Does Working-Memory Training Improve General Intelligence?

Does working-memory training improve general intelligence?

Working-memory training reliably teaches people to do the practiced exercise better; improvement on a similar, untrained task is also possible. But current evidence has not established reliable improvement in general intelligence. These are three different outcomes: learning the trained task means getting better at that exercise; near transfer means doing better on a new task that draws on similar operations; a gain on an independent reasoning measure would be broader evidence, though it still would not by itself prove a general change in intelligence. The distinction matters because a larger score on a familiar drill can be useful practice without showing that reasoning ability has changed across settings. The early positive fluid-intelligence result is a real finding and deserves careful consideration as a meaningful and intriguing scientific signal, but it does not settle whether such broader gains are dependable.

Why did the first positive result matter?

The early positive result mattered because it tested a plausible route from a narrow exercise to a broader ability. In “Improving Fluid Intelligence with Training on Working Memory,” participants practiced adaptive dual n-back: they had to retain information from a changing sequence and update what they remembered as new items appeared. As the task adjusted in difficulty, successful performance called for keeping recent information available while responding to new input. That combination made working memory an intuitively relevant candidate for training: if a person became more effective at maintaining and updating information, perhaps that practice could support performance on reasoning problems that also require holding information in mind. This is a rationale for asking whether transfer occurs, however, not evidence that the trained operation and general intelligence are interchangeable. A shared demand can motivate a prediction; only an outcome measure can show whether the prediction is borne out.

The report drew attention because it described more than an improvement on the exercise participants had repeated. The study reported that adaptive working-memory training was followed by gains on a fluid-intelligence measure. That result challenged the easy assumption that practice effects must stay confined to the practiced activity. If an intervention built around a particular updating task could also shift performance on a separate reasoning measure, then training might affect something that mattered beyond proficiency with the drill. For readers, the distinction is consequential: a product that improves its own scores may teach its own format, while a change on an independent measure would offer a more interesting signal. The early report therefore raised a testable possibility rather than merely showing that participants could learn a demanding game.

The proper weight of that finding is neither to wave it away because later evidence may disagree nor to treat it as a settled promise. It establishes that a positive result was reported for a particular adaptive program, studied participants, and fluid-intelligence measure. It does not by itself establish that the gain lasts, appears across different reasoning measures, or recurs when another group runs a controlled test. Nor does an average study result guarantee that an individual who practices will see a meaningful change. The mechanism that made n-back interesting cannot answer those questions: retaining and updating sequence information may overlap with some reasoning demands, but overlap alone does not show broad transfer. The result’s importance lies in the hypothesis it made worth testing. A stronger conclusion requires repeated evidence that separates improvement on the trained exercise from change on independent outcomes; the following controlled replication addresses whether the early pattern extends beyond its original setting.

Sources: Improving Fluid Intelligence with Training on Working Memory

What does ‘transfer’ mean in this research?

To interpret a training result, ask how much the outcome task resembles the exercise people practiced. Improvement on the trained task is task learning: performance rises on that same exercise through repeated exposure and feedback. Near transfer describes improvement on a new, unpracticed task that shares important demands with the exercise. Far transfer describes improvement on outcomes more removed from the trained activity, such as a distinct reasoning measure. These terms describe relative distance between tasks; they are not separate kinds of score and do not, by themselves, tell us why performance changed. The useful question is what the participant had to do in each task, not simply whether researchers gave the tasks different names.

Consider an explicitly illustrative example, not a study result. A learner repeats a drill that asks them to keep a short sequence in mind and update it whenever a new symbol appears. They then try a fresh task with different symbols, timing, and response buttons, but it still requires holding a sequence and updating its contents. That second task is unpracticed, yet its central operation remains close to the drill. Altering surface details can make an exercise look new without changing the mental work it requires. A gain there can show that performance extends beyond exact item repetition, while leaving open whether the learner has acquired a broadly useful ability or learned a strategy that fits both formats.

Now imagine the learner takes a separate reasoning task that asks them to infer a rule among visual shapes, with no running sequence to maintain. Because its demands differ more from the drill, improvement would count as evidence at a greater outcome distance. It still would not automatically establish a general change in intelligence: the result would depend on what the reasoning task measures, how well it measures it, and whether the change can be attributed to training. Calling an outcome far transfer marks a claim that needs broader evidence; the label cannot supply that evidence. Likewise, a task called near transfer may measure a genuinely shared skill, not merely invite reuse of a trick. Similarity makes strategy reuse plausible and limits how far the finding reaches, but it does not make a gain meaningless.

