I want to make three claims about your future that you probably haven't heard stated this plainly.
The first is mathematical. Most of the traits you were born with - intelligence, conscientiousness, height, bone density, grip strength, resting heart rate - follow a bell curve. They are Gaussian: symmetric, mean-reverting, self-averaging. The wealth you were born into is not. Wealth follows a power law. The top 1% of American households holds more than the bottom 50% combined. The mean is five times the median. These are not the same kind of object. When you multiply a bell curve by a power law to produce a life outcome, the power law dominates. Per standard deviation, inherited wealth predicts lifetime income more strongly than cognitive ability. And unlike cognitive ability, wealth does not regress toward the mean when it passes from parent to child — it compounds. At the extremes, where the power-law tail of wealth extends far beyond what any Gaussian bell curve can reach, the comparison becomes no contest at all.
The second is structural. The historical pathway from intelligence to income - the wire that ran from IQ through credentials through high-paying knowledge work - is being severed by artificial intelligence in real time. Large language models already match median professional performance on the routine tasks that constitute a large fraction of professional billing across legal research, financial analysis, software engineering, and diagnostic reasoning. The IQ premium in the labour market is collapsing. The capital premium is not. The coefficient on inherited wealth in the income equation is rising as the coefficient on cognitive ability falls — and the transition is measured in years, not decades.
The third is a prediction about what kind of society your children will live in — and about a window that is open right now and closes once.
For most of human history the wire connecting intelligence to wealth simply didn’t exist. A peasant born with an exceptional mind had no mechanism to convert that endowment into heritable capital. The channels ran separately, by law and by caste. Then the French Revolution, industrial capitalism, and the credential systems of the nineteenth and twentieth centuries built a bridge: IQ → credentials → income → heritable wealth. Not a perfect bridge — mobility was always slower than the meritocracy mythology suggested — but a real one. For the first time at scale, cognitive ability could escape the class it was born into.
Artificial intelligence is dismantling that bridge. As the IQ→income wire is severed, and as assortative mating intensifies around wealth rather than credentials, the two inheritance systems described in this essay fuse. The Gaussian biological channel and the power-law legal channel stop running in parallel and start running together, toward the same families, across generations. Traits regress. Wealth does not. Three or four generations of that fusion produces something that has a historical name: aristocracy — not by decree, but by mathematics.
But here is what that prediction means right now, today, for anyone who is not already in the top tier of capital ownership. The transition is not complete. The integration of AI into medicine, law, logistics, education, engineering, finance — every domain where human expertise currently commands a premium — is still immature. Domain knowledge combined with AI fluency is still scarce. That scarcity commands a price. The people who build companies at that intersection in the next five to ten years are capturing a transition premium that is, for this brief period, not gated by inherited wealth. It is gated by speed and domain depth — things a person with knowledge and modest starting wealth can actually have.
After the window closes, the legal clock runs alone. Even redistributive mechanisms — UBI, aggressive capital taxation — can compress the distribution from both ends. They cannot create new entrants at the top in a world where labour no longer commands a premium sufficient to accumulate capital from scratch. The window is the next five to ten years. It is open because the transition is still happening. It closes when the transition completes.
You do not need a genetics database to test this prediction. Watch the labour share of GDP against capital returns. Watch whether professional income and inherited wealth grow more correlated in the same households over the next decade. These numbers are published quarterly. The current trajectory already points in one direction.
The rest of this essay builds all three claims carefully — with interactive simulations you can run on your own numbers, a model calibrated to empirical data, and a political argument about which levers are actually available. It takes about an hour. What it offers in return is a precise map of a transition window: where it came from, how long it stays open, and what the world looks like after it closes. I built this framework to understand my children’s futures. I think it will change how you see yours.
There is a person you know - perhaps have always known - for whom things seem to accumulate. The talent that opens the first door. The confidence that follows from early success. The looks that made teachers kinder and strangers more generous. The money that arrived, eventually, as though drawn by some quiet gravity. You watch from nearby and feel something complicated: not quite envy, but a dawning suspicion that the universe is not neutral. That some lives are tilted toward abundance and others toward an endless subtle friction. You wonder if this is luck, or structure, or something so deep it has no name.
Begin with the simplest version of the question. Why does this person seem to have everything? The mathematics has an answer — and it starts with a bell curve.
What It Means to Be Normally Distributed
Take height first. It arises from hundreds of genes, each contributing a tiny nudge upward or downward from some baseline. No single gene determines whether you are tall; it is the accumulation of small effects that matters. And because of a theorem so central to probability theory that it is called the Central Limit Theorem, the sum of many small, independent influences converges to a single, symmetrical, bell-shaped distribution. Not approximately. Exactly, in the limit. The Gaussian curve is not merely a description of height. It is a mathematical inevitability wherever many small additive forces conspire to produce a single outcome.
This is the infinitesimal model, first formalized by R.A. Fisher in 1918, and it is why human height, IQ scores, bone density, grip strength, and dozens of other traits distribute themselves in elegant bells across any large population. Now extend the picture. A person is not a single number but a profile of measurements - intelligence, physical vitality, emotional resilience, social ease, drive. The right mathematical object for all of this at once is the multivariate Gaussian: a joint distribution over many variables, each marginally bell-shaped, related through a covariance matrix Σ.
