working working memory with dual n-back

May 9, 2010 2:30 PM

dual n-back is a simple working memory game of unbounded difficulty.

The phrase crystallized intelligence describes static or systematic knowledge that is accumulated piece-by-piece over the course of a lifetime, career, or practice within a specific subject. Methods to improve it abound; standardized tests, study groups, flash cards, rote learning, and highly specialized apprenticeships could all be understood as mechanisms to promote crystallized intelligence. But fluid intelligence has proved difficult to develop directly, as training methods which promote the development of working memory capacity often fail to offer generalizable results. In response, 2008 study explored methods to improve fluid intelligence (previously mentioned on mefi) via a short-term memory game of unbounded difficulty, dual n-back, in which players must remember increasingly long series of positions and sounds. Study participants who played the game for 19 days showed marked improvement in fluid intelligence capacity relative to controls--- the authors argue that they have explored a method to improve fluid intelligence. A subsequent study has further demonstrated that working training games can in turn improve crystallized intelligence acquisition among learning-disabled students.

The provided link points to the project page for Brain Workshop - a Dual N-Back game, which provides an open source version of the game used in the 2008 paper on fluid intelligence. A web-based flash version also exists.

posted by melatonic (31 comments total)

That's cool. It's like Simon™ grew up, went to college and got his PhD.
posted by kuujjuarapik at 2:45 PM on May 9, 2010 [3 favorites]

Not taking sides here, but this should be mentioned:

Brain training games will not make you smarter.
posted by WalterMitty at 2:46 PM on May 9, 2010 [1 favorite]

Just in case it's not clear from the first link, what's particularly interesting about this study is that they achieved a transfer effect from one test of intelligence to another. That is, the participants didn't just get better at taking the test by practicing that test, which distinguishes it from the apparent improvement people usually see in brain training games like Brain Age. What's interesting to me, then, is that the training game and the intelligence test have to be similar enough (so you know you're measuring the same kind of intelligence) but also different enough (so people aren't just improving at that one task). In the study, they describe the actual test of fluid intelligence they used:

Transfer tasks: We used standardized fluid intelligence tests, consisting of visual analogy problems of increasing difficulty. Each problem presents a matrix of patterns in which one pattern is missing. The task is to select the missing pattern among a set of given response alternatives.

But I'd be interested to hear more of the details of how that particularly compares to the dual n-back game.
posted by albrecht at 2:59 PM on May 9, 2010 [1 favorite]

I suspect mice may be involved. Also reminds me of the game, but then so do the people on the subway playing sudoku.
posted by jardinier at 3:26 PM on May 9, 2010

@WalterMitty: Thanks for the link; I didn't want to log in so I found a similar article in the telegraph Popular brain training games 'do not make users any smarter'

I'm curious about what the games in this study entail. Does anyone know of a listing of example games and rules. I'd be interested if some of them have the same structure as dual n-back.
posted by melatonic at 3:30 PM on May 9, 2010

Thanks for the post.
posted by anotherpanacea at 3:39 PM on May 9, 2010

@albrecht: dual n-back isn't so much about visual analogy pattern completion as it is about the memory and processing of two sequences of signals. You have to hold in mind n items, depending on how many 'back' the game goes.

In dual 1-back, you would signal if the position or sound of the current trial is the same as the last trial. In dual 2-back you would signal if either were the same as the trial previous to the last. In dual 3-back you would signal if it were the same as three trials previous, and so on.

The effect is that you have to constantly memorize and forget two sequences of length n in order. You can see this without playing the game because you can't remember only the n-previous signals and somehow ignore all the other items. You have to remember all the previous signals up to n, shift the n+1-previous ones out of your mind as the game progresses, and memorize the current signals while responding.

The best way to understand is by trying the game. It sounds simple but this aspect makes it surprisingly difficult.

I get the feeling of juggling one too many balls. Provided you play a game that always increases n when you become proficient, this sensation will never stop and you will always be increasing the number of items you can hold in memory. If the techniques and mental phenomenon that allow you to do this can be generalized to other working memory tasks, you have a pretty good system for improving working memory capacity. If they can't, you get a game you can 'beat' without generalizable skills.

I'd be interested what someone who has played the brain age games thinks of dual n-back.
posted by melatonic at 3:42 PM on May 9, 2010

Trying to play this leads me to believe it could be explained better. I had a couple of annoying sessions just trying to figure out what the hell it wanted. When an annoying voice is going "Ha! Cee! Tee! Cee! Ha!" at the same time, well, it was a little annoying. For anyone else who might happen to have been annoyed, this is how I understand the game:

For "dual 2-back":

You're given a series of images and sounds, each at the same time. Your job is to remember the sequence. When the sound you just heard matches the one you heard the time before last, press a given key. When the image you see matches the one you heard, again, the time before last, press the other key.

