What is dual n-back, and does it do anything?

What the task is

N-back is a working-memory task. You are shown a sequence of items one at a time, and at each step you have to say whether the current item matches the one shown n steps earlier. At 1-back you compare with the previous item; at 2-back with the one before that; at 3-back with the one three steps ago. The higher the n, the more items you have to hold in mind, and the more you have to keep reshuffling as each new one arrives.

Dual n-back runs two independent sequences at once, most commonly a position on a grid and a spoken letter. At every step you make two separate decisions: does the position match the one n back, and does the letter match the one n back? Either, both or neither can be true. That doubling is what makes the task demanding: you are tracking two streams, each with its own history, and updating both every couple of seconds.

A 2-back, step by step

Suppose the grid has nine cells and a letter is read out with each one. At 2-back, nothing can match on the first two steps because there is nothing two steps back yet. From the third step onward, every item is compared with the item from two steps earlier.

Look at step 7 carefully, because it shows why the task is hard. The position does match: step 5 was bottom right, and step 5 is two back from step 7. But step 6 was centre, so if you are only remembering “the last one” you will say no. And step 4 was also bottom right, which is three back and irrelevant, so if your memory of the order has blurred you may say yes for the wrong reason. Doing this correctly requires keeping a precise, ordered buffer of the last two items in each stream, and shifting it by one every time a new item arrives.

At 3-back the buffer is three deep in each stream. Most people who start at 2-back find 3-back a wall for a while. That is normal.

Why it became famous

Dual n-back had been used in research for years before a 2008 study made it a phenomenon. In that study, people who trained on dual n-back for a few weeks improved on a test of fluid intelligence — the ability to reason through unfamiliar problems — and the gain was larger the more sessions they had done. Fluid intelligence was widely regarded as essentially fixed in adults, so a training task that appeared to raise it was extraordinary news, and it spread through the press, forums and the first wave of brain-training apps within months.

It also had an intuitive story behind it. Working memory and fluid reasoning are strongly correlated, and n-back plainly loads working memory hard, so the idea that training one would lift the other seemed plausible rather than magical.

Why the follow-ups were mixed

Plausibility is not evidence, and the replications did not go cleanly. Some later studies found a similar improvement; others, often larger and more tightly controlled, found nothing beyond improvement on n-back itself. The criticisms centred on three things.

• Control groups. In many positive studies the comparison group simply did nothing, or did something less engaging. A group that expects to improve and is putting in effort tends to score better on the follow-up test for reasons that have nothing to do with working memory. Studies with an active, equally demanding control task found smaller effects.
• Outcome measures. A single reasoning test, sometimes given with a shorter time limit than usual, is a narrow window on “intelligence”. Improvement on it can reflect strategy or test familiarity rather than a broader change.
• Sample size. Small groups produce noisy results, and noisy results that happen to be positive are more likely to be published.

When meta-analyses pooled everything, the far-transfer effect was small and shrank further when only the well-controlled studies were counted. Near transfer — getting better at n-back and at closely related working-memory tasks — was solid throughout. The honest summary today is that dual n-back trains dual n-back very effectively, and that whether it trains anything else is unresolved and probably minor. Does brain training work? covers the wider picture.

How to practise it sensibly

If you want to do n-back, do it the way the research did it, without the expectations.

1. Start at 2-back, not 1-back. 1-back is a reaction task; the working-memory load only really begins at two.
2. Short sessions, most days. Around fifteen to twenty minutes is what the training studies used. Longer sessions add fatigue, not learning.
3. Let the level move both ways. An adaptive scheme that raises n after a strong block and lowers it after a weak one keeps you at the edge of what you can do, which is where the practice is. Sitting at a level you can already manage is comfortable and useless.
4. Do not lean on tricks. Rehearsing the letters as a chant, or only attending to one stream, will raise your score and empty the task of the thing it was supposed to exercise. If a strategy makes it feel easy, it is probably bypassing the load.
5. Expect variation. Sleep, caffeine and time of day move n-back performance a lot. A bad day is data about the day, not about your progress.

Braintide’s dual n-back uses a position stream and a sound stream, and the brief on its first screen says which research task it descends from.

Plateaus

Almost everyone stalls, usually somewhere between 3-back and 5-back. Early progress comes from learning the format and settling on a way of holding the items; that runs out, and what is left is the slow, unglamorous business of a working-memory buffer under load. Plateaus can last weeks.

Three things help. Take a few days off rather than grinding; performance often steps up after a break. Vary the task — a different working-memory exercise for a week, then back. And stop watching the level. If the level is the goal, a plateau is a failure; if consistent practice is the goal, a plateau is just what practice looks like for a while.

If you have plateaued and the practice has stopped being interesting, that is also a reasonable point to stop. There is no evidence that pushing from 4-back to 5-back does anything for you beyond 5-back.

Who it is for

Dual n-back suits people who enjoy a hard, austere task and want a single number that reflects effort honestly. It does not suit anyone hoping for an easy win, and it is not the right place to start if you have never used a brain-training app: begin with a wider rotation of tasks and other working-memory exercises, and come to n-back once you know you will stick with it.

Frequently asked questions

Does dual n-back increase IQ?

The 2008 study that made it famous reported gains on a fluid-intelligence test, but later, better-controlled replications mostly did not, and meta-analyses find the effect shrinking towards zero as study quality rises. It reliably improves n-back itself and closely related working-memory tasks. Beyond that, the evidence is weak.

What is a good dual n-back level?

Most people start around 2-back and, with regular practice, settle somewhere between 3-back and 5-back over a few months. There is no meaningful target beyond that; the level tells you how well you do this particular task, not how intelligent you are.

How long should a dual n-back session be?

Around fifteen to twenty minutes, most days, is what the research protocols used. Longer sessions tend to add fatigue rather than learning, and consistency matters more than duration.

Why is 3-back so much harder than 2-back?

Because the buffer you have to keep in order grows from two items to three in each of two streams, and every new item forces you to shift all of it by one. The difficulty rises faster than the number suggests, and stalling at 3-back for a while is normal.

Not a medical tool: Braintide is a game, not a cognitive assessment or a treatment. These guides summarise published research for general interest and are not medical advice. If you are worried about your memory or thinking, talk to a doctor.