Cognitive Enhancement

The question is narrow and it is worth stating precisely: can a rested, healthy, unimpaired adult be made to think better by an intervention that is not practice? Three routes have been tried at scale. Drugs — modafinil, methylphenidate, amphetamine, and a long tail of supplements. Current — transcranial direct and alternating stimulation, delivered through scalp electrodes by researchers and by a consumer device market. Training — working-memory tasks sold on the promise that they generalise. All three have large literatures, and all three return effect sizes in a narrow band that a reader should memorise before going further: Cohen’s d or Hedges’ g between roughly 0.1 and 0.3, when they return anything. For scale, that is smaller than the effect of a night’s sleep on the same tasks, and about the size of the gap between a passive and an active control group.

Two things separate this subject from the rest of Category IV. The first is that the gap between public belief and published effect size is wider here than anywhere else on this site, so the brief’s job is arithmetic rather than rhetoric. The second is that one of the three routes has a complete, quantitative, published case that it is a null field — a mechanistic measurement showing that almost none of the applied current reaches a neuron, a quantitative review finding no significant effect across fifty-nine analyses, and a p-curve estimating the statistical power of the working-memory studies at a level indistinguishable from randomly generated data. That case is assembled here in one place, together with the strongest available rebuttal, because it is not assembled anywhere else that a general reader will find it.

Scope. The genetic route belongs to Biological Enhancement and the critical-period route to Neuroplasticity Engineering; this brief cross-references both rather than restating them, and owns the pharmacology, the stimulation literature and the training literature.

Established The prescription-stimulant meta-analysis is the anchor, and its authors’ own summary is that the effect is probably modest overall. Ilieva, Hook and Farah pooled forty-eight studies and 1,409 participants in the Journal of Cognitive Neuroscience in 2015: small but significant enhancement of inhibitory control and short-term episodic memory, a working-memory effect reaching significance on only one of two analytical approaches, and a medium-sized delayed-episodic-memory effect qualified by publication bias — which is a way of saying the literature it was pooled from is not clean. Frontier Their closing suggestion is the one that matters and is almost never quoted: that users may be boosting energy and motivation more than cognition. This brief prints no per-domain d values for that paper, because the numbers behind those characterisations were not obtainable. Established The broader survey by Smith and Farah in Psychological Bulletin in 2011 found declarative memory improvable by some stimulants, consistent with enhanced consolidation, executive and working-memory effects less reliable, and the whole enterprise not yet characterisable definitively despite the volume of research done. Frontier Fifteen years on, that last sentence still describes the field better than any single effect size.

Established The single most important distinction in this subject is the state of the user, and almost all popular writing collapses it. Repantis and colleagues, in Pharmacological Research in 2010, reviewed modafinil and methylphenidate together and wrote that expectations regarding the effectiveness of these drugs exceed their actual effects. Established For methylphenidate: an improvement of memory, and no consistent evidence for other enhancing effects. Established For modafinil: improved attention in well-rested individuals, and maintenance of wakefulness, memory and executive function in sleep-deprived individuals to a significantly higher degree than placebo. Established And then the sentence that should be this field’s epigraph — repeated doses of modafinil were unable to prevent deterioration of cognitive performance over a longer period of sleep deprivation, while maintaining wakefulness and possibly even inducing overconfidence in a person’s own cognitive performance. Frontier A drug that keeps you awake, lets you get worse, and raises your confidence that you have not, is not a cognitive enhancer in any sense a user would recognise. Speculative Whether that overconfidence generalises to the rested case is untested, and the experiment is cheap.

Established The review that produced the “first well-validated smart drug” coverage is a systematic review with no pooled effect size, and it reports harms. Battleday and Brem, in European Neuropsychopharmacology in 2015, covered modafinil in healthy non-sleep-deprived subjects from 1990 to 2014. Established On basic paradigms, most studies show enhanced executive function, only half show improvements in attention and learning and memory, and a few report impairments in divergent creative thinking; on more complex assessments they report consistent enhancement, with no preponderance of side effects. Established They also name two flaws in their own literature: discrepant methods, and reliance on psychometric tests designed to detect cognitive effects in ill rather than healthy populations. Frontier There is no pooled estimate anywhere in that paper, so any citation of it for “modafinil improves cognition by X” is citing a number that does not exist.

Frontier The supplement market divides into a short supported list and a long unsupported one. Supported, per meta-analyses indexed by a fetched tertiary survey: Established caffeine increases alertness and attention; Frontier amphetamine shows improvements in inhibitory control, episodic memory, working memory and aspects of attention, with methylphenidate in the same cluster; nicotine is associated with improved alertness, attention, memory and motor performance, with a 2020 review flagging industry funding concerns; and citicoline shows potential effectiveness in older people with mild cognitive decline, a different population from the one this brief is about. Established Unsupported: one meta-analysis found choline bitartrate ineffective at improving any measure of cognitive performance, there is little evidence that racetams enhance cognition in the unimpaired, and Centella asiatica, Ginkgo biloba, Panax ginseng, folate, omega-3 fatty acids and vitamins B6, B12 and E all lack convincing evidence in healthy populations. Frontier These are tertiary-indexed rather than primary-read, and each deserves tracing before it is leaned on.