Here is the first beautiful mathematical fact: the marginals of a multivariate Gaussian are themselves Gaussian. Pull out any single trait, look at it alone, and the bell curve re-emerges. But the covariance matrix - the grid of numbers relating every trait to every other - is where all the interesting structure lives.
One caveat before going further, and it matters. This mathematical framework works well for the traits we have been describing - height, IQ, conscientiousness, physical health. These are approximately Gaussian: sums of many small effects, roughly symmetric, tending toward the mean — though the approximation works better for some traits than others, and breaks down entirely for outcomes that involve wealth. But not all the things we care about in a life live in this space. Social status is a ranking - it is by definition zero-sum, ordinal, and structurally incapable of a normal distribution, because every point gained by one person requires a point lost by another. And wealth follows something closer to a power law: the mean is five times the median, the top 1% holds more than the bottom 50% combined, and the "average" wealth is a statistical ghost that nobody actually has. These are not bell curves. They are different mathematical objects entirely. Mapping them onto the same Gaussian covariance matrix isn't imprecise - it is categorically wrong. What happens when Gaussian biology meets power-law economics turns out to be the more interesting question, and we will return to it after establishing the biological structure first.
II — The Independence ProblemThe Independence Assumption - And Why You Don't Live In It
Imagine that the covariance matrix were diagonal - zeros everywhere off the main axis. This is the world of independence: knowing how tall you are tells you nothing about how quick your mind is, which tells you nothing about the symmetry of your face, which tells you nothing about how hard you work. In this world, probability is ruthless and clean. If being two standard deviations above average in any given trait occurs with roughly 2.3% frequency, then being exceptional in two independent traits simultaneously occurs with probability 0.023 × 0.023 - about one person in 1,900. Three dimensions: one in 83,000. Four: one in 3.6 million. The math does not merely say such people are rare. It says they are essentially impossible.
If traits were truly independent, the person who seems to have everything — the intelligence and the confidence and the ease and the money — should be a statistical impossibility. The mathematics says so plainly. The fact that they exist is itself the proof that the covariance matrix is not diagonal.
But they do exist. This is not merely an anecdote - it is an empirical observation with mathematical content. If independence predicts near-impossibility and the actual world contains observable frequency, then the off-diagonal elements of Σ are not zero. Human traits covary. The visualization below makes this collapse visceral.
III — The CausesFour Reasons the System Tilts - Three Biological, One Legal
The correlation between desirable human traits is not one thing. It is the composite residue of at least four distinct processes, each operating at a different timescale and by a different mechanism. The first three are biological and social. The fourth is legal - and it operates on entirely different mathematics.
The first is assortative mating - the ancient and powerful tendency of humans to pair with those who resemble them. The mate-selection market, for all its apparent chaos, is organized around overall desirability: people near the top of the distribution tend to pair with people near the top, drawing from whatever dimensions are locally valued. An intelligent man of high status pairs with a beautiful and capable woman. Their children inherit genes for intelligence and for attractiveness simultaneously, not because any single gene codes for both, but because the alleles for each traveled together through the pairing. Repeat this across ten generations, and what were initially independent trait distributions begin to develop correlations - not because nature linked them biologically, but because humans linked them socially, over and over, until the links became hereditary.
The second mechanism is pleiotropy - the biological reality that many genes do not specialize in a single function. A body developing under favorable genetic conditions tends to develop well across multiple systems simultaneously. Health is not one thing. It is a general regime of developmental integrity, and when that regime is present, it elevates many traits at once.
The third mechanism is the most philosophically uncomfortable of the three biological-social mechanisms: social compounding. The world treats attractive people as though they are intelligent. It invests more in children who seem promising. Those investments return dividends that are indistinguishable, in outcome, from raw biological ability. The correlation was not in the genes. It was manufactured by a world that could not stop projecting one quality onto the others.
The fourth mechanism sits outside the Gaussian framework entirely, and it is the one the other three cannot account for: legal inheritance. Biological traits transmit to children through meiosis and recombination - a stochastic process subject to regression, to the shuffling of chromosomes, to the fundamental randomness of sexual reproduction. Each generation, extreme values are diluted. The system is noisy and self-correcting. Wealth transmits through property law - through wills, trusts, the step-up in basis at death, the institutional memory of alumni networks and board seats and donor records. This process is not stochastic. It does not regress toward the mean. A trust fund is not subject to independent assortment. It passes whole. It compounds. It is, in the mathematical sense, a completely different inheritance channel running in parallel with the biological one - and actively working against the self-correcting tendency of the genetic system. Two people with identical trait profiles but different starting wealth positions are not located at the same point in any meaningful joint distribution of human outcomes. They are operating under different physics. That difference - the fourth mechanism - is the subject of Section V. This fourth mechanism operates on entirely different mathematics from the first three. It does not belong in the covariance matrix. It belongs in the power law. And understanding why it is categorically different from the biological mechanisms — not just quantitatively larger but structurally distinct — is the hinge on which the rest of this essay turns.