Higher-order games (dual 3-back, dual 4-back), if I'm understanding it correctly, require you to remember the sound & symbol from progressively further back.

It seems like I've been having some problems with short-term memory lately, so I look forward to trying this out.
posted by JHarris at 3:46 PM on May 9, 2010

@albrecht: I think I get your point a bit more. Dual n-back involves a combination of auditory and visual stimuli. Thinking about it, I would find the paper much more convincing if they used only audible or tactile stimuli, and then tested using a visual test. There is a control problem: perhaps just looking at structured visual stimuli makes you do better at visual pattern tests.
posted by melatonic at 3:48 PM on May 9, 2010

melatonic: I played the first Brain Age a fair bit. It's not bad, but I think it's not varied enough. There are relatively few minigames on the card, so I suspect people learn the games more than develop their cognitive flexibility.
posted by JHarris at 3:48 PM on May 9, 2010

(For an example of my problems with short-term memory, notice how I used "annnoyed" or a version of it three times in my first paragraph. Ouch!)
posted by JHarris at 3:50 PM on May 9, 2010

I can feel it working already! I'm so fluidly intelligent I know not to download a random binary from the internet based on claims of increasing my penisbrain size!
posted by DU at 3:56 PM on May 9, 2010 [4 favorites]

WalterMitty, the point of this post is that the dual n-back game appears to be an exception to that finding, see for example in the disabled students link this statement: "Although there is evidence that Gf can be improved through memory training in adults, the efficacy of memory training in improving acquired skills, such as Gc and academic attainment, has yet to be established. Furthermore, evidence of transfer effects from gains made in the trained tasks is sparse."
posted by invitapriore at 4:09 PM on May 9, 2010 [1 favorite]

I've always had my doubts about brain games. They're like vitamin pills -- both are easy to swallow, and both do no more than make you think you're doing something worthwhile. Of course, once in a while, there's an exception (which this apparently is).

Honestly, just go out and learn something that you don't understand. Challenge yourself. Be confused. Get frustrated. Then keep at it until you understand it, and you'll increase your cognitive capacity.
posted by spiderskull at 4:23 PM on May 9, 2010 [1 favorite]

Spatial recognition and sound/music are both right-hemisphere-oriented functions. Maybe multi-mode n-backs would be worth exploring, with an eye toward hemispheric integration.
posted by psyche7 at 4:34 PM on May 9, 2010

Thinking about it, I would find the paper much more convincing if they used only audible or tactile stimuli, and then tested using a visual test. There is a control problem: perhaps just looking at structured visual stimuli makes you do better at visual pattern tests.

Yeah, that's more along the lines of what I was thinking about. I don't know enough about these things to say with any certainty, but it seems like there's kind of an art form in picking the right combination of training/test to demonstrate an actual transfer of some kind of cognitive skill, rather than an increased ability at taking that test. If you take a broad enough definition of "fluid intelligence," then it ought to be measurable by lots of tests that are very different from the training process. It seems in this study, though, that the test was fairly close to the dual n-back training game, although maybe sufficiently dissimilar that the transfer is indicative of something. I'd just be curious to hear an analysis of why they think that test is the right combination of similar and different.
posted by albrecht at 5:20 PM on May 9, 2010

It seems in this study, though, that the test was fairly close to the dual n-back training game, although maybe sufficiently dissimilar that the transfer is indicative of something. I'd just be curious to hear an analysis of why they think that test is the right combination of similar and different.

I agree that the measure is just distinct enough from the training method to suggest a degree of transference to the gains made through the training method, but not so distinct as to be completely compelling.

The authors are not solely targeting working memory, which is a relatively well-defined and testable entity, but a combination of working memory and the ability to refresh it rapidly and precisely. They then must make the leap to argue that another unrelated test is distinct enough to only benefit from the improvement in these simultaneous capacities. While this makes sense intuitively it is very, very hard to prove in a small study.

The authors would do well to issue a battery of tests of various kinds. This would help to situate the specific benefits of this type of training. They are currently working on a longitudinal study which hopefully will broaden the scope of capacities measured.