Established Now transcranial direct current stimulation, where the case is complete and quantitative. Horvath, Forte and Carter, in Brain Stimulation in 2015, collected single-session tDCS data in healthy adults aged 18 to 50 for every cognitive outcome measure reported by at least two different research groups. Established Of the fifty-nine analyses conducted, tDCS had no significant effect on any, regardless of inclusion laxity, including no effect on any working-memory outcome or language production task. Established Their companion paper in Neuropsychologia states its result in its title: little-to-no reliable neurophysiologic effect beyond motor-evoked-potential amplitude modulation in healthy subjects. Frontier That abstract could not be retrieved for this brief and only the title is quoted, which is the honest position. Established The most sympathetic meta-analysis in the field — Mancuso, Ilieva, Hamilton and Farah in 2016, written expressly to correct what its authors saw as under-estimation by the null reviews, across thirty-one articles — found a small significant overall effect, Established found that the left dorsolateral prefrontal effect fell to non-significance after correction for publication bias, and concluded that the potential of tDCS probably lies in its use during working-memory training.

Established Then the p-curve, and then the mechanism. Medina and Cason, in Cortex in 2017, took a random selection of tDCS cognitive studies together with the working-memory studies inside Mancuso and colleagues’ meta-analysis and found no evidence that they had evidential value, with estimated statistical power of approximately 14% for the cognitive studies and 5% for the working-memory studies — and they state that 5% is what randomly generated data would produce. Frontier A body of work returning a stream of positive results at 5% power is not reporting a small true effect; it is reporting whatever survived selection. Established A year later Vöröslakos, Buzsáki, Berényi and colleagues, in Nature Communications, measured what actually reaches the brain: in rodents and human cadaver brains, approximately 75% of scalp-applied current is attenuated — 58 ± 7% by skin and soft tissue, 16 ± 8% by skull resistance. Established From recordings in rats, at least 1 mV/mm is necessary to affect neuronal spiking and subthreshold currents, which puts the requirement at approximately 6 mA applied transcutaneously; in living subjects, 4.5 to 6 mA or higher induced measurable phase-dependent modulation of alpha-band EEG. Established Conventional and consumer protocols use 1 to 2 mA — between a third and a sixth of what the measurement says is required.

Established Hold the clinical literature apart, because it is a different question with better evidence. An individual-patient-data meta-analysis of placebo-controlled tDCS trials in major depressive disorder — 572 participants across nine studies through December 2018 — reports response of 30.9% against 18.9% sham, odds ratio 1.96, number needed to treat 9; and remission of 19.9% against 11.7%, odds ratio 1.94, number needed to treat 13. Frontier Those figures reach this brief through a fetched encyclopaedia index citing Moffa and colleagues rather than from the paper itself, and should be traced. Established tDCS carries CE approval for major depressive disorder in the UK, the EU, Australia and Mexico. Established A technique can work for a disorder and do nothing for a healthy person; collapsing that distinction is the commonest error on both sides of this argument.

Established The training literature reached its verdict earlier and more decisively, and the market did not notice. The largest single trial, carried by Neuroplasticity Engineering, returned trained-task effect sizes of 0.73 to 1.63, transfer effect sizes of 0.01 to 0.22, and no evidence of generalised improvement across 11,430 completers. Established Melby-Lervåg, Redick and Hulme, in Perspectives on Psychological Science in 2016, analysed 87 publications with 145 experimental comparisons and found reliable improvement on intermediate transfer immediately after training, no convincing far transfer against a treated control, and, in mediation analysis, no relation between the degree of improvement on working memory and the magnitude of far transfer. Frontier That last result kills the causal story and is almost never quoted: even inside the studies reporting transfer, the people who improved most at the trained task were not the people who transferred most. Established Au and colleagues, restricting to n-back training with a fluid-intelligence outcome across 1,022 participants, return Hedges’ g = 0.24 overall — roughly 3 to 4 IQ points on their own conversion — but 0.44 against passive controls and 0.06 against active ones at p < .01, with international studies at 0.44 against US studies at 0.06, and participant remuneration negatively correlated with effect size at p = .05. Frontier An effect of 0.44 against a passive control and 0.06 against an active one is, to a first approximation, an expectancy effect, and paid participants have no reason to want the training to have worked. Established Simons and colleagues then evaluated every peer-reviewed study cited on the leading brain-training companies’ own websites and found extensive evidence for the trained tasks, little for anything distant, and none conforming to best practices.

