Hey guys, latest talk of Jaeggi, talking about "brain training", n-back, fluid intelligence and transfer effects.
http://vimeo.com/65064957
Details/Slides of the issues talked:
The Issue
How can we promote the ability to reason and to solve new problems
that require more than just the retrieval of previously acquired
knowledge?
Commonly called "fluid intelligence" (Gf) [sic]
Contrasted with "crystallized intelligence" (Gc) [sic]
Many well-respected psychometric measures with high intercorrelations
among them.
Assumed to be highly heritable.
Highly predictive for educational and professional success.
=================================
Types of Intelligence - Examples
Crystallized (gC)
General knowledge
What is the capital of France?
How much feet [sic] in a mile?
WHo was Mahatma Gandhi?
Language / Vocabulary:
What does 'discernible' mean?
What does 'being smart' mean in German?
Fluid (gF)
Inferences:
Bill, a member of the basketball team, is 6 feet, 2 inches tall and
weights 195 pounds. To qualify for the team, a person must be at least
5 feet, 10 inches tall.
1. The larger a man is, the better basketball player he is.
2. Basketball players are often underweight.
3. Some players on the team are more than 6 feet tall.
4. Bill is larger than the average man.
5. The best basketball players come from the ranks of larger-than-
average men.
Matrices (see next slide) [noted among "spatial measures"]
=================================
Measuring Fluid Reasoning
An Example
[image can be seen in video at 5:44-7:04]
=================================
Measuring Fluid Reasoning
Another Example (easy item)
[image can be seen in video at 7:04-7:56]
=================================
Measuring Fluid Reasoning
Another Example (hard item)
[image can be seen in video at 7:56-8:52]
=================================
Improving Intelligence - An Overview
[noted to be a "very old topic"]
Flynn effect ["the observation that if you look at poeple across a
generation usually what people see is that people's performance in
intelligence measuring tend to increase over time"; various
explanations are mentioned]
Head start and related projects (e.g. Abecedarian, Perry school)
[notes that it has been observed that the initial improvements in I.Q.
tend to "fade away" in later years, but that there are "important long-
term effects"]
Reasoning/Problem Solving/Enrichment training:
Baltes et al, 1988
Berman Nutley et al, 2011
Feuerstein, 1979, 1981
Guinagh, 1971
Hayslip, 1989, 1995
Herrnstein et al, 1986
Klauer et al, 2002, 2008
Mackey et al, 2010
Reese et al, 1976
Tomic & Klauer, 1996
Tranter & Koutstaal, 2007
Sanz de Acedo Lizarraga et al, 2003, 2009
Stankov (Kvashchev), 1986
[notes that use of Raven's items and improvement on such tests would
be criticized as "teaching them to the test" / "task specific"
learning]
=> Demonstrating transfer after cognitive training has been difficult
(e.g., Caruso, Taylor & Detterman, 1982, Detterman, 1993, Saloman &
Perkins, 1989, Shipstead et al., 2012).
[notes that there are some conclusions that suggest it is difficult to
find transfer from tasks that are very different might not even be
possible, if not merely difficult]
[time in video @ ~13:00]
=================================
Improving Intelligence - An Overview
Nevertheless, there is accumulating evidence that certain targeted
cognitive interventions lead to transfer
cf. Morrison & Chein, 2011, Buschkuehl & Jaeggi, 2010, Buschkuehl,
Jaeggi & Jonides, 2012 for reviews
Training on Attention, WOrking Memory, and/or Executive Control:
Anguera et al, 2012
Basak et al, 2008
Borella et al, 2010
Buschkuehl et al, 2008
Carremi et al, 2012
Green et al, 2012
Jaeggi et al, 2008, 2010, 2011
Jausovec & Jausovec, 2012
Kerns et al, 1999
Klingberg et al, 2002, 2005
Karbach & Kray, 2009
Loosli et al, 2012
NOvick et al, in press
Qin et al, 2009
Roughan & Hadwin, 2011
Rueda et al, 2005, 2012
Rudebeck et al, 2012
Schmiedek et al, 2010
Schweizer et al, 2011
Shalev et al, 2007
Stepankova et al, in revision
Witt, 2011
Zhao et al, 2011
Zinke et al, in revision
[notes that these seem to suggest that transfer to other long-term
measures is possible]
[time in video @ ~13:45]
But see also Redick et al, 2012, Chooi & Thompson, 2012
=================================
Working Memory
A cognitive system responsible for storing and manipulating a limited
amount of information for a short time.
Examples:
multiplying 56 by 13 in your head
reading a complex paragraph
taking notes while following a lecture
Predicts performance on many cognitive tasks (including intelligence,
reading comprehension, math....)