Researchers classify outcomes by comparing the actual operations, content, response demands, and scoring of the training and assessment tasks. A task can share one important demand with training while differing in others, so the boundary is not always a clean dividing line. Two assessments may also look different while relying on the same underlying operation. The categories are therefore a practical map of inference distance, not a verdict about the mechanism behind an observed score change. In the evidence discussed here, the distinction helps separate learning the practiced exercise from performance on related measures and from broader claims about intelligence. Each step outward asks for evidence on that outcome itself; a higher score at one distance does not establish gains at every distance beyond it.

Sources: Failure of Working Memory Training to Enhance Cognition or Intelligence

What did a controlled replication find?

The controlled replication reported in “Failure of Working Memory Training to Enhance Cognition or Intelligence” tested whether dual n-back practice produced broader gains beyond improvement on the exercise itself. Its comparison design included three groups: participants assigned to dual n-back, participants assigned to an active multiple-object-tracking task, and a no-contact group. The active group mattered because it spent time learning another demanding task. Comparing dual n-back with that group asks whether the particular working-memory program outperforms an alternative engaging activity, not merely whether people improve after receiving attention, expectations, scheduled practice, and repeated assessment. A no-contact comparison provides a different reference: it shows what happened over the same period among participants who did not undertake either training program.

The study randomly assigned young adults to 20 training sessions, with the active multiple-object-tracking group scheduled for a comparable amount of practice; the no-contact participants completed assessments without the training schedule. The design thus separates two questions that a simple before-and-after score cannot answer. A change shared across trained and untrained participants may reflect passage of time or familiarity with testing, while an advantage specific to dual n-back over another active task would be more consistent with a training-specific effect. The no-contact group helps place observed change against the retest and time pattern, though it does not control as closely for engagement and expectation. These comparisons strengthen interpretation by making the alternatives visible; neither comparison makes an imperfect measure or a limited set of outcomes comprehensive.

The reported pattern was clear at the level of the exercises: participants improved substantially on the tasks they practiced, and much of that learning remained evident at a six-month follow-up. Yet the researchers found no corresponding transfer to the measured untrained cognitive abilities relative to the comparison groups. In other words, the training could teach participants to perform its own demanding activities better without producing a detectable training-specific advantage on the broader outcomes selected for this experiment. The result matters because task learning was present, so the absence of broader measured gains cannot be explained simply by saying that participants failed to learn the exercise. Practice and transfer are different outcomes, and this experiment observed one without demonstrating the other.

The active comparison makes the inference more demanding than asking whether dual n-back participants scored higher after training than before it. If repeated testing or general participation were sufficient to create the broad pattern, similar changes might appear in the other groups; if dual n-back had a distinctive broad benefit, it should have produced an advantage beyond an equally scheduled alternative. The study did not find that pattern on its selected measures. The no-contact group also helps contextualize change that occurred without either intervention, although differences in contact mean it cannot answer every expectancy question on its own. Taken together, the comparisons support a bounded conclusion: this replication found retained learning on trained tasks but no measured broader transfer attributable specifically to dual n-back.

That conclusion is narrower than saying working-memory training cannot help anyone or that no program could ever affect a broader outcome. The experiment concerned healthy young adults, particular training schedules, and particular assessments; an effect not detected on those measures could differ in another population, program, or outcome. At the same time, the presence of strong trained-task improvement gives no warrant to assume that general reasoning improved invisibly. The study’s value is its direct test of a broad prediction with both an active practice comparison and a no-contact reference, and its result did not reproduce broad gains under those conditions. That is meaningful counterevidence to a general promise, while leaving a more specific future claim open to direct testing.

Sources: Failure of Working Memory Training to Enhance Cognition or Intelligence

What do the broader reviews find beyond one experiment?

The two broad reviews point in the same cautious direction, but their scope and comparisons matter. “Is Working Memory Training Effective? A Meta-Analytic Review” (2013) brought together 23 studies and 30 comparisons, including clinical and typically developing children and adults. Across that varied evidence base, it found short-term improvement on working-memory tasks but no convincing generalization to broader outcomes including verbal and nonverbal ability, attention, reading, and arithmetic. This is useful because the conclusion does not rest on one young-adult n-back experiment: it spans different ages, participant groups, programs, and outcome families. Yet that breadth is also a limit. A pooled conclusion across heterogeneous studies is not a precise forecast for every age group, intervention, or current training program; it describes the pattern in the studies the review could combine.

The later updated review, “Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review,” makes the comparison group especially important. It separated studies using untreated controls from those using treated controls: participants assigned another activity intended to match some experience of training. Against treated controls, the review found gains on working-memory outcomes resembling the trained tasks, but the broader ability outcomes it assessed did not show reliable differences from zero in the reported analyses after outlier handling. The authors also reported that working-memory gain size did not track far-transfer gain size. That pattern suggests that getting better at closely related tasks did not, across these studies, predict a corresponding broader ability gain.