Established Against all of that, here is what does work, and it is unglamorous. The best-replicated cognitive intervention in this brief is playing a learned cue quietly to a sleeping person: a meta-analysis of targeted memory reactivation across 91 experiments, 212 effect sizes and 2,004 participants returns Hedges’ g = 0.29, 95% CI 0.21 to 0.38, with NREM stage 2 at 0.32 and slow-wave sleep at 0.27, Established and nothing at all in REM sleep or during wakefulness — the internal control that makes it credible. Frontier The honest summary of the whole subject is that the two interventions with the best evidence are sleep and practice, they are free, they have no commercial sponsor, and neither generalises. Speculative Aerobic exercise is routinely named beside them; this brief could not verify a meta-analytic effect size for exercise on cognition in healthy adults from a primary source and therefore states none, which is itself a finding about how casually that baseline is invoked.

Frontier The tDCS field’s future depends on direct pre-registered replications, and they are starting to appear. Jiang, Jones and von Bastian posted a replication study of tDCS effects on visual working memory as a Research Square preprint in 2023. Frontier The preprint was provenance-blocked to the research behind this brief and no result is quotable here, which is stated rather than papered over. Frontier The structural point stands regardless: whether this literature survives depends on results of that shape rather than on further meta-analyses of the existing corpus, because meta-analysing a set of studies with 5% power produces a more precise estimate of a selection process, not of an effect.

Frontier Temporal interference stimulation is the first genuinely new non-invasive modality in two decades and its first cognitive result is positive. Violante and colleagues, in Nature Neuroscience in 2023, applied two kilohertz-range fields — 2.005 kHz and 2 kHz, a 5 Hz difference frequency — so that only the envelope at the difference frequency reaches the target, with current densities of 0.9 and 1.35 mA/cm². Established They validated the field by electric-field modelling and by measurement in a human cadaver, then ran n=20 for fMRI and n=21 for behaviour. Frontier Temporal interference stimulation of the hippocampus increased the odds of correct recall by 12%, 95% CI 0.02 to 0.21. Frontier One group, one protocol, and an effect at the low end of what the tDCS literature also claimed before it collapsed — but obtained by measuring the field in a human head rather than modelling it, which is exactly what the tDCS literature never did.

Frontier The mechanistic critique has hardened and has not been rebutted. The 2018 measurement is now eight years old, open-access, heavily cited and published in a leading journal. Frontier The research behind this brief found no published demonstration that 1 to 2 mA scalp stimulation produces intracranial fields above 1 mV/mm in a living human. Speculative Two counter-arguments exist in the literature and both are coherent. The first is that sub-threshold fields could bias network-level activity without driving spikes, which would make the 1 mV/mm spiking criterion the wrong test. Speculative The second is that effects accumulate over minutes of stimulation rather than acting instantaneously, so an instantaneous threshold is not the relevant quantity. Frontier Neither has been demonstrated to produce a behavioural effect that survives an active control, and the efficacy literature has largely proceeded as though the measurement did not happen — which is a finding about a field’s behaviour rather than about a number, and the more damning of the two.

Established In the training literature the question has moved from “does it transfer” to “why did anyone think it did,” and the answer is control-group design. Established Au and colleagues give 0.44 against passive controls and 0.06 against active ones; Established Melby-Lervåg and colleagues find no convincing far transfer against treated controls; Established Simons and colleagues find that none of the industry-cited studies conformed to best practices. Frontier Frontier status here is unusual: the science is settled among methodologists and unsettled in the market. Established US consumer spending on brain training ran from about $2 million in 2005 to roughly $80 million in 2007, $1 billion by 2012 and $1.3 billion in 2013. The industry did not shrink when the evidence did.

Frontier Closed-loop and personalised stimulation is where the electrical route is heading, and nothing has landed. The precedent is clinical rather than cognitive: adaptive deep brain stimulation is approved and deployed, and closed-loop stimulation triggered on a decoded encoding state produced the best human memory result of the DARPA era. Frontier No closed-loop non-invasive cognitive-enhancement result exists in the literature assembled for this brief. Speculative The obvious experiment — trigger alternating-current or temporal-interference stimulation on a decoded cognitive state rather than applying it open-loop — has not been reported, and it is the one design that would answer the sub-threshold-accumulation steelman on its own terms.

Frontier What is conspicuously missing is cheaper than what has been done. There is no large pre-registered multi-site trial of tDCS for cognition in healthy adults. Frontier There is no dose-response study relating measured intracranial field strength to a behavioural outcome in humans. Frontier There is no enhancement study with a real-world outcome — grades, job performance, error rates — rather than a laboratory task. Frontier And there is no study measuring whether users’ confidence rises more than their performance, despite that specific risk being flagged in print in 2010. Speculative The last of these requires adding one self-rating scale to an existing protocol and would probably be the most informative study in the field.

Established The workback plan to durable, general enhancement in a healthy adult has an unusual shape: the binding link is not something nobody has tried, but something the field has repeatedly measured and repeatedly failed. Ten links, each with a status.