IV — The EvidenceWhat the Studies Show - Once You Remove the Bad Ones
In cognition, the evidence is most robust. The positive manifold - the finding that all measured cognitive abilities correlate positively with one another - has been described as arguably the most replicated result in all of psychology.1 No matter how different two cognitive tests are, scores on them tend to move together. This structural property emerges from factor analyses of thousands of test batteries across dozens of countries and cultures, making it uniquely immune to the usual confounder concerns: it is a mathematical property of the score matrix, not a relationship between measured outcomes. From it emerges g, the general factor of intelligence, which typically accounts for forty to fifty percent of the total variance in any diverse cognitive battery.2
Between domains, the picture becomes more complicated - and more honest methodology matters enormously. The height–IQ correlation (r ≈ 0.15–0.20) has been examined in a large nuclear twin-family design that explicitly models and controls for assortative mating, partitioning the covariance into its genetic and environmental components and separating genuine pleiotropy from the statistical artifact of correlated alleles built up through generations of mate choice.3 Even after this rigorous decomposition, roughly half the correlation survives as shared additive genetic variance - modest, but real. Intelligence predicts income with replicated correlations around 0.35–0.40, a relationship that holds across multiple datasets and partly survives controls for education, age, and socioeconomic background.4
The attractiveness–IQ pairing requires particular care, because it is where the literature has been most misleading. Early studies, including the frequently-cited work of Kanazawa (2011) using large British and American cohorts, did control for social class, body size, and health - and found correlations of r = 0.13–0.38.5 However, a critical flaw undermines the stronger of those estimates: in the British cohort, the same teacher who assessed a child's intelligence also rated their physical attractiveness, making the two measures non-independent and inflating the correlation through evaluator bias. The most methodologically rigorous study to date - Mitchem et al. (2015), using a large twin and sibling sample with independently collected measures of facial attractiveness and IQ, and a design capable of partitioning genetic from environmental covariance - found no phenotypic or genetic correlation between the traits.6 Their meta-analysis further found that reported effect sizes decrease as sample size increases (r = −0.41 between log N and effect size), a signature of publication bias: small studies showing a correlation got published; equally small studies showing nothing did not. The honest current estimate for the attractiveness–IQ correlation is probably near zero or at most weakly positive - and most of what earlier studies captured was likely methodological artifact.
The overall picture is a covariance matrix that is not diagonal - but also less strongly off-diagonal than a casual reading of the literature suggests, and almost certainly less so for attractiveness than the heatmap below implies. The correlations that survive rigorous confounder control and large-sample pre-registration tend to be modest: r = 0.15–0.40 for the better-established pairs, near zero for others. The space of human possibility is genuinely high-dimensional, with desirable traits positively but loosely correlated - when good methodology is applied.
The correlation that matters most for the argument that follows is the one between wealth and everything else. Wealth–IQ (r ≈ 0.35–0.40) is large partly because wealth buys the environments that develop cognitive ability — nutrition, schooling, stability, the absence of chronic stress — and partly because high-IQ parents earn more and pass both the genes and the capital to the same children. This bidirectional causation is the first hint that the biological and legal channels are not independent. They are already coupled. The question Section V asks is what happens when they couple further.
The conditional distribution - what observing someone to be exceptional in one dimension tells you about the others - is perhaps the most practically important consequence of this structure. If ρ = 0, learning that someone is brilliant tells you nothing about whether they are also beautiful or wealthy. If ρ = 0.3, it shifts your expectation modestly but far from deterministically.
Two Inheritance Systems, Running in Parallel
The framework built so far quietly assumes that all human traits live in the same mathematical space. They do not. IQ, height, and conscientiousness are approximately Gaussian - sums of many small effects, bell-curved, self-averaging. But social status is a ranking: zero-sum, ordinal, structurally incapable of being normally distributed because every point gained requires a point lost elsewhere. And wealth follows a power law, or something close to it, where the mean is a fiction and the tail contains a disproportionate share of everything.
Consider what this means concretely. The mean US household net worth in 2022 was approximately $1.06 million. The median was $193,000.9 The gap between them - a factor of more than five - is the signature of a power-law distribution, where a small number of extreme values pull the mean far above the middle. For a Gaussian distribution, mean and median are identical. The fact that they diverge so dramatically for wealth tells you immediately that wealth is not Gaussian. The "average person" in the wealth distribution is a statistical ghost. The figure below shows what the real distribution looks like: not a bell, but an asymmetric curve with a vast low tail and an almost invisible high tail that nevertheless contains an enormous fraction of all wealth.
This is the first rupture in the Gaussian framework. The second is more consequential: the mechanism by which these non-Gaussian goods are transmitted across generations is entirely different from the genetic mechanism we have been discussing. Biological traits transmit to children through meiosis and recombination - a stochastic process that guarantees regression toward the population mean. The child of two people with IQs of 130 will, on average, have an IQ closer to 115. Genes are shuffled. Extreme values are diluted. The biological inheritance system is, in the long run, self-correcting.