The curious among us will just have to experiment on ourselves.
posted by melatonic at 7:05 PM on May 9, 2010

I practiced with a dual n-back system for a while when the paper first came out. In fact, I was planning an iphone app when the iphone app store first debuted, but alas, one of the first apps was a dual n-back. Anyway, when you get up to 5 or so, no other brain training game (of the few I've skeptically tried) is remotely as difficult. The authors claim to have controlled for working memory per se, but mostly it still feels like it's just exercising working memory. One odd effect is that, as the chains you have to remember get longer (and recall that each time a new location/sound is added, you essentially have to pop the oldest element off your remembered array and push the next pair onto the front), you no longer quite consciously recollect the exact sequence, you just get a feeling for when a new location or sound appears that feels like it was about 5 ago; but even if it's just a feeling, it will be quite accurate. For a while I imagined that my working memory muscles were indeed strengthening, the main sensation being that I could retain the various threads of a complicated conversation better as they dangled and were forgotten by the other conversationalists. But that was probably just wishful thinking. Because it's boring and difficult, I haven't stuck with it, though I keep intending to.
posted by chortly at 7:42 PM on May 9, 2010 [2 favorites]

i have done this while in an fMRI scanner - not easy. and i was only doing 2-back! it is used to activate your working memory part of your brain, i think. but i doubt it is a test for any intelligence nor a method of improving memory. pshaw.
posted by lucysun at 8:04 PM on May 9, 2010

Until now, I've never played a game so fucking hard that the demo made me quit.
posted by spamguy at 9:25 PM on May 9, 2010

That's Letterwang!
posted by Strange Interlude at 9:55 PM on May 9, 2010 [6 favorites]

Strange Interlude wins the thread.

If the theory behind the promotion of this game is correct, something like Go should work just as well (if not better), with the advantage of being entertaining and social. The only argument I can see for the n-back is it's purity.

It would be really evil to combine this with ear training. Press the button if the tone three tones ago was a perfect fifth from the current tone.
posted by phrontist at 11:30 PM on May 9, 2010

I think it's interesting to see how hard it is for a human brain to do something like this; an equivalent computer program playing this setup would be trivially easy to implement and (given enough memory) would do this test indefinitely without ever failing.

Personally, I failed catastrophically at this - I really hate doing things simultaneously, I prefer to do tasks in order and focus on each step in turn; this test asks for too many things at once (listen new sound, recognize new image, remember new sound, remember new image, compare new sound with old, compare new image with old, press correct button(s)), making the exercise incredibly frustrating. I wonder if anyone who has had experience programming parallel processes running in multiple threads simultaneously has a better fitting mental landscape that would easier deal with this kind of complexity.
posted by PontifexPrimus at 1:36 AM on May 10, 2010

If the theory behind the promotion of this game is correct, something like Go should work just as well (if not better), with the advantage of being entertaining and social. The only argument I can see for the n-back is it's purity.

The thing about the computer game is that you can do it little and often...they recommend 20x per day.

I think this is important as it feels like a difficult thing to get better at...you have to be really consistent.
posted by Not Supplied at 6:03 AM on May 10, 2010

Not taking sides here, but this should be mentioned:

Brain training games will not make you smarter.

Hmm, PNAS or the, uh, Financial Times? It's hard to know what sort of outlets to trust on matters like these. Good thing we've got some options...

it is used to activate your working memory part of your brain, i think. but i doubt it is a test for any intelligence nor a method of improving memory. pshaw.

Which part might that be? The frontal-parietal-temporal-occipital network, perhaps?
posted by solipsophistocracy at 7:04 AM on May 10, 2010

First, I would love this to hold true, and think it warrants further study. That said...

The linked study has a serious flaw - Which the authors note, so no dishonesty here, but I do think they gloss it over more than they should.

The control group showed significant improvement on the transfer task (the "other" test to see if results transferred - A visual analogy battery, so fairly distinct from the n-back task). The training group showed significantly more improvement, however, and the authors subtract the control group from the training group to claim "real" transfer occurred.

The flaw here comes from presuming the sequence effects merely interact additively, which we already know does not always hold true. Novelty in itself has a priming effect on learning rate across all domains - Rats with more stimulating environments learn to run their mazes faster.

As a follow-up, I'd like to see n-back compared with some task orthogonal to fluid intelligence; unfortunately, crystallized intelligence generally doesn't change rapidly enough to give a meaningful basis of comparison, so it would take quite some care to come up with a control task that A) correlates poorly with Gf, B) changes measurably over the course of a few days; and C) doesn't itself count as exciting enough to overwhelm the training task itself.



PontifexPrimus : I think it's interesting to see how hard it is for a human brain to do something like this; an equivalent computer program playing this setup would be trivially easy to implement and (given enough memory) would do this test indefinitely without ever failing.

Humans tend to have a short-term memory capacity for seven "things", where our brains really don't differentiate much based on the complexity of the things in question. Most mnemonic tricks work by making the definition of "things" progressively more complex - For example, seven digits of a phone number contain far less information than seven new name/face pairs at a party, but we have roughly the same ability (and limitation) to remember either of them. If, therefore, you can synthesize a face with 8 freckles, 6 teeth, 7 moles, 5 forehead wrinkles, 3 chins, 0 eyebrows, and 9 nose hairs, you have compressed Jenny's phone number into one "thing".