Frontier 1. An instrument that is not a deficit test. The field measures with psychometric batteries designed to detect impairment in ill populations, which Battleday and Brem name explicitly as a core problem. Frontier A test with a ceiling cannot measure headroom. Status: not done, and invisible because everyone uses the same instruments. Frontier 2. A separation of performance from confidence. Status: essentially never done, despite being flagged in 2010. Established 3. A demonstrated effect against an active control in a pre-registered protocol. Status: for tDCS in healthy cognition, no; for stimulants, small effects surviving in some domains and not others; for training far transfer, no. Frontier 4. A dose-response relation tied to a measured biological quantity. Status: never done for tDCS. Vöröslakos and colleagues measured the field but not the behaviour; Violante and colleagues measured the field in a cadaver and the behaviour in living subjects, the closest anyone has come.

Frontier 5. Persistence beyond the intervention. Status: the training literature says gains decay and never transferred anyway; the drug literature does not test it; the one human critical-period trial states that duration of effect was not tested. Established 6. Generality across domains. Status: 87 publications and 145 comparisons say no, and the mediation analysis says the mechanism does not look like a transfer effect. Speculative 7. An identified biological brake and a way to modulate it. Status: perineuronal nets are identified, chondroitinase ABC works in animals, there is no human delivery route, and the one human trial is a 24-person valproate crossover whose second arm was null. Frontier 8. Absence of a compensating decrement. Status: not attempted; the only signal is the divergent-creative-thinking impairment noted in the modafinil review. Established 9. A real-world outcome: none. 10. A regulatory and social frame: no approval anywhere for a cognitive-enhancement claim, two settlements against enhancement marketing, and a professional-body position discouraging prescription to the healthy.

Established Which link binds. Link 6, generality, is the binding scientific link, and the field’s own strongest evidence says it fails: working-memory training does not transfer, and the degree of training improvement does not predict the transfer that is claimed. Frontier Every general-enhancement programme has to explain why its intervention would generalise when practice — the most powerful cognitive intervention known — does not. Frontier Link 3 binds in practice: nothing in the tDCS enhancement literature has cleared an active control in a pre-registered study, and until something does the field cannot distinguish itself from expectancy. Speculative Link 1 binds the whole measurement enterprise, and link 2 is the cheapest and would be the most informative — report confidence alongside performance and the field would learn in a year what it has not learned in twenty.

Frontier This brief depends on results that other fields are producing, and on one that nobody is. The scientific dependency runs to Neuroplasticity Engineering, which owns the only mechanism with a serious advocate that could evade the negative results assembled here: an identified developmental brake on adult learning, and an agent that releases it. Speculative If adult cognition is stabilised rather than optimised, the headroom is whatever the brake is costing, and the brake is that brief’s subject rather than this one’s. Speculative The genetic and germline route belongs to Biological Enhancement; nothing in the pharmacology or stimulation literature bears on it and this brief makes no claim there.

Frontier The measurement dependency is the one that blocks everything. Until a cognitive battery exists with adequate individual-level reliability, no ceiling in healthy adults, and demonstrated sensitivity to a known manipulation, no enhancement claim in this brief can be checked at the level of a person rather than a group mean — a problem shared with Intelligence Measurement, which is assembling exactly that methodology for a different purpose. Frontier A night of sleep deprivation is the obvious calibrator, because it is the one manipulation in this subject with an effect large enough to validate an instrument against.

Established The meta-scientific dependency is institutional rather than technical. Pre-registration, active controls, and publication of null results are not instruments that need inventing; they are conventions that need enforcing, and their absence is the direct cause of the 5% power estimate. Frontier No new physics, no new chemistry and no new hardware is required to answer the central question of this brief. What is required is a different publication norm, which is why the constraint appears in this brief’s tech tree as an institutional one.

Speculative Six experiments, in the order that would extract the most information per dollar. None is expensive by this category’s standards, and the cheapest is the most informative.

Speculative One: the confidence ratio. In any enhancement study, administer the intervention, measure objective performance and self-rated performance on the same task, and publish the ratio. Frontier Almost no study in this literature reports both, despite the overconfidence risk being named in print in 2010. Speculative If subjective ratings rise more than objective scores, the mechanism is arousal and motivation rather than cognition, and the population-level consequence of widespread use is worse decisions made more assertively.

Speculative Two: the trial the tDCS field owes itself. A pre-registered, multi-site, active-controlled trial in healthy adults, powered for d = 0.2, with a behavioural primary endpoint, individual current-flow modelling, and ideally in-vivo field measurement. Frontier It has never been run, and it would cost less than a single brain-computer-interface implant trial. Speculative Three: dose-response against a measured field. Relate the measured intracranial field to the behavioural effect in the same individuals; for drugs, relate plasma concentration or receptor occupancy to the effect. A monotonic relation is the pass condition and no modality in this brief has one.