Wealth transmits through property law. A trust fund does not do regression to the mean. A house valued at $2 million does not become a house valued at $1 million when it passes to the next generation - it appreciates. Capital earns returns. The legal mechanisms that transfer wealth across generations - inheritance law, gift tax thresholds, estate planning, the step-up in basis at death - are specifically designed to preserve and concentrate, not to randomize and disperse. Jeff Bezos's parents invested $250,000 in Amazon in 1995.11 That was not a trait transmitted through DNA. It was capital transmitted through a bank wire, subject to none of the biological regression that governs cognitive inheritance. Similarly, Harvard's legacy admission rate is roughly six to seven times the standard rate - a non-genetic advantage transmitted not through chromosomes but through alumni directories and donation records.12
The mathematical reason the two systems interact multiplicatively rather than additively is the threshold. Below approximately $20,000 in net worth — where roughly a quarter of American households live — the compounding mechanism of the legal clock simply does not operate. Income is consumed by subsistence. Savings cannot activate. The IQ-adjusted savings premium, which in the model adds roughly four percentage points of savings rate per standard deviation of IQ above the mean, is worth zero when there is nothing left to save. Above the threshold, both clocks run — and the larger one, the legal one, runs independently of whatever traits its owner has. The floor is not merely hard. It is, mathematically, a different country with different rules.
The result is two inheritance systems operating simultaneously, at different speeds, on different mathematics. The biological system is Gaussian and mean-reverting. The legal system is power-law and compounding. And they interact multiplicatively, not additively. A high-IQ child born into poverty is not simply a high-IQ adult with constrained resources — they are a high-IQ adult whose trait advantages must fight against a compounding disadvantage that grows each year through interest on debt, through the opportunity cost of not having capital to deploy, through the friction of institutions that were designed for people who look and sound and were schooled differently. And below the $20k net worth threshold, the IQ advantage cannot even activate the savings channel: income is entirely consumed by subsistence, so the second route from intelligence to intergenerational wealth is blocked before it begins. Conversely, a modest-ability child born into substantial wealth does not merely have a financial head start — they have a multiplier applied to every trait they possess, because capital converts even mediocre traits into above-average outcomes through access, network, and institutional legitimacy.
Y — household income (log-normally distributed; anchored to a $40k individual income median)
zIQ — IQ standardised to mean 0, SD 1. Intelligence is genuinely Gaussian — the sum of thousands of small genetic effects. It enters the equation as a raw z-score. Heritability ≈ 0.60 (a conservative lower bound for adulthood; adult twin studies typically find 0.65–0.80); regresses toward the population mean each generation, though the rate of regression is shallower under assortative mating — which is why the AM dynamics in Section VI matter for the long-run trajectory.
zC — conscientiousness standardised to mean 0, SD 1. Also approximately Gaussian and heritable (h² ≈ 0.45). Predicts income through persistence, reliability, and long-horizon planning — independently of IQ.
zW — starting wealth, rank-transformed to a z-score via the empirical wealth CDF. Wealth is not Gaussian — it follows a power law, with a mean five times the median. It cannot be treated as a raw z-score. Instead, each person’s wealth percentile is mapped through the inverse normal CDF, preserving ordinal position while allowing entry into the linear model. The implication: a one-unit move in zW near the top of the distribution corresponds to vastly more absolute dollars than the same move near the median. The model is linear in rank; the underlying variable is not.
ε — residual noise, N(0, 0.70²). Captures luck, path-dependence, and the enormous individual variation that the three structural variables cannot explain. The model is a skeleton, not a destiny.
η — inheritance fraction (≈ 0.85). The legal channel transmits most wealth intact — what does not reach children goes to taxes and dissipation.
g(Wt) — capital growth multiplier, threshold-gated:
• Wt < $1k — g ≈ 0.85 (debt zone: interest erodes wealth)
• $1k ≤ Wt < $20k — g ≈ 1.0 (subsistence: income consumed month to month; nothing to invest)
• Wt ≥ $20k — g ≈ 1.5 (accumulation: capital appreciates at ≈1% real/yr over a 40-year career)
s(zIQ) — savings rate, IQ-adjusted: max(0, 0.12 + 0.04 · zIQ). Higher intelligence predicts lower discount rates, better portfolio decisions, and debt avoidance (Grinblatt et al. 2011; McArdle et al. 2009; Shamosh & Gray 2008). At average IQ the baseline rate is 12%; each standard deviation above adds roughly 4 percentage points. This term is zero below the $20k threshold — you cannot save what you must spend to survive.
ν, T — the fraction of lifetime savings that reaches the child (≈ 15%) and the career length (40 years).
The simulator below lets you test the arithmetic yourself. Three presets tell the essential story. IQ vs Wealth: high-IQ at median wealth versus median-IQ at 90th-percentile wealth — notice how close the medians are. Dynasty: exceptional traits at low starting wealth versus average traits at top-1% wealth — run it, then set the IQ→income β to 0.10 to model an AI economy and watch the gap become a chasm. The Floor: identical average traits, one group below $20k, one group at comfortable middle class — run it and watch what the threshold does across a generation. Then switch to the dynasty simulation and push the assortative mating slider to 0.9. Watch five generations. That is the prediction this essay makes. The simulator makes it visible.
This is not an argument that traits are irrelevant. The residual variance in the model is large: two people with identical traits and identical starting wealth will have very different outcomes, because the world is noisy and path-dependent and full of choices that the model cannot capture. But it is an argument that the Gaussian framework - the idea that human advantage is primarily a biological story of normally-distributed traits - misses the second inheritance system entirely. And it is an argument that the second system, being legally constructed rather than biologically transmitted, is in principle alterable in ways that the biological one is not.
