The cognitive enhancement divide is usually discussed as a future scenario: brain-computer interfaces, genetic editing, neural implants arriving someday and splitting humanity into enhanced and unenhanced castes. The divide is already here. It arrived through a subscription model.
A $20 AI subscription provides analytical capacity that, five years ago, required an expensive consultant or a graduate degree. A $200 subscription accesses meaningfully more capable models with higher usage limits. Enterprise deployments, costing $85,000 per month on average in 2025, integrate AI into workflows that learn from proprietary data, operate continuously, and compound organizational intelligence in ways no individual subscription can match. The person using free-tier AI to draft a cover letter inhabits a different cognitive world from the firm using enterprise AI to screen ten thousand applicants per hour. This is cognitive enhancement stratified by price, and the performance data confirms the gap is real. A landmark study published in the Quarterly Journal of Economics (Brynjolfsson, Li, and Raymond, 2025) tracked 5,172 customer support agents after the introduction of a generative AI assistant and found a 15 percent average productivity increase, with gains of 30-35 percent for the least experienced workers. AI compressed the skill distribution within the occupation, functioning as a cognitive equalizer at the task level. A Harvard Business School field experiment found that management consultants using AI completed tasks 25 percent faster at 40 percent higher quality when operating within AI’s capability frontier.
These within-occupation gains are genuine. They are also misleading if extrapolated to economy-wide effects. A meta-analysis in the California Management Review (October 2025) pooling 371 estimates found no robust, publication-bias-free relationship between AI adoption and aggregate productivity growth. The task-level equalization coexists with occupation-level contraction and capital-level concentration. AI may close the gap between the worst and best customer service agent inside customer service, while the customer service occupation itself shrinks and the productivity surplus flows to shareholders. The Anthropic labor market study (March 2026) found that workers in the most AI-exposed occupations earn 47 percent more than the zero-exposure group. The cognitive enhancement is reaching the already-advantaged first, and the returns from that enhancement are accruing to capital. There is also a question of whether AI-assisted productivity constitutes genuine cognitive enhancement or cognitive outsourcing, a distinction explored in Shade #6 (The Cognitive Atrophy Trap). The Brynjolfsson study found that AI “captured and disseminated the patterns of behavior that characterize the most productive agents,” but whether lower-skilled workers internalized those patterns or merely had them performed on their behalf determines whether the gains survive if the tool is removed. The divide may not close when access widens if the “enhancement” was dependency all along.
The AI-as-cognitive-enhancement channel is the most immediate, but three other pathways are converging on the same divide.
The first is neuropharmacological. GLP-1 receptor agonists, originally approved for diabetes and obesity, are now under investigation for neuroprotective effects across Alzheimer’s, Parkinson’s, and general cognitive impairment. A review in Frontiers in Neuroscience (2025) documented that liraglutide enhanced brain activation and restored cognitive domains in clinical trials, with Phase III EVOKE trials evaluating oral semaglutide for early Alzheimer’s now underway. These drugs are already prescribed to tens of millions of people worldwide for weight management. The cognitive enhancement potential is a side channel of a mass-market pharmaceutical. Meanwhile, the use of prescription stimulants by healthy individuals for cognitive performance is widespread and growing. A narrative review in Biology (September 2025) documented that methylphenidate, modafinil, and amphetamines are the most commonly self-administered smart drugs, used under academic and professional pressure, though evidence for their effectiveness in healthy users remains unclear. The access pattern is stratified: professionals with good healthcare, disposable income, and awareness of these compounds use them. Workers without those resources do not.
The second pathway is genetic. Polygenic embryo screening for cognitive ability is already commercially available in the United States. Genomic Prediction has offered screening for disease risk since 2019. In October 2025, Herasight launched CogPGT 1.0, a polygenic predictor claiming to detect an average 8.5 IQ point difference between three embryos. A study in Cell (Karavani et al., 2019) calculated that current polygenic predictors yield an average gain of approximately 2.5 IQ points per selection cycle, accompanied by wide confidence intervals. The New England Journal of Medicine (2021) warned that this technology is essentially unregulated in the U.S. and urged a “society-wide conversation” about ethical and regulatory frameworks. Most European countries restrict genetic testing to disease conditions. The United States has no federal restriction on polygenic screening for non-disease traits. At least one company’s co-founder has publicly speculated about offering screening for above-average cognitive ability and skin color. The current gains are modest. The predictive power of polygenic scores improves with larger genomic datasets. The technology becomes more powerful over time while governance remains absent. Unlike AI subscriptions or pharmacological enhancement, genetic selection compounds across generations. An advantage embedded at conception is inherited. The regulatory divergence across jurisdictions compounds this risk. Most European countries restrict embryo screening to disease conditions. The United States has no federal restriction. China banned all clinical research in germline genome editing as of July 2024, but polygenic screening of existing embryos for non-disease traits operates under a different and less clearly defined regulatory framework. The Hastings Center has noted that polygenic screening is essentially unregulated in most countries, creating conditions for jurisdictional shopping: wealthy parents travel to permissive regimes, or permissive regimes produce populations with compounding generational advantages that restrictive regimes cannot match. This connects to #7 (The Geopolitical AI Arms Race) and #25 (Sovereignty Erosion), where regulatory divergence becomes competitive divergence.