Speculative Four: a battery broad enough to detect what got worse. The field’s test batteries are selected for the expected gain, which is why trade-offs are invisible; the divergent-thinking impairment is the only decrement anyone has stumbled into. Speculative Five: persistence. Measure the effect 30 and 90 days after the last dose or session. Frontier The longest post-intervention interval anywhere in the human memory-stimulation literature reached for this brief is fifteen to twenty minutes. Speculative Six: a real-world outcome. Error rates in air-traffic control, diagnostic accuracy in radiology, defect rates in code review — an outcome with adequate individual-level reliability over days. Frontier Consent, confounding and cost are why nobody does it; that the sleep-deprived case is the only one where the drugs clearly work, and also the only case resembling how people actually use them, is why somebody should.

Speculative And one weekend’s work with high option value: apply p-curve analysis to the transcranial alternating-current and transcranial-ultrasound literatures. Speculative If the structural features that produced the tDCS collapse — small samples, flexible analysis, no active control, no pre-registration, a plausible mechanism story and a commercial incentive — are present in a neighbouring field, the same tool should find the same thing, and nobody has published it.

Established The hardware in this subject is trivial and that is precisely the problem. A tDCS device is a current source, two saline sponge electrodes of about 35 cm², and a timer. Established Protocols delivering up to 4 mA over two weeks in sessions of 20 to 30 minutes were considered within the safety envelope as of 2017. Established Reported adverse effects are skin irritation, phosphenes, nausea, headache, dizziness and itching; people susceptible to seizures should not receive it; long-term adverse effects were unknown as of 2017. Established In rats, cathodal stimulation at a current density of 142.9 A/m² or a charge density of 52,400 C/m² produced brain lesions — more than two orders of magnitude above the protocols in human use. Frontier The engineering conclusion is uncomfortable and rarely stated: these devices are almost certainly safe, and the same measurement that makes them safe is the one that suggests they are inert.

Frontier The engineering that would matter is measurement, not delivery. Three capabilities are missing. Frontier First, in-vivo intracranial field measurement in living humans at the intensities actually used, which currently exists only opportunistically in patients with implanted electrodes. Frontier Second, per-subject current-flow modelling from individual anatomy, so that a trial does not average across heads in which the field lands in different places — the strongest technical form of the pro-tDCS rebuttal. Speculative Third, ambulatory cognitive measurement with individual-level reliability, which is a software and psychometrics problem rather than a hardware one and is the binding requirement for any real-world trial.

Frontier Temporal interference stimulation raises the hardware bar and the raised bar is the point. It requires two independently controlled kilohertz-range current sources, precise electrode montage, field modelling validated against a physical measurement, and a target deep enough that no simpler modality reaches it. Frontier That is a substantially more demanding instrument than a two-electrode direct-current box, which is one reason its first result is more credible than a decade of tDCS results and also why it will not become a consumer product on the tDCS timescale.

Frontier Neuromodulation for disorders is the adjacent field that calibrates what a large effect looks like. Accelerated intermittent theta-burst stimulation reported 79% remission in a 14-participant trial and 16.1% acute remission in a 155-participant one; reading those two together before writing any sentence about neuromodulation effect sizes is the single most useful discipline available in this area. Established The pattern — a spectacular small-n result followed by a modest large-n one — is the same pattern the tDCS enhancement literature produced, with the difference that the depression literature ran the larger trial.

Frontier The interface route is adjacent and its ceiling is informative. The highest reported information transfer rate for a non-invasive brain-computer interface in a recent clinical review is 360 bits per minute at a pooled speller accuracy of 74% — roughly 60 to 70 English characters per minute, which is about what a healthy adult achieves by typing. Frontier Brain-Computer Interfaces and Neural Interfaces carry the invasive side; the honest read for enhancement is that a non-invasive interface adds no bandwidth a healthy adult does not already have.

Frontier Memory stimulation is adjacent and its numbers belong beside these ones. Memory Engineering carries the closed-loop result — a 15% relative increase in item recall probability at p = 0.02 with an odds ratio of 1.18, in 25 epilepsy patients with implanted electrodes. Frontier Set that against 12% odds increase from scalp electrodes in 21 healthy people, and against g = 0.29 from a sound played during slow-wave sleep in 2,004 people. Frontier Either the invasive route is being wasted, or all three are measuring the same small non-specific effect. Speculative Psychedelics as plasticity agents and the pharmacological reopening of critical periods sit in Neuroplasticity Engineering, and they are where the only serious “headroom” argument lives.

Established The regulatory asymmetry is the central institutional fact and almost nobody states it. Where approval for transcranial stimulation exists, it is for a disorder: CE marking for major depressive disorder in four jurisdictions. Established There is no regulatory approval anywhere for a cognitive-enhancement claim. Frontier A December 2025 approval of a home tDCS device for depression in the United States has been reported; this brief could not verify it independently, attributes it rather than asserting it, and notes that it would again be an approval for a disorder.

