very large efficiency gains for overall fluid intelligence.
.
Bottlenecks & Efficiency Gains
If we factor in potential bottlenecks in multi-component processes. Bottlenecks are by definition the causes of high inefficiency, and gains can be very large from brain training that releases such bottlenecks. Bottleneck clearing operates like the Pareto 80:20 principle, where e.g. 20% brain training effort can result in 80% training benefits in overall cognition. Here we have a little work for large gains.
.
When there are no bottlenecks in your Gf-CODE, the training helps with small incremental gains (a lot of work for small gains) for each component, but overall based on the multiplier effect you will see impressive performance benefits.
.
Brain Training as Both Gf Efficiency Gains and Learning to Learn
Effective Gf-CODE brain training results not only in efficiency gains in the ‘atomic’ functions and processes of Gf as described above, but also the ability to rapidly learn from tasks by abstracting more general task performance skills that can then be applied in new contexts to new tasks (far transfer), while- with practice -becoming less attention-demanding, more automatic and effortless.
.
In other words, Gf-CODE training can improve rapid instructed learning (our ability to rapidly reconfigure our minds to learn new rules in new contexts), and once established, to transfer new task-skills to other tasks with similar structures. With practice these general purpose skill sets becomes less attention demanding, less effortful and more automated, improving performance.
.
Greater efficiency of our basic fluid intelligence (Gf) system thus allows us to learn more efficiently and apply our learning more flexibly. Effective brain training thereby allows us to better learn how to learn.
.
Cognitive Skills AND Motor Skills: Embodied Cognition
From the most general level processes of fluid intelligence (e.g. input gating, focusing, disengaging, output gating, etc) to more specific skill sets learned for certain types of tasks, we can look at brain training as improving or acquiring skills– and these skills have a code that can be utilized both in cognition (as in math or language comprehension) and in your decision-making actions, and motor skills (as in musical instrument playing or sports). As a 2019 review paper recently states:
.
Brain training is the acquisition of cognitive skills akin to learning motor skills.
.
What is exciting about Gf-CODE training is that because the same Gf functions and processes served by the same frontoparietal hub networks – and the same underlying Gf code – are involved in both academic/symbolic and action/motor based skills, benefits are seen across the board, whether you are doing an IQ test, reading music, fixing an engine, or learning a new sport. The training is fully embodied in action if it needs to be.
.
PART 2
.
Premise 1. Far transfer brain training depends on learning new skills or routines at high levels of abstraction – independent of specific applications of those skills.
.
Gf-CODE training involves learning cognitive skills which enables far transfer.
.
Research Papers
.
.
This University of Cambridge study is a meta-analysis looking at multiple brain training studies to find the the real underlying effects at play.
.
The researchers tell us how to get substantial transfer:
.
substantial transfer from [brain] training is a consequence of the development of new routines that are applied to new tasks. It is a form of learning that follows well-established principles of the acquisition of complex skills.
.
This idea is based on Taatgen’s seminal paper The Nature And Transfer Of Cognitive Skills. Taatgen talks of high level ‘machine code’ as the basis of far transfer: hence Gf-CODE training. To quote:
.
The central idea is that skills are broken down into primitive information processing elements that move and compare single pieces of information regardless of the specific content of this information. A learning process therefore combines the elements in increasingly larger units. If there is overlap between tasks, this means the larger units learned for 1 task can be reused for the other task, producing transfer.
.
So what are these ‘information processing elements’ that can be combined?
.
Premise 2. We can selectively train the sub-processes of the Gf / working memory system – the highest level elements of the ‘machine code’ of cognitive performance.
.
As already shown above, here is my model of the executive function sub-processes identified in the last few decades of cognitive neuroscience research. Supporting research can be found in my review here.
.
.
Training combinations of these elements is built into IQ Mindware app programs.
Research Papers
.
This research team looked specifically at the effects of training attention control for either input or output gating in the model above. They conclude that the transfer from the training:
.
is [based on] the dynamic structure of the task. Collectively, our results highlight the importance of WM gating policies in particular, and control policies in general, as a key component of the task knowledge that supports flexible behavior and task generalization. ….Abstract knowledge about the tasks we encounter enables us to rapidly and flexibly adapt to novel task contexts.
.
.
There is known to be a link between our working memory capacity (WMC) and our fluid intelligence (Gf) – as measured by IQ tests. According to research summarized in the 2018 review above, there are two ‘information processing elements’ or core functions involved (both shown in the Gf-CODE model above), coordinated by attention – also in the model. To quote:
.
“the ability to maintain access to critical information and the ability to disengage from or block outdated information. In the realm of problem solving (Gf), high working memory capacity allows a person to represent and maintain a problem accurately and stably, so that hypothesis testing can be conducted. However, as hypotheses are disproven or become untenable, disengaging from outdated problem solving attempts becomes important so that new hypotheses can be generated and tested.” (ref)
.
WMC tasks depend relatively more on the maintenance function while Gf tasks depend relatively more the on the disengagement function – e.g. clearing from your mind any hypotheses that you have thought through but ruled out as you think through a problem, shown in this diagram from the 2018 review paper below.
.
So TRAINING the disengaging function – as in IQ Mindware apps – is more likely to transfer to fluid intelligence gains than training working memory capacity (the dual n-back).
.
From Shipstead & Engle, 2018
.
Premise 3. A core ‘hubs’ network (the FPN) centered on the dorsolateral prefrontal cortex, is targeted by Gf-CODE training, and this network underlies both cognitive and motor skill learning.
An exciting implication of the Gf-CODE cognitive training framework is that because the same Gf functions and processes served by the same FPN network – and the same underlying Gf code – are involved in both academic/symbolic and action/motor based skills, benefits are seen across the board, whether you are doing an IQ test, reading music, fixing an engine, or learning a new sport. This is an example of ’embodied cognition’.
.
Research Paper
.
According to this meta-analysis study of all fMRI brain imaging studies on working memory training to date:
.
Comparisons with perceptual-motor (PM) learning revealed that WM training engages domain-general large-scale networks for learning …Also the dynamics of the training-induced brain activation changes within these networks showed a high overlap between WM and perceptual-motor training. …Overall, our findings place WM training effects into a general perception-action cycle, where some modulations may depend on the specific cognitive demands of a training task.
.
Summary
In Part 2 I have briefly reviewed the state-of-the-art (2018-2019) evidence for the Gf-CODE training framework, which I am now implementing in my IQ Mindware apps and coaching programs. This framework – I believe – represents an exciting new wave in brain training moving forward.
.