Related to academic achievement
[notes that for designing interventions WM is more predictive towards
academic achievement than intelligence for children]
[time in video @ ~15:30]
Alloway et al, 2009 (Child Development)
Jonides et al, 2008 (Annu Rev Psychol)
Pickering, 2006
Shah & Miyake, 1999
=================================
The Optimal Training Paradigm
Our Rationale
*the training should engage skills related [sic] to working memory
-> underlying mechanism for many cognitive tasks
*the training should require the use of multiple processes
-> shared processes, overlapping neural networks
*the training should discourage the generation of task-specific
strategies
-> engagement of executive processes
[contrasted with digit span specializers; training should "keep them
on their toes"]
*the training difficulty should match participants' capacity
->adaptivity
[notes that if it is too difficult/demanding, people might disengage,
become disinterested, and just give up; if it is too easy/automatic,
then there is a risk of using merely task-specific strategies]
[time in video @ ~18:30]
Jaeggi et al, 2008 (PNAS)
Dahlin et al, 2008 (Science)
Lustig et al, 2009 (Neuropsychol Rev)
================================
The N-Back Task
[as a task that would fulfill the above criteria - notes that they
would use the dual n-back task]
[example image of a 2-back task can be seen at 18:47-20:11]
[time in video @ ~20:00]
e.g. Jaeggi et al, 2007, 2009, 2010, 2011, 2012
================================
N-back Task - Example
Get ready for 2-back!
[proceeds to give a brief demonstration of the visual-spatial variant
of the task]
[time in video @ ~20:30]
================================
Behavioral and Neural Effects
[images show (she describes): accuracy (%) drops with increased n-back
and increased number of simultaneous modalities; WM network: activity
in DLPFC, Superior Parietal lobes, and some cingulate activations;
higher the n-level the higher the activation]
[time in video @ ~22:30]
================================
Neural Correlates of Proficiency
[images described below]
High Performer - fairly minimal activation
very good at distributing their resources / efficient
Low Performer - fairly extensive and high activation
[time in video @ ~23:30]
=================================
Training Study Designs
[images described below]
Pretest (Gf) [sic]
image of matrix items
Train on Dual n-back
Posttest (Gf) [sic]
Control Group, active & passive
Training can also implement the audio or visual versions of n-back
separately
[time in video @ ~24:45]
=================================
Improvement in N-back
[image of Mean n-back level vs Traning session, showing general
improvement, starting at 3-back]
[asks what others think what the highest n-level someone could achieve
at DNB - answers 14-back is the highest record]
[time in video @ ~27:00]
=================================
Transfer to Gf [APM / BOMAT] - Young Adults
(Dual N-back Training)
[image of increase in Matrix Reasoning from Jaeggi, Buschkuehl et al
2008 PNAS]
[image of Dose-Response Curve also from same source]
[time in video @ ~29:00]
=================================
Transfer to Gf [RAPM / BOMAT] - Young Adults II
(Dual vs. Single N-back Traning - Spatial)
[image shows spatial single n-back was more effective from source:
Jaeggi et al, 2010, Intelligence]
[notes that they would prefer to train those "who are in need of
training"]
[time in video @ ~30:30]
=================================
Transfer to Gf [WAIS Block Design & Matrix Reasoning] - Old Adults
(Single N-back Training - Verbal)
[image of Increase in Spatial Ability, with standardized Gf [sic]
Score vs pre and post]
[image of Dose-Response Curve for Standardized gain vs Training
length]
[time in video @ ~32:00]
Stepankova, Lukavsky et al, in revision
=================================
Michigan Training Study I
Can we replicate the previous findings with an active control group?
[notes that it is a non-trivial issue to find an active control group
engaged in a task where they have a reasonable chance of "getting
smarter" from that task]
Sample: N=44 (20 women); mean age: 21 years (SD = 3.0)
Transfer tasks:
Near transfer:
Single n-back
Dual n-back
Working Memory:
OSPAN
VAC
Other:
ANT
DST
Fluid Intelligence:
APM
BOMAT
Card rotation
Letter sets
[time in video @ ~34:30]
Anguera et al, 2012, BBR
=================================
Transfer Effects - but not on Gf
[image of Overall Transfer: Standardized Gain vs either Dual N-back or
Knowledge Trainer; notes that the increased gain for DNB training
group was due to the working-memory measures included in the composite
of the Overall Transfer]
[image of Transfer to Gf [sic] notes that there is no difference
between groups]
["scratched our heads"]
[time in video @ ~35:30]
Anguera et al, 2012, BBR
=================================
Why no Transfer to Gf?