Established Enforcement has happened, and it happened in the training market rather than the device market. The FTC settled with Lumosity in January 2016 — $2 million, reduced from a $50 million judgment — over claims that its games could prevent memory loss, dementia and Alzheimer’s disease, with an order prohibiting claims of improvement at school, work or in athletics without competent and reliable scientific evidence. Established It settled with LearningRx the same year for $200,000 of a $4 million judgment over “clinically proven” claims, with future claims requiring randomised controlled trials. Established The FDA and FTC warned nootropic-supplement manufacturers about advertising fraud in 2019, and the American Medical Association discouraged prescribing nootropics to healthy people in 2016. Frontier Every one of those figures reaches this brief through an encyclopaedia index rather than a regulatory document, and should be traced before the dollar amounts are relied on.

Frontier The gap is the consumer stimulator. A device sold as wellness rather than as a treatment for a condition falls outside device regulation in most jurisdictions, which means the one product category whose mechanism has been measured and found insufficient is also the one facing no efficacy review at all. Speculative Bringing wellness stimulators inside device regulation would not require proving them harmful — the safety record is good — only requiring that an enhancement claim be substantiated to the standard a therapeutic claim already must be.

Established The second institutional constraint is a publication norm. A p-curve estimating 5% power is not a finding about tDCS; it is a finding about what a literature looks like when small studies with flexible analysis and no pre-registration are published only when positive. Frontier Pre-registration as a condition of publication, with active controls and mandatory reporting of nulls, is the intervention that would have prevented a decade of this and is the only one that will prevent the next one. Frontier On neurorights and neural-data law more broadly, the research behind this brief established no jurisdiction, statute or case as fact, and this brief therefore states none.

Established The standing policy argument is fifteen years old and it is a good one. Greely and colleagues, in Nature in 2008, argued for regulated acceptance of cognitive-enhancing drugs by the healthy rather than prohibition, treating enhancement as continuous with education, caffeine and eyeglasses, and treating safety, medicalisation and equity objections as manageable through oversight. Frontier A widely quoted informal poll of researchers published alongside it is often cited for a use-prevalence figure; this brief could not verify those numbers and does not print them.

Speculative The strongest objection is not the one the ethics literature usually makes, and it does not depend on the drugs being dangerous. If the effect is on arousal, motivation and confidence rather than on cognition, then widespread use raises assertiveness without raising accuracy, and the resulting equilibrium is worse than the status quo for everyone including the users. Speculative That is an empirical claim with a cheap test attached — the confidence ratio — and it inverts the policy case rather than merely qualifying it. Speculative The second objection is the trade: if cognitive resources are allocated rather than created, a population that has traded divergent thinking for executive control has made an irreversible collective choice through individually rational decisions, and the only signal anyone has looked for is a single line in a modafinil review.

Frontier The equity argument cuts differently once the effect sizes are on the table. A great deal of writing worries about unequal access to enhancement; at g = 0.06 against an active control, the thing being unequally accessed is close to nothing, and the more defensible worry is the transfer of money and attention from interventions that work — sleep, practice, treating the sleep-deprived — to interventions that are marketed. Established The population most clearly helped by these drugs is the sleep-deprived, and a society that responds to sleep deprivation by medicating it rather than reducing it has answered a question it did not ask.

Speculative If enhancement worked, the interesting consequence would be compositional rather than individual. Three or four IQ points in one person is a rounding error; the same shift across a population changes the tail, and the tail is where research output, catastrophic error and institutional competence live. Frontier That is why the effect sizes matter more than the ethics: at 0.24 against a passive control and 0.06 against an active one, there is no population-level shift to argue about.

Speculative The reframe this brief recommends is that the interesting question is not whether people can be made smarter but whether practice can be made cheaper. The evidence is consistent with a specific narrow claim: little headroom for general enhancement by a single systemic intervention in a rested, healthy, unimpaired adult, and substantial headroom for specific capability acquired by practice — which is exactly what the transfer literature shows when it is read as a positive result rather than a negative one. Speculative Halving the hours to expertise would be a larger civilizational intervention than any molecule in this brief, and it is Neuroplasticity Engineering’s target rather than this one’s.

Frontier The durable lesson may be meta-scientific rather than cognitive. tDCS accumulated a decade of positive small-n studies, a favourable meta-analytic consensus, a consumer-device market and clinical enthusiasm, and then a p-curve estimated its working-memory power at 5%. Speculative The structural features that produced that are present in several neighbouring fields, and a civilization that funds research faster than it audits it will produce this outcome repeatedly. Frontier The cheapest civilizational upgrade available in this subject is not a drug; it is making pre-registration a condition of publication.

What these horizons track is not the arrival of an enhancement technology but the arrival of the evidence that would tell us whether one is possible. The dates below are dates for results, not for products.