There is one more turn of the screw. As assortative mating intensifies — and the evidence suggests it is intensifying, particularly among the highly educated, with Greenwood et al. estimating that 2005-level mating patterns applied to 1960 would have produced substantially higher inequality even without any change in the underlying income distribution8,14 — the two inheritance systems are becoming increasingly aligned. The person who is in the upper tail of the trait distribution is now more likely than in previous generations to mate with someone also in the upper tail, and both are increasingly likely to have significant financial assets. This means the biological and legal channels of advantage are converging on the same families. The Gaussian biological world, which is self-correcting, and the power-law financial world, which is not, are being fused together by mating patterns. The result, compounded across generations, is a world where multi-dimensional biological advantage and non-regressing legal wealth accumulate in the same dynasties. Whether this is already visible in the data is an empirical question. The theoretical expectation is that it should be.
Figure 7 makes this concrete at the family level. Place a founding couple, set their traits and starting wealth, then simulate five generations. Watch what regresses and what compounds.
What We Are Left With, and What to Do With It
If desirable human traits are correlated - even modestly, even for the reasons described - then the world is organized in a way that is not fair, in the deep sense of that word. Not merely unfair in the way that random chance is unfair, where outcomes vary but no process systematically advantages one kind of person over another. Structurally unfair: the advantages compound. The person who is more attractive receives better treatment, develops more confidence, performs better on tests, earns more money, and mates with others similarly advantaged - producing children who begin life with correlated advantages they did nothing to earn.
The brilliant, beautiful, wealthy person did not assemble themselves from free choices. They arrived at the intersection of a dozen correlated advantages, most of which were determined before they drew their first breath.
This is not an argument against individual agency. The variance around any cross-trait prediction is enormous. The person who is disadvantaged on multiple correlated dimensions can still - through effort, through circumstance, through the unpredictable grace of a single opportunity - move the needle on outcomes. Regression to the mean is a statistical tendency, not a destiny.
But the humility this mathematics demands of the exceptional is real. Your intelligence is not purely your achievement. Your ease in rooms is not purely a testament to your character. Some of what you are is the accumulated consequence of genetic lotteries and social investments that preceded you by generations. The appropriate response is not guilt - which is as unearned as the advantages - but a certain lightness about one's own excellence, and a certain gravity about the structures that produced it.
There is a darker implication that follows from Section V, and the reckoning is incomplete without it. Increasing assortative mating does not merely widen inequality in the familiar sense - it does not simply stretch the distribution outward at both ends. It changes the shape of the distribution. The top tail locks in: high-trait, high-wealth couples produce children who inherit both channels simultaneously, with below-average regression in traits and near-zero regression in wealth. The bottom tail locks in too: low-trait, low-wealth couples produce children who inherit the debt, the institutional illegibility, and the constrained possibility — with no channel offering escape. The wealth transmission model is specific about why: below $20k net worth, capital erodes rather than compounds, savings cannot activate, and even a high IQ contributes nothing to the savings channel. The two routes by which cognitive ability could have broken the cycle — higher income converting to savings, and better financial behaviour compounding that savings — are both blocked below the floor. The middle disperses. What grows between these concentrating poles is not a gap but a valley - a thinning of the middle class in the precise statistical sense, a distribution drifting toward bimodality. We do not yet have definitive evidence that this has fully emerged. But the theoretical expectation is unambiguous. And the direction of travel in educational sorting, geographic concentration, and occupational stratification is consistent with it. We are not merely unequal. We may be becoming structurally unequal - in a way that self-reinforces rather than self-corrects.
Here is what most conversations about inequality miss entirely, and what the mathematics of this essay makes precise. The biological inheritance channel is not alterable. You cannot tax regression to the mean. You cannot legislate heritability downward. Any policy targeting the Gaussian side of the equation is fighting biology, which does not negotiate. But the legal inheritance channel is made of something completely different. It is made of words. Specific, findable, rewritable words. The step-up in cost basis at death — a provision allowing inherited assets to be revalued at the moment of transfer, erasing the capital gain accumulated over a lifetime — is words in the Internal Revenue Code. The dynasty trust laws in sixteen US states, which allow wealth to compound across centuries without probate, are words in state statutes. The estate tax threshold of $13.6 million, below which the entire legal compounding mechanism operates untouched, is a number chosen by a legislature and changeable by one. None of this is physics. It is policy — written at the behest of people who understood exactly what it did. The political conversation about inequality has for decades concentrated on income: the top marginal rate, the minimum wage, the capital gains rate. The model in this essay says that is exactly backwards. The income tax fights the symptom. The inheritance mechanism is the cause. And unlike the biological mechanism, it can be changed — if enough people understand precisely what it is and where to push.
What Artificial Intelligence Does to the Equation
For most of human history, the wire connecting cognitive ability to economic advancement did not exist. Before the bourgeois revolutions of the late eighteenth century, intelligence was a survival tool within your caste, not an engine of mobility between them. A peasant born with an exceptional mind had no mechanism to convert that endowment into heritable wealth — the channels ran separately, and by law. Wealth transmitted through blood and title. Traits recirculated within class.