I wonder if anyone who has had experience programming parallel processes running in multiple threads simultaneously has a better fitting mental landscape that would easier deal with this kind of complexity.

Nope, I sucked at it too. :)

Keep in mind, though, the claimed findings have nothing to do with absolute levels of performance, just increases in performance from each individual subject's baseline.
posted by pla at 7:35 AM on May 10, 2010

I suck frighteningly bad at this.
posted by BrotherCaine at 3:43 PM on May 10, 2010

Humans tend to have a short-term memory capacity for seven "things", where our brains really don't differentiate much based on the complexity of the things in question. Most mnemonic tricks work by making the definition of "things" progressively more complex

I don't think the 7+/-2 "slots" for STM is really in vogue these days beyond intro to cognition courses (and shouldn't even show up there except for historical purposes). It is true that this number of average slots in STM was well replicated for many years, but contemporary thinking suggests that the "magic number" is much more a product of time than space. It just so happens that 7+/-2 is about the number of digits, chunks, or whatever that you can reliably repeat before decay sets in for the first one. Although you don't necessarily say it out loud, when rehearsing information in the phonological loop, the neural code for the articulo-motor output you would be producing is referenced/simulated and "perceived" by auditory cortex associated with the acoustic buffer, and is essentially bound by rate of speech. This process can go on indefinitely, but does not in any way contribute to the kind of elaborative rehearsal needed to encode information in long term memory. 7+/-2 is a coincidental number of stuffs people can rehearse in the maximum amount of time that information can exist in this temporary auditory store, not a number of "things."

In fact, we don't even really refer to STM much these days, except by means of comparison to newer models (although there are die-hards out there who cling to it, and refuse to acknowledge that its most likely a byproduct of the auditory tasks used to measure it).

We call it working memory: the phonological loop, the central executive* and the visuo-spatial sketchpad.

Similar buffers have been proposed for motor learning, among others, but these two are the most well understood. The capacity of visual working memory (the sketchpad, visual buffer, whatever you choose to call it) is about 3-4 objects. Actual objects. Complexity and number of the objects are orthogonal properties, and do not tend to limit each other. WM capacity correlates much better with general fluid intelligence, IQ, academic success, and high scores on more culturally sensitive measures of intelligence than old school STM.

It has been argued further that WM and attention are essentially indistinguishable. WM is pretty much the number of things you can actively process, or attend to.

Is visual working memory going to be supplanted by some newer theory? Maybe, but the neural evidence for this idea is pretty tight. Executive control has always been of more interest to me, but it's difficult to divorce that from attention as well. No matter where you come down on the current myriad schizms in cog neuro, though, seven plus or minus two is wicked old school.

*It is true, however, that this is not really a solution so much as a better name for an old problem. Ole big bad Baddely himself even said that the central executive is "a residual area of ignorance."

I am all about a lively debate, and will be more than happy to provide citations if you so require, but I elected not to clog the thread with links that are likely inaccessible to a significant portion of folks without a request to do so. A quick googlin' will back up the majority of these claims, however.
posted by solipsophistocracy at 9:22 PM on May 10, 2010 [3 favorites]

Not taking sides here, but this should be mentioned:
Brain training games will not make you smarter.

For what it's worth, this survey did not look at dual n-back games, which have been considered useful specifically because they seem to have different effects than standard brain training games. Also, more criticism here.
posted by Adam_S at 7:40 AM on May 11, 2010

I've been working with the dual n-back program for a bit of time now. Improvement is slow, but seems to be happening; I just had a 68% run at dual 3-back. Observations like this are not really scientific and hellishly subject to bias, but I think I may be noticing it slightly easier to think effectively.

I'd report more, but I think the expiration date on this thread is approaching.
posted by JHarris at 9:06 PM on May 27, 2010

Brain training games will not make you smarter.

This is true, but neither are they worthless I think. It's like, you could be able to multiply four digits to four digits in your head, but if you're still a Tea Partier, there is not an awful lot that could be done for you. Some people are willfully unintelligent; until they manage to let go of that rock dragging them down, it will be a bad influence on their judgment. The best reasoning computer in the world still cannot deduce truth from invalid premises.

Something I've been considering for a long time now (as a layman; I have no training in neuroscience or even biology) is the possibility of metaknowledge as an aid to thought. When I was younger I was fairly good at in-head calculation and things like that. I'm out of practice now, but I think I'm much better able to think about the world. College (liberal arts!) helped out a lot there, and participating with Metafilter I really think helped out more. I wonder, if I were to get back into math solidly (I got like a 730 on the math portion of the GRE after all), would I do better at it?
posted by JHarris at 9:16 PM on May 27, 2010