• 10 yr: Frontier A pre-registered, multi-site, active-controlled tDCS trial in healthy adults with individual current-flow modelling either lands or continues not to exist; on present funding patterns the second is more likely than the first. Frontier Temporal interference stimulation is independently replicated by a second group, or it is not, and that single result determines whether non-invasive deep stimulation is a field. Speculative The confidence-versus-performance ratio starts being reported, because it costs one scale and one extra column.
• 25 yr: Speculative A cognitive battery with individual-level reliability above 0.8 and no ceiling in healthy adults exists and is standard, which would retroactively make most of the existing literature uninterpretable rather than merely underpowered. Speculative The first real-world enhancement outcome study — error rates in an actual high-stakes task over weeks — is published, most plausibly in a safety-critical industry that already measures errors for other reasons. Handwave Consumer stimulators are either regulated as devices or have disappeared from the market for commercial reasons.
• 50 yr: Speculative If any general enhancement exists at all, it comes from the plasticity route rather than the pharmacological or electrical one: an identified brake on adult learning, a deliverable modulator, and demonstrated selectivity. Handwave Each of those three is independently unsolved today and the delivery problem for the best animal candidate has no human route. Speculative The more likely fifty-year outcome is that “cognitive enhancement” has been retired as a category and replaced by accelerated skill acquisition, which is a different and more tractable target.
• 100 / 250+ yr: Handwave Durable, general, drug-free elevation of an adult’s cognitive ceiling requires five things to hold at once: an identified biological brake, a systemically deliverable agent, selectivity of the reopened window, persistence after withdrawal, and generality across domains. Handwave Links two, three and five are each independently unsolved, and link five is the one that 87 publications and 145 comparisons say is false. Cheerfully far from evidence, and included because pretending otherwise would be the more dishonest option.

• Depends on This brief depends on Neuroplasticity Engineering more heavily than on anything in its own three literatures. The pharmacological, electrical and training routes have each been measured to their limit; what would change the answer is a mechanism that evades a well-replicated negative result, and the only candidate with a serious advocate is critical-period reopening — an identified developmental brake on adult learning, an agent that modulates it in human cortex, and selectivity in what the reopened window admits. That brief owns all three and has one small human trial to its name. The second dependency is quieter: every claim here is a group mean on an instrument built to detect deficits in ill populations, so until a battery exists with individual-level reliability and no ceiling in healthy adults, no enhancement claim can be checked at the level of a person — which is Intelligence Measurement's methodology, built for a different purpose. Biological Enhancement carries the genetic route, on which nothing here bears. None of this dependency structure is technological: no new instrument, material or synthesis is required to answer the central question.
• Requires (not on this map) Neither constraint is scientific and neither is expensive. The first is a publication norm: a p-curve returning 5% estimated power is not a fact about transcranial stimulation but a fact about what gets published when small, flexibly analysed, unregistered studies reach print only when positive. Pre-registration as a condition of publication, with active controls and mandatory reporting of nulls, is the single change that would have prevented this decade and the only one that will prevent the next. The second is a regulatory boundary: stimulators sold as wellness products sit outside device regulation in most jurisdictions, so the one product category whose mechanism has been measured and found insufficient faces no efficacy review at all, while the same technology marketed for depression must clear a therapeutic bar. Closing that gap requires no finding of harm — the safety record is good — only that an enhancement claim be substantiated to the standard a therapeutic claim already meets.
• Enables A negative result of this quality is itself an enabler, in two directions. Inside the subject, it redirects effort from general enhancement toward accelerated skill acquisition, which the transfer literature says is where the headroom actually is. Outside it, the tDCS case supplies a worked template that Memory Engineering, Brain-Computer Interfaces and Neural Interfaces can all use: a literature with small samples, flexible analysis, no active control, no pre-registration, a plausible mechanism story and a commercial incentive will produce a favourable consensus that a p-curve can dissolve in an afternoon. Applying the same tools to the transcranial alternating-current and ultrasound literatures is a weekend's work with large option value, and is named here as the highest-leverage unclaimed result in the area.
• Adjacent Temporal interference stimulation is the nearest adjacent technology and the only one whose first cognitive result came from measuring the field rather than modelling it. Adaptive deep brain stimulation supplies the closed-loop precedent that non-invasive stimulation has not borrowed. Accelerated theta-burst protocols for depression supply the calibration for a genuinely large neuromodulation effect, and the discipline of reading a 14-participant result beside a 155-participant one. Targeted memory reactivation during slow-wave sleep is adjacent in an awkward way: it is the best-replicated cognitive intervention here, it costs nothing, and it has no commercial sponsor — the clearest illustration of how poorly funding tracks evidence in this subject.

Frontier “Modafinil is the first well-validated smart drug.” The review that generated that coverage is a systematic review with no pooled effect size. Its own findings: on basic paradigms most studies show executive-function enhancement, only half show improvements in attention and learning and memory, and a few report impairments in divergent creative thinking. Established A separate review and meta-analysis of the same drug concludes that expectations regarding effectiveness exceed actual effects. Frontier Never attach a number to the first of those papers, and let the divergent-thinking finding travel with every citation of it.

Established “Smart drugs work” and “smart drugs do nothing” are wrong in the same way: both ignore the state of the user. Modafinil improves attention in well-rested people and maintains wakefulness, memory and executive function in sleep-deprived people better than placebo, Established but under sustained sleep deprivation repeated doses keep the user awake without preventing cognitive deterioration, and possibly induce overconfidence in their own performance. Frontier The sleep-deprived case and the rested case are different questions with different answers.

