Landmark study, good experiment design. This paper basically says that you can train your fluid intelligence, which is far more important than crystallized intelligence. In light of the earlier post I made, training working memory is essential, but it is not enough to stop there (and this paper affirms it). There are evidence for focus and other factors.
It’s possible that kids who saw greater gains in fluid intelligence were those who started out at lower ability levels and simply had more room to improve.
Well, in Denmark, the Flynn Effect was found to be almost entirely the left tail shifting towards the mean, which is precisely what you’re asking about here.
Randall Engle found that improving working memory using complex span task was possible, but that fluid intelligence was largely unaffected. Does this study add to that? (haven’t read it yet)
I was satisfied with this answer
Summary.
- Fluid intelligence is trainable.
- The training and subsequent gains are dose-dependent—meaning, the more you train, the more you gain.
- Anyone can increase their cognitive ability, no matter what your starting point is.
- The effect can be gained by training on tasks that don’t resemble the test questions.
… These five primary principles are:
- Seek Novelty
- Challenge Yourself
- Think Creatively
- Do Things The Hard Way
- Network
===
I think, typical mind training games do not offer these 5 principles. The result depend on inner and outer challenges 
and individual approach might work better than mass approach
and I like to read this type of articles
http://www.brainy-child.com/article/thinking-skills-for-children.shtml
http://www.brainy-child.com/article/critical-thinking-skills.shtml
Sorry, I haven’t been able to reply in the last few days. Frukc’s article summarizes the study pretty well: it is a landmark study that really proves that fluid intelligence can be trained and is dose-dependent.
By the way, the Dual-N-Back exercise is available for download here. The webpage currently features 4 different replication studies that have been done since the publication of the paper. It is replicable.
Back to the write-up. The blog article that Frukc linked had 5 points (see Frukc’s answer) that directly follow the paper. However, I would caution that these points (seek novelty, self challenge, creative thinking, do things “the hard way”, and “network”) are highly informal and probably may never be directly testable through research. Also, some of the papers cited for plausibility are somewhat tenuous—at least to me.
What we know so far is that Gf (fluid intelligence) is an amalgamation of many different skills, with working memory as the nexus (or perhaps the essential “reservoir”). Such skills are controlled through “central circuitry” called the “executive function” (EF), whose definition is still not yet settled (though there are several models). To increase Gf, we will need to increase the efficiency and the capacity of the EF, for sure, in addition to increasing the capacity and efficiency of the working memory. The problem is: How? That’s especially true for the first part (EF training). Once we are exposed to brain-training exercises, we somehow develop some strategies against them and we execute them over and over again. As such, these exercises lose their novelty quickly—I think this is why previous papers show that there are no evidence in brain-training exercises improving Gf. This is why novelty requirement (and therefore self-challenge and do things the hard way) kicks in—it forces our brain to grow. The problem is how to define novelty: What is novel? Is synthesis over several different things enough to be considered as novel? I guess that the novelty context varies amongst individuals and the novelty that forces the brain to grow is what we are seeking.
On “Thinking Creatively”: The cited Sternberg paper (this one) is quite interesting. Especially the “What is teaching for successful intelligence” section. He outlines several key ideas:
- Teaching for memory learning (Recall, recognize, match, verify, repeat)
- Teaching for analytical learning (Analyze, evaluate, explain, compare/contrast, judge)
- Teaching for creative learning (Create, invent, explore, imagine, suppose, synthesize)
- Teaching for practical learning (Put into practice, use, implement, apply)
I can see that these teaching ideas would work (and he shows that they work by citing previous papers). The main idea is banking on the memory training and force the EF to organize around the memory reservoir. Since it somehow resonates with some EF model (esp. the Working Memory Model), such strategies would work. However, it is not sufficient to stop right there—especially more so since we have quite a number of EF models defined (see wiki). For sure we would need emotional and social intelligence since EF also involves self-inhibition / regulation. On top of that, there are attention / focus, task switching, etc. However, there’s a paper that hinted that all these are manifestations of the same root ability, I am still a little skeptical about it (primarily about the analysis method—they used a version of factor analysis, which can be problematic to explain).
Anyway…
So, to conclude, I think the consensus on training Gf (and EF) is still not there yet. The five points are a good start.