[image of plot: Study Payment in US$ vs either Bern, Taipei, Prague,
or Ann Arbor, where each provided higher and higher payout that the
last in increasing order]
Extrinsic reward can undermine people's intrinsic motivation.
If extrinsic reward is crucial then [sic] its influence should be
visible in our data.
[time in video @ ~37:00]
Murayama et al, 2010, PNAS
Deci et al, 1999, Psychol Bull
=================================
Training Performance as a Function of Incentives
[image of plot for Average N-Back Level vs Session # with Bern
(unpaid), Taipei (unpaid), and UofM (paid), noting little difference
among groups]
[no source provided for image, assumed to be Anguera et al, 2012, BBR]
[time in video @ ~38:00]
=================================
Motivation & Engagement
[image of Motivation for Rating vs Week # with slight downard trend -
"clearly drops"]
[image of Engagement for Rating vs week # with linear downard trend]
Note: self-reported
[time in video @ ~39:00]
Anguera et al, 2012, BBR
=================================
Refined Michigan Training Study
->Participants did not receive any payment
Sample: N = 78 (36 women); mean age: 25.2 years (SD = 6.5)
Interventions:
Dual N-Back, Verbal Single N-Back, Knowledge Trainer
Transfer tasks (2 composites based on pre-test):
Verbal Gf [sic]:
Inferences, Reading Comprehension, Verbal Analogies, CFT
Visuospatial Gf [sic]:
BOMAT (untimed), Raven's APM (untimed), Surface Development, Form
Board, Space Relations
[time in video @ ~40:00]
Jaeggi et al, in revision
=================================
Motivation & Engagement
[now differentiates between paid and unpaid participants, where
Motivation shows only a slight increase in motivation over 4 weeks
(for unpaid vs paid); where Engagement shows no decline but overall
higher value (for unpaid vs paid)]
[time in video @ ~40:30]
Anguerra et al, 2012, BBR
Jaeggi et al, in revision
=================================
Transfer to Gf
(Dual vs Single N-back Training - Verbal)
[image of Standardized gain for Verbal Factor and Visuospatial Factor
(separately), across three groups: Active Control Group, Single NB
(auditory), and Dual NB (auditory + visual); where Verbal Factor shows
no significant improvement; where Visuospatial Factor shows greater
improvement for SNB and DNB groups]
[notes they have not determined why there is no demonstrable
improvement for the Verbal Factor]
[notes that in these reasoning measures, there are not always
standardized "parallel versions" available, forcing them to cut tests
in half, meaning that the the tests become unreliable (due to fewer
items)]
Jaeggi et al, in revision
=================================
Payment and Transfer
(only 4-week n-back training)
[image Study Payment in US$ vs studies with or without transfer to Gf
[sic]; suggests that higher payment will increase chance for no
transfer to Gf; can be seen at 42:50-43:33]
=================================
Transfer to Gf as a function of payment
[image of Effect Size (averaged) vs paid (n=192) and not paid (n=227),
where not paid has a higher Effect Size than paid]
Note: Data are coming from 14 studies using a total of 52 effect sizes
(the Gf measures were verbal and visual, ranging from APM to BETA-III,
with APM being the most prevalent).
=================================
N-back for Children
[image of lilypads and frog for four trials in a task of 1 round = 15
+ n-trials]
[time in video @ ~45:00]
Jaeggi et al, 2011, PNAS
=================================
[presents demonstration of the task]
[time in video @ ~46:00]
=================================
Knowledge Trainer
Key Features
A crystallized knoweldge intervention
Contains general knowledge, vocabulary, and some trivia questions.
Once a question is answered correctly, it is no longer asked (thus,
the game is always increasing in difficulty).
[time in video @ ~46:30]
=================================
Training Features
To maximize motivation and compliance, the tasks were designed based
on research that features of video game that make them engaging.