Established “Stimulants make healthy people smarter.” Forty-eight studies and 1,409 participants return small significant effects on inhibitory control and short-term episodic memory, a working-memory effect significant on only one of two analytical approaches, and a delayed-episodic-memory effect qualified by publication bias; the authors’ own conclusion is that the effect is probably modest overall and may be energy and motivation rather than cognition. Frontier One methodological note: the machine-readable record for that paper in a major bibliographic database returns the correct abstract attached to the metadata of an unrelated Chinese-language library-science article. The citation here was confirmed against a second source; in this cluster, no full citation should come from a single database record.

Established “tDCS enhances cognition — hundreds of studies show it,” and “a 2 mA headset changes your brain.” Across fifty-nine analyses covering every cognitive outcome reported by at least two independent groups in healthy adults, single-session tDCS had no significant effect on any, regardless of inclusion laxity; Established a p-curve estimates the power of those studies at approximately 14% for cognition and 5% for working memory, the latter being what randomly generated data would produce; Established the companion review found little-to-no reliable neurophysiological effect beyond motor-evoked-potential amplitude; Established and roughly 75% of scalp current is attenuated, at least 1 mV/mm is needed to affect spiking, and about 6 mA is required to reach it, against the 1 to 2 mA that consumer and research devices deliver. Frontier Hundreds of studies at 14% power is not hundreds of pieces of evidence; it is one selection process observed hundreds of times.

Speculative The strongest steelman for tDCS, given its due. Frontier The depression evidence is a different literature with individual-patient-data meta-analysis, numbers needed to treat of 9 and 13, and CE approval in four jurisdictions — and a technique that does nothing cannot produce that. Frontier The null review’s inclusion rule discards the multi-session and within-subject literature, which is where the sympathetic meta-analysis locates the effect. Speculative Sub-threshold fields may bias network dynamics without driving spikes, making the spiking criterion the wrong test; and individual anatomical variation in current flow is large, so averaging across unmodelled heads would dilute a real effect toward zero. Frontier The adjudication: the enhancement literature in healthy adults is most likely null or so small as to be practically indistinguishable from null; the clinical literature in depression is a live question with better evidence; and those rebuttals are reasons to run one specific trial rather than reasons to believe the existing corpus. Nobody has run it.

Established “Brain training makes you smarter” — and its mirror, “133 scientists signed a letter saying it does.” The n-back meta-analysis returns g = 0.24 overall, 0.44 against passive controls and 0.06 against active ones; Established 87 publications and 145 comparisons find no convincing far transfer against treated controls, and no relation between working-memory improvement and far transfer; Established the evaluation of the industry’s own cited studies found none conforming to best practices. Frontier The 133 signed in 2014, months after more than seventy scientists signed the opposite, through a site hosted by a brain-training company’s chief scientific officer. Speculative One real objection survives on that side: an active control involving engaging cognitive activity may itself produce transfer. Settling that requires knowing what an inert cognitive placebo is, and nobody does.

Established “The regulators have not weighed in” — and “nothing improves cognition in healthy adults.” The FTC settled with Lumosity in January 2016 for $2 million, reduced from a $50 million judgment, and with LearningRx the same year for $200,000 of a $4 million judgment; the FDA and FTC warned nootropic manufacturers in 2019 and the American Medical Association discouraged prescribing to healthy people in 2016. Frontier Those figures are indexed rather than read from the orders and should be traced. Established On the sceptic’s side: targeted memory reactivation during slow-wave sleep gives g = 0.29 across 2,004 participants and nothing in REM or wakefulness, and deliberate practice gives trained-task effect sizes of 0.73 to 1.63. Both are real; neither is general; and g = 0.24 is about 3 to 4 IQ points on the authors’ own conversion, dropping to 0.06 against an active control.

Speculative The last misconception is the one this brief is least sure about: that the absence of measured enhancement proves the absence of headroom. The case for no headroom is strong — cognition is under intense selection, so a free lunch would probably have been found, and every apparent enhancement looks like restoration toward an individual’s own ceiling, or a trade. Speculative But the measurement argument cuts both ways: if the instruments are deficit tests with ceilings, they cannot detect enhancement or rule it out, and the field has been arguing about the sign of a quantity it cannot measure. Frontier The strongest pro-headroom argument is not pharmacological — adult cognition is stabilised rather than optimised, critical periods close by an active mechanism, and removing the brake restores juvenile plasticity in animals. The second is an existence proof: expertise, literacy and musical training produce large durable differences, so something makes a grandmaster. Speculative Declare the tie honestly. There is little headroom for general enhancement by a single systemic intervention in a rested, healthy, unimpaired adult, and substantial headroom for specific capability acquired by practice. The interesting question was never whether people can be made smarter. It is whether practice can be made cheaper.