The French Revolution, industrial capitalism, and the credential systems of the nineteenth and twentieth centuries built a bridge. Gregory Clark’s surname-tracking research shows that measured mobility rates remained roughly constant across regimes — what changed was the mechanism, and the mechanism matters. For the first time at scale, cognitive ability could be converted into credentials, and credentials into heritable wealth. The two inheritance channels began to couple. The meritocracy was always imperfect — the bell curve and the power law were never fully joined — but the bridge existed, and it changed who the locked-in class was. It is worth remembering what ended Phase 1. Not a change in the distribution of human traits — the peasants of 1789 were not more intelligent than the peasants of 1689. What changed was the legal architecture: the abolition of feudal title, the creation of property rights accessible to non-aristocrats, the establishment of credit markets that allowed cognitive ability to be capitalised. The bridge was not a natural phenomenon. It was words. Which means it can be unwritten with new ones.
Artificial intelligence is dismantling that bridge. The question is not whether this is happening — the productivity substitution data is unambiguous. The question is what it resembles, historically, when the cognitive-work premium collapses and the conversion mechanism shuts down. The answer is uncomfortable: it resembles Phase 1. This is not a distant prediction. The coefficient is already moving. Watch what has happened to entry-level knowledge-work salaries relative to capital returns since 2020. The repricing has begun.
Everything written so far describes a static structure. The Gaussian trait distributions, the covariance matrix, the two inheritance channels, the multiplicative model: these are descriptions of how the world currently converts starting positions into outcomes. But the model has coefficients. And the coefficients are about to change - faster, and more asymmetrically, than at any point in the last century.
The income model from Section V reads: log(income) = 0.35·IQ + 0.25·conscientiousness + 0.45·wealth + ε. That IQ coefficient of 0.35 is a historical artefact. It reflects a century of industrialised economies that were willing to pay a premium for cognitive work precisely because cognitive work was scarce. The thing that made a high-IQ person economically exceptional was simple: they could do cognitive tasks faster, better, and more reliably than others. That scarcity is ending. Not at the margins, over decades. Rapidly, at the core, right now.
Large language models already match or exceed median professional performance on the routine, high-volume tasks that constitute a significant fraction of professional billing hours across legal research, financial analysis, diagnostic triage, software engineering, and content production. This is not a prediction about the 2030s. It is a description of 2025. The junior lawyer billing for research that a model now does in seconds is not being made more productive. They are being made unnecessary. The management consultant producing slide decks at 2am, the analyst building Excel models from scratch, the radiologist triaging routine scans — these are not jobs being augmented. They are jobs being substituted. What remains of those roles is the part that requires judgment, client relationships, and embodied accountability — a smaller and less compensated fraction of what those jobs used to contain. The distinction matters enormously for what happens to B_IQ in the next decade. Andrej Karpathy’s AI job-exposure visualiser — mapping 342 occupations across twelve major categories — puts software engineers, data analysts, paralegals, and translators all at 8–9 out of 10 on the exposure scale. The roles anchoring the low end are roofers, plumbers, and electricians — defined precisely by their physical irreducibility. That is not a coincidence. It is the shape of Wave 1.
If cognitive work can be purchased at near-zero marginal cost from a model running on a data centre, the premium for human cognitive ability collapses toward the premium for anything else that can be purchased cheaply. The coefficient B_IQ does not fall to zero - human judgment, embodied presence, and genuine creativity retain value - but it falls substantially. A plausible trajectory over the next fifteen years puts it somewhere between 0.10 and 0.20. The Gaussian bell curve of IQ does not move. The trait is still there, still real, still heritable, still normally distributed across the population. What changes is how much the economy rewards it.
The coming disruption does not flatten the bell curve. It severs the wire connecting the bell curve to the income distribution.
But the model contains one asymmetry that points in the opposite direction, and it is time-limited enough that naming it plainly is urgent. The transition from a world where cognitive work commands a premium to a world where AI does cognitive work cheaply is not instantaneous. It is a decade-long repricing. And repricing events create arbitrage windows.
The arbitrage here is specific. AI tools now exist that can dramatically amplify the output of a domain expert — a nurse, a structural engineer, a logistics manager, a teacher, a lawyer, a radiologist — who understands both their domain deeply and what the tools can and cannot reliably do. That combination is currently scarce. It will not remain scarce. As the integration matures, as workflows standardise, as vendors commoditise the implementation, the premium for being early will compress toward zero. But the window is open now, probably for five to seven more years in most domains, and the economics during that window are extraordinary. The people who build companies at that intersection — who take a decade of domain knowledge and apply it to AI integration before the market has priced it — are capturing a transition premium that is, for this brief period, not gated by inherited wealth. It is gated by speed, domain depth, and the willingness to act before the outcome is obvious. Those are things a person with cognitive ability and modest starting wealth can actually have.
The model says the window closes. It also says the window is open now. That distinction is the most important thing in this essay for anyone who is not already in the top tier of capital ownership. The bridge is being pulled up. You have roughly five to ten years to cross it.