A treated control changes what a difference between groups can mean. Both groups spend time on an activity, receive contact and instructions, and may expect improvement; comparing them better accounts for engagement, attention, expectations, and the experience of doing an alternative task than comparing training only with no activity. If both groups improve, the trained group must show an advantage over that active alternative before the difference supports a training-specific interpretation. Such controls do not automatically match every feature of an intervention or remove all bias: the activities may differ in credibility, effort, adherence, or other ways that affect outcomes. The updated review’s treated-control analysis therefore makes the broad-transfer test more demanding, while its estimates still inherit limitations in the underlying trials and in how their outcomes and comparison activities were chosen.

Read together, the reviews support a measured conclusion rather than a universal rule: the pooled literature reports improvement most consistently on practiced or similar working-memory outcomes, while it has not established dependable gains across the broader abilities examined. The later review sharpens that judgment by showing why results against no-treatment groups may overstate what is specific to the training. It does not prove that no targeted program can ever help a particular person or outcome, and the 2013 review’s mixed populations cannot settle effects for each subgroup separately. Better matched studies could change the picture. For a reader deciding what a practice result means now, however, these reviews give no sound basis for treating a higher score on a drill or close analogue as evidence that general intelligence has risen.

Sources: Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer’: Evidence From a Meta-Analytic Review; Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies

Why does task similarity change the size of the claim?

The n-back-specific synthesis makes task similarity visible by comparing outcomes at different distances from the exercise people practiced. “Working Memory Training Revisited: A Multi-Level Meta-Analysis of N-Back Training Studies” examined randomized trials of n-back updating training in healthy adults. Its clearest pooled transfer was to an untrained n-back task: a new test within the same paradigm family. Results for other working-memory tasks, fluid intelligence, and cognitive control were reported as similarly very small. The distinction matters because an untrained n-back form is not simply the practiced items again; success on it indicates that performance can carry to another version of a task with closely related demands. But it remains much closer to n-back practice than an independent measure of general reasoning.

This pattern supports a useful interpretation rule. Improvement on a fresh n-back version is evidence of generalization within a task family, since the exact practiced test has changed while much of the central format and updating demand remains. It is not a direct measurement of general intelligence, nor does the synthesis establish that the gain transfers to unrelated situations. The conclusion is specifically about n-back training in healthy adults and the outcomes included in those trials; it should not be stretched to every kind of working-memory program or every population. The broader ability claim would need corresponding improvement on appropriate, independent measures. A boundary on interpretation does not make task-family learning trivial: it can matter when the goal is to perform n-back-like tasks better, and it helps researchers identify where further transfer should be tested.

This comparison also clarifies what a training claim should name. “Transfer” is incomplete unless it identifies the destination: transfer to another n-back task, to a distinct working-memory measure, or to a broader reasoning outcome are different findings. When the strongest result is concentrated in the first destination and does not appear at similar size in the others, the safest account is that practice generalized across related task versions, not that a single underlying capacity expanded everywhere. That is an interpretation of the outcome pattern, not proof of the mechanism: shared task demands may help explain the closer result, but the pooled abstract cannot determine exactly which strategy or process carried over. For practical purposes, someone seeking better performance on n-back-like exercises may value that narrower gain. Someone asking whether working-memory drills raise general intelligence needs evidence on broader measures, not an inference from resemblance alone.

Sources: Working Memory Training Does Not Improve Executive Functioning or Fluid Intelligence

Does a longer training schedule settle the question?

A longer schedule can test whether gains persist or reach outcomes beyond the practiced exercises, but duration alone cannot make that transfer happen. “Training Working Memory for Two Years—No Evidence of Transfer to Intelligence” is useful because its researchers followed a program far longer than a brief laboratory training block and assessed broader constructs rather than relying on one reasoning score. The participants were 112 students who began in ninth grade; they practiced a heterogeneous set of working-memory tasks every two weeks. A separate group of 113 students, also initially in ninth grade, served as a pretest-posttest control. This design makes the study a test of a specific sustained school-age program, not simply a demonstration that students can improve when they repeat exercises.

The study reported improvement on practiced tasks and on a latent working-memory factor, while finding no observed transfer to latent fluid-intelligence or crystallized-intelligence factors. A latent factor combines information from multiple measures intended to reflect a broader construct. That approach can reduce dependence on the quirks of a single test, so the intelligence outcome was not just one isolated score. The contrast is the key finding: the program was associated with improvement in the working-memory domain it targeted, yet broader intelligence factors did not show corresponding transfer. The result therefore puts pressure on a simple duration explanation—that if people practice long enough, a narrow skill gain will eventually become a general ability gain.