The third pathway is neural interfaces. Neuralink has implanted devices in approximately 20 participants across the U.S., UK, Canada, and UAE as of early 2026, all for medical conditions: paralysis, ALS, spinal cord injury. The FDA granted Breakthrough Device Designations for speech restoration (May 2025) and vision restoration through the Blindsight system (June 2025). Participants have demonstrated thought-controlled computer use, robotic arm manipulation, and video game play. Musk announced plans for high-volume production and automated surgical procedures in 2026. The technology is real, advancing rapidly, and entirely medical. Enhancement of healthy brains remains speculative. The leap from restoring lost function to augmenting normal function requires neuroscience breakthroughs that may be decades away. The 35 percent likelihood for this shade already accounts for that uncertainty. What matters for the governance question is that the regulatory framework is being built around a treatment paradigm. The FDA approves devices that restore function to patients with diagnosed conditions. There is no approval pathway for making a healthy person’s memory faster or their attention span wider. When the technology does cross from treatment to enhancement, the regulatory vacuum will already exist.
Dario Amodei’s “Machines of Loving Grace” (2024) envisions AI-driven neuroscience compressing a century of progress into 5-10 years, including effective treatments for most mental illness and potential enhancements to everyday cognitive function. He acknowledges that without deliberate policy, such advances could be “only for the rich.” The convergence of these four pathways (AI tools, pharmacology, genetics, neural interfaces) makes the policy question urgent even though the most dramatic technologies are not yet mature. Each pathway independently creates cognitive stratification. Together, they create the conditions for a divide that compounds across every dimension of advantage simultaneously: the wealthy family accesses enterprise AI, prescription cognitive enhancers, polygenic embryo screening, and eventually neural augmentation, while the family without those resources accesses free-tier AI with usage caps and whatever the public education system provides. The compounding across channels matters more than any single channel. A modest AI productivity gain, a modest pharmacological boost to focus, a modest polygenic IQ advantage, and eventually a modest neural interface acceleration, each measured in single-digit percentages, interact multiplicatively during the same developmental window. The child receiving all four is not 4x advantaged over the child receiving none. The child is operating within a different cognitive trajectory, one where each enhancement reinforces the capacity to benefit from the others.
The 7-point governance dividend is the highest in the collection after the Intelligence Explosion (#21). Universal cognitive enhancement would be the greatest equalizer in human history. Restricted enhancement would be the greatest divider. The mechanisms for universal access are familiar: public investment, healthcare policy, IP law, international access frameworks. The Annals of Medicine and Surgery (August 2025) warns that cognitive enhancers already raise concerns about disparities in access, fairness, and coercion in competitive environments. The case for cognitive enhancement as a public right is analogous to the case for public education in the 19th century. Democratic governance required a literate citizenry, so societies built public schools. The parallel is not abstract. Public education is already the existing cognitive enhancement infrastructure, and it is already failing to provide equal cognitive development. Per-pupil spending in the wealthiest U.S. school districts exceeds spending in the poorest by a factor of three or more. AI tutoring layered onto that baseline amplifies the disparity: a child with a $200/month AI tutor calibrated to their learning pace and available during critical developmental windows is receiving a qualitatively different cognitive environment from a child in an underfunded school with no devices. This is the cognitive enhancement divide in its most immediate, least speculative, and most actionable form. Augmented governance will require a cognitively capable citizenry, and the question is whether societies will build the equivalent infrastructure before the private market makes the divide permanent.
The strongest counterargument draws from technology diffusion history. Cochlear implants began as experimental devices costing tens of thousands of dollars and are now covered by approximately 90 percent of employer health plans as well as Medicare and Medicaid. LASIK was a luxury procedure in the 1990s and is now widely affordable. Smartphones went from executive status symbols to global ubiquity within 15 years. AI tools are already following this curve: open-source models are free, and the cost per token is falling rapidly. The counterargument has real force. The rebuttal is about timing. Previous technology diffusion cycles took decades. During those decades, the early adopters accumulated advantages, but those advantages did not compound biologically or cognitively in ways that made the gap permanent. A generation that grows up with AI tutoring calibrated to their learning pace, pharmacological cognitive support, and eventually genetic optimization will have developed capabilities that a subsequent generation receiving the same tools cannot retroactively acquire. The developmental window for many cognitive capacities is finite. Brain plasticity peaks during critical periods in early development, after which molecular brakes constrain further reorganization, and the resulting neural architecture becomes difficult or impossible to replicate through later intervention (Hensch, PNAS, 2020). Enhancement during those windows produces advantages that late access cannot fully replicate.
Key tension: The policy frameworks that will govern cognitive enhancement are being built now, in IP law, FDA classification, IVF regulation, and AI access policy, before the most dramatic technologies arrive. The AI-based cognitive divide is already measurable. The pharmacological and genetic channels are commercializing ahead of governance. Each year of stratified access compounds into advantages that become harder to reverse, and some of those advantages, once embedded in development or germline, cannot be reversed at all.