[frog and lilypads, pirates, rockets and aliens, vampire and castle
with bats, treasure chest with gold; all of which are essentially
visuospatial in nature]
points
high scores
bonus rounds
real prizes
different themes and story lines
[time in video @ ~47:30]
Gee, 2007
Malone & Lepper, 1987
Prensky, 2001
Squire, 2003
=================================
Transfer Task - WOrking Memory
Object 2-back task
[cactus, flower, pumpkin; includes lure trials]
[time in video @ ~48:30]
3 rounds with 20+n trials on 2-back level
Task duration: approx. 10 minutes
=================================
Transfer Task - Attention
Continuous Performance Test
[letters in 1s, 2s, 4s intervals; hard for kids with ADHD; "it's a
very boring task"]
Go vs no-go ratio: 90:10
18 blocks with 20 trials each (= 360 trials in total)
Task duration: approx. 20 minutes
[time in video @ ~49:30]
after Conners et al, 2003
=================================
Transfer - Attentional Control
[image of 2-back task (non-trained) for False Alarms (proport.) vs
pre, post, and follow-up for 3 months after training on nback (N=29)
and KT (N=27); fairly clear trend of nback group increasingly making
fewer and fewer errors than KT]
[image of Continuous Performance Test for Percent Errors vs pre, post,
and follow-up for 3 months after training; similar but not as drastic
trend]
[time in video @ ~50:00]
Buschkuehl, Jaeggi et al, in preparation
=================================
Transfer - Attentional Control
Reaction Times
[image of 2-back task (non-trained) for RT vs pre, post, follow-up
{same as previous slide}; fairly clear trend that nback group became
slower "more careful"]
[image of Continuous Performance Test for RT vs pre, post, and follow-
up {same as previous slide}; same trend as 2-back task - slower RT for
nback group]
[time in video @ ~51:30]
Buschkuehl, Jaeggi et al, in preparation
=================================
Individual Training Differences
[image of Average n-back level vs Session; some children clearly
improve, others not at all]
[time in video @ ~52:00]
Jaeggi et al, 2011, PNAS
=================================
Individual Transfer Differences
[image of Immediate Transfer on Reasoning for Standardized gain vs
Small training gain, Large training gain, Active control; STG shows no
standardized gain on Reasoning, AC shows a slight gain, and LTG shows
a large gain]
Jaeggi et al, 2011, PNAS
[image of Immediate Transfer on CPT [Continuous Performance Test] for
Standardized gain vs STG, LTG, AC, but where STG outperforms AC, but
still well below LTG]
Buschkuehl, Jaeggi et al, in preparation
=================================
Relations of Gf and CPT Gains
N-back Group
Gain in CPT vs Gain in Gf => r = -.30 ["inverse relationship"]
Control Group
Gain in CPT vs Gain in Gf => r = -.06 ["no relationship"]
[notes that the apparently improved ability to deal with interference
may be one of the mechanisms to explain the apparent improvement in
fluid intelligence measures]
[time in video @ ~53:45]
Buschkuehl, Jaeggi et al, in preparation
==================================
Parent Questionnaires
(ADHD Symptoms)
[image of N-back and Active Control for CBCL ADHD Score vs Pre and
Follow-Up; where N-back group has a slightly decreased score at Follow-
Up, whereas AC shows no improvement in ADHD score]
[image of correlation for Change in COnner's Oppositional Behavior vs
Training Gain (N-back) with r = -.48; after 3 months training]
[time in video @ ~55:00]
==================================
Summary
There is accumulating evidence for improved performance in Gf measures
after targeted cognitive training.
-> even more evidence for near transfer effects.
The effects have been demonstrated throughout the lifespan - in young
adults, in old adults, in typically developing children, as well as
children with ADHD and other deficits.
Improvements in fluid intelligence seem to rely on improvements in
interference resolution (recognizing and avoiding lures) and
impulsivity, perhaps not on WM "capacity".
There are certain moderating and mediating factors that might drive
any transfer effects.
[time in video @ ~56:30]
==================================
Potential Moderators & Mediators
Training time
(Jaeiggi et al, 2008; Stepankova et al, in revision; Basak et al,
2008)
Training success
(Chein & Morrison, 2010, Jaeggi et al, 2011; Rudebeck et al, 2012;
Zhao et al, 2011; Von Bastian & Oberauer, in revision; Stepankova et
al, in revision)
Pre-existing individual differences
(Jaeggi et al, 2008, 2011, in revision; Rudebeck et al, 2012; Zinke et
al, in revision)
Age
(Schmiedek et al, 2010; Zinke et al, in revision; Borella et al, in
revision)
Personality
(Studer-Luethi et al, 2012)
Motivation / Beliefs
(Anguera et al, 2012; Jaeggi et al, 2012, in revision)
[time in video @ ~58:30]
==================================
Conclusions & Future Directions
The question should not be whether but for whom and why cognitive
training works.
-What are the underlying cognitive mechanisms?
Differential training interventions
-What are the underlying neural correlates?
-For whom does it work best?
Individual differences
-What needs to be done in order to make the effects stronger?
adaptivity, spacing, scaffolding, group vs. individual trining,
amount of interference, motivational features
-How can we best maintain effects (e.g. booster sessions)?
-To what extent do the improvements affect academic achievement and
other real-life consequences?
==================================
Have fun.
Forum 