Latent outcomes do not make the conclusion unlimited. They represent the particular measures, model, students, and intervention used in this study; they cannot stand in for every possible definition of intelligence or every assessment setting. Nor does an unobserved transfer effect prove that the true effect was exactly zero. The comparison group and repeated assessments help interpret the reported pattern, but this study does not test adult training programs, different training content, or outcomes outside its design. Its contribution is narrower and still important: among students entering ninth grade, two years of this regular practice produced working-memory improvement without observed transfer to the measured fluid or crystallized intelligence factors. Extra time is not itself evidence of broader change.

For someone deciding whether a prolonged routine is likely to raise general intelligence, this study does not support treating the calendar as the mechanism. A long intervention can create more opportunity to learn the practiced operations and can give researchers a chance to observe broader outcomes. Those are advantages of the design, not a guarantee about what the outcomes will show. Here, the broader latent measures remained distinct from the working-memory gains. The finding is consistent with the larger distinction in this article: improvement should be named at the level where it was measured. Practice may be worthwhile for the skills or tasks it targets, but a claim about intelligence needs evidence on intelligence outcomes. The ninth-grade sample and this particular program limit how far the result travels; they do not turn duration into proof of transfer.

Sources: Training Working Memory for Two Years—No Evidence of Transfer to Intelligence

What can one person’s higher retest score establish?

A higher score after practice establishes that performance was higher on that occasion under those conditions. By itself, it does not establish why the score changed. Consider a hypothetical person who completes a reasoning test, practices, and later earns a higher score. They may remember the item formats, discover a strategy that fits those formats, attend more closely on the second sitting, benefit from ordinary familiarity with being retested, or show genuine improvement on the task. Several of these can happen together. One before-and-after difference cannot separate them, and describing it as a training effect goes beyond what the observation alone shows.

This is why individual change cannot be inferred directly from an average study result, or vice versa. “Working Memory Training Does Not Improve Executive Functioning or Fluid Intelligence” reports a randomized study in which both the training and control groups improved on most retested outcomes, despite no training-specific transfer. That group-level result illustrates how scores can rise across repeated measurement without identifying a training-specific cause for every person. It does not tell us what caused the hypothetical individual's gain or quantify the retest change that any particular person should expect. It shows why a comparison is informative: if a suitable group that did not receive the target training also improves, a rise in the trained group alone is not enough to attribute the change to that training.

A stronger causal claim needs outcomes that do not simply repeat the practiced material, testing conditions kept as consistent as practical, a measure adequate for the ability being discussed, and a comparison group that helps estimate what would have happened without the training. Fresh outcomes reduce the chance that remembering particular items explains the result; comparable conditions reduce avoidable variation; the comparison helps distinguish the intervention from retesting and other shared influences. These features support a more careful inference, but they do not guarantee a result or repair a measure that lacks evidence for the interpretation someone wants to make. A repeatable gain on a well-controlled, validated assessment can be meaningful evidence about the measured skill. The point is to match the claim to the quality and scope of that evidence.

The site's 50-question result needs the same distinction between observed performance and a population claim. It is a provisional estimated IQ range produced by an unvalidated conversion with no validated adult reference population. Its displayed band is not a confidence interval or an established rank against adults, and a higher retest result would not establish that a person's IQ changed. The free test can provide low-stakes reasoning practice and worked answers; its result should not carry a personal or consequential interpretation. “Free IQ Assessment: 50 Adaptive Questions, Estimated IQ Range | Test IQ Free” states that the optional paid report adds explanation, not accuracy or validity. Choose that report only if explanation is useful; it cannot turn the provisional estimate into a validated measure.

Sources: Working Memory Training Does Not Improve Executive Functioning or Fluid Intelligence; Free IQ Assessment: 50 Adaptive Questions, Estimated IQ Range | Test IQ Free

What is the proportionate next step?

Choose practice by the skill you want to use. If you want to hold and update information, work on a fresh sequence task; if you want stronger number, spatial, or verbal reasoning, practice that operation directly and check each solution. Check whether you can explain the method on new problems.

For low-stakes exploration, take the free 50-question test and review its worked answers. “Free IQ Assessment: 50 Adaptive Questions, Estimated IQ Range | Test IQ Free” describes the estimated range as provisional: it comes from an unvalidated conversion with no validated adult reference population, so it cannot establish an adult IQ or population standing. The optional $9 report can add explanation and structured practice, but payment does not make the estimate valid. A browser exercise is still a reasonable way to try reasoning questions when curiosity or practice is the goal.

If the answer could affect education, employment, diagnosis, or another consequential decision, ask a qualified professional about an assessment suited to that decision . Choose a fresh operation to practice, or arrange the assessment your consequential decision requires.

Sources: Free IQ Assessment: 50 Adaptive Questions, Estimated IQ Range | Test IQ Free