What AI cannot yet do is physical. It cannot fix your boiler, navigate a building site, insert a catheter, or sit with a dying person. A plumber cannot be replaced by a model. An electrician, a nurse, a care worker - these roles require embodied presence that robotic systems cannot yet reliably replicate. In the personal and care domains — the boiler, the building site, the bedside — general-purpose robotic dexterity lags further behind. That lag creates a transitional window that looks superficially like equalisation. The paralegal and the radiologist fall toward the income of the skilled tradesperson. The Gini coefficient within the bottom ninety percent compresses. This gets described as AI democratising opportunity. It is more accurately described as AI levelling down the professional-managerial class while the top 0.1% diverges faster than ever. The equalisation is real but it is the wrong kind. Economists have a name for this pattern: the K-shaped economy — a distribution that splits rather than shifts, upper arm rising, lower arm falling, tracing the two arms of the letter K.21
There is, however, an important correction to the framing of physical work as a shelter from the first wave. It already understates what is happening. Automation of physical labour is not approaching — it has been under way for a decade, operating at a scale that would unsettle anyone who had not visited a modern fulfilment centre or factory floor recently. The International Federation of Robotics reports that global manufacturing robot density doubled in seven years, from 74 units per ten thousand employees in 2016 to 162 in 2023, with 542,000 new industrial robots installed globally in 2024 alone and a worldwide operational stock now exceeding 4.66 million.22 Amazon runs over one million robots across its fulfilment centres. These numbers are not projections. They describe the current state of the spaces that package your deliveries and manufacture your goods. Most people have not registered this because automated spaces are not yet part of their visible daily lives. When robotic systems cross the threshold into the domains that are — the care home, the building site, the delivery van, the high street — the adjustment will not feel gradual. The infrastructure will have been built long before the visibility arrives.
The more important point is economic rather than temporal. Physical jobs — construction, care work, manufacturing, logistics — have never been the low-capital path to heritable wealth. They were the survival floor: income sufficient to live on, but structurally insufficient to generate the capital surplus that clears the $20k threshold and enters the compounding tier. The knowledge-work premium mattered precisely because it was the route where cognitive ability could generate income large enough to begin that compounding process. A care worker’s income, however indispensable the role, has not historically done this. When AI removes the knowledge-work route and robotics eventually removes the physical-work income floor, the remaining path to capital accumulation for people without inherited wealth is not narrowed. It is closed. The two waves do not arrive simultaneously — the AI wave is cresting now, the robotics wave is building — but they are running on the same current. The combined timeline extends the consolidation window from a single decade to perhaps two or three.
The investment community has priced this in. In the first seven months of 2025 alone, robotics startups raised over $6 billion — more than the entirety of 2024 — with capital from Nvidia, Amazon, Sequoia, and Andreessen Horowitz.23 Elon Musk has described the humanoid robot as “the biggest product in history” and is retooling Tesla’s Fremont factory to produce Optimus units at scale, targeting public sales by end-2027 at $20–30,000 a unit. The economic logic is not obscure: a general-purpose robot that performs the full range of tasks currently requiring a human worker, produced at consumer-electronics scale, captures a labour share of global GDP that currently runs to tens of trillions of dollars annually — not for the worker, but for the company that builds the robot. This is Phase 1 restated in silicon and servos. The mechanism of value extraction is updated; the beneficiaries are the same class as always — those who own the means of production rather than those who constitute it.
The lawyer who billed at $400 an hour for work a model now does in seconds does not become wealthier. The analyst whose edge was processing information faster than competitors loses that edge. The junior doctor whose role was pattern-matching symptoms to diagnoses finds the pattern-matching outsourced. Within the knowledge-economy tier - the credential-holding, university-educated, white-collar professional class - the AI transition is a flattening event. The income distribution within that tier compresses. In the Gini-coefficient sense, measured within the bottom ninety percent, this looks like equalisation. It is not experienced as equalisation. It is experienced as loss.
This is not equalisation in the sense that matters most. The transitional compression within the bottom ninety percent does not create new entrants at the top. It reshuffles the middle while the top diverges. And here is the specific mechanism: in a world without a labour premium sufficient to generate capital accumulation from scratch, redistribution can compress the distribution from both ends — UBI can floor the bottom, aggressive capital taxation can slow the compounding at the top — but neither mechanism creates new dynastic wealth for people starting without it. The only route to the compounding tier is to accumulate capital before the transition completes. That is what the window is for.
Because while B_IQ falls, B_WEALTH rises. Capital owns the AI. The productivity gains from AI do not primarily accrue to the workers whose labour it replaces — they disproportionately accrue to the shareholders of the companies that own the models and the infrastructure. Some gains flow to consumers through lower prices and to workers in adjacent sectors through productivity spillovers; those effects are real but historically modest relative to the capital share. This is not a prediction; it is the observed pattern of every prior capital-for-labour substitution at scale. Piketty's core argument - that when the return on capital r exceeds economic growth g, wealth concentrates - becomes r >> g when AI-driven productivity gains compound into capital returns at the rate currently observed.16 The legal inheritance channel does not merely stay non-regressive. It accelerates. B_WEALTH, already the dominant coefficient in the current model at 0.45, plausibly rises toward 0.65 or higher in a world where AI has commoditised knowledge work. The multiplication still holds — but now one factor dominates almost entirely.
Run this through the simulator in Section V. Set B_IQ to 0.10. Set B_WEALTH to 0.65. Load the Dynasty preset. The gap between exceptional traits at low wealth and average traits at high wealth, which was already uncomfortable in the current model, becomes dramatic. Then load the Floor preset. With B_IQ at 0.10, the high-IQ family below the $20k threshold now loses on both axes simultaneously: the income premium from intelligence has collapsed, and — since income barely covers subsistence — the savings channel cannot activate. The threshold trap closes from both sides. The high-IQ child of a low-wealth family does not merely face a headwind. In an AI economy, the very thing that made their starting position promising has been most comprehensively devalued.
There is, however, one more turn - the one that connects AI back to the assortative mating argument and closes the loop in the darkest possible way. Assortative mating in the current era sorts primarily by educational credential: people meet at universities, in graduate programmes, in professional networks, and pair with those who share their institutional location. The credential was an imperfect but functional proxy for the biological lottery - a rough signal that someone won the IQ draw and did something with it. If AI makes credentials economically meaningless - if a Harvard law degree no longer predicts income when AI does legal research for $20 a month - then the primary sorting mechanism of the current mating market loses its signal value.
The empirical baseline here matters. The primary sorting mechanism today is credential — educational attainment, professional network, institutional tier. But Fagereng, Mogstad and Rønning find that individuals already assort significantly on their own pre-marriage wealth (rank correlation ≈ 0.45–0.50), independently of education, suggesting that wealth has always been part of the sorting signal, partially concealed behind the credential that proxied for it.19 The feedback from assortative mating to inequality is not theoretical: Erola and Kilpi-Jakonen document that changes in partnering patterns directly mediate income inequality growth, a finding that holds even in Finland, where redistribution should suppress it most.20 The K-shaped divergence — upper tail separating, middle compressing — is already visible in the data. Assortative mating is the mechanism that amplifies it across generations. The question the two-futures framing poses is not whether this is happening, but whether removing the credential signal will arrest it or accelerate it.
Two futures follow from this, and they are very different. In the first, assortative mating weakens: with credentials gone as a sorting signal, people pair more randomly across the trait distribution, the two inheritance channels partially decouple, and the dynasty problem stabilises. In the second, assortative mating intensifies - but now around wealth directly rather than around its credential proxy. In a world where wealth is the dominant predictor of income, wealth-based mate selection becomes more rational than it has ever been. The person with capital looks for a partner with capital. The sorting becomes cruder and more explicit. And if that is the direction of travel - if AI removes the cognitive-trait sorting signal and replaces it with wealth sorting - then the fusion of the biological and legal channels does not weaken. It intensifies. The Gaussian world becomes less economically relevant precisely as the power-law world becomes more dominant. The dynasty problem, which the current model makes visible, becomes the dynasty certainty. The available evidence points toward the second future. Wealth was always in the sorting mechanism — the credential was its legible proxy. AI is removing the proxy. What remains is the thing itself: a new class of capital-holders intermarrying with capital-holders, while the rest are left to assort among themselves. The K-shaped economy describes a bifurcation of income trajectories. Wealth-based assortative mating, compounded across generations, is the mechanism that makes it heritable.
This prediction is specific enough to be falsifiable and observable enough to test without a genetics database. Within the next decade: the income share captured by capital will rise to levels not seen since before the New Deal. The correlation between professional income and inherited wealth within the same households will increase measurably. Assortative mating will shift from sorting on educational credentials toward sorting on net worth directly, as the credential screen loses its signal value. Watch the Bureau of Labor Statistics wage data against S&P 500 earnings-per-share growth. Watch the labour share of GDP. These numbers are published quarterly. The model says they will diverge. The current trajectory suggests they already are. And if they diverge as predicted — if the wire severs on the timeline the model implies — then the window described above is not a metaphor. It is an arithmetic deadline. And if the robotics trajectory holds — if the second wave lands on the timeline the investment data implies — then the deadline is not merely for crossing the knowledge-work bridge before it closes. It is for doing so before the physical-work floor is removed as well. The first wave compresses income from the top of the distribution downward. The second removes it from the bottom up. The two waves, completing across two to three decades, leave capital ownership as the only durable source of income that does not erode. That is not a metaphor either. It is what the model predicts when you set B_IQ to its AI-economy value and run the dynasty simulator forward five generations without a labour-income floor beneath the starting position.
It is worth noting what ended Phase 1 the first time. Not biology. Not the gradual natural regression of aristocratic traits toward the mean — though that happened too. What ended it was law. Specifically: enough people understanding precisely which words in which documents were maintaining the structure, and rewriting them. The French Revolution was not a mood. It was a legal event. The question Phase 3 forces is whether the political will exists to write new words before the concentration becomes self-reinforcing enough to capture the legislative process itself — which is, historically, what sufficiently concentrated wealth eventually does. The window for that is also closing. These two windows — the economic one and the political one — are running on roughly the same clock.
VIII — The LetterWhat You Tell Your Children
You tell them this: the world is not random, but it runs on not one lottery but two - with different mathematics, different timescales, and different moral implications. Most people conflate them. The conflation is not innocent: it allows those who won both to treat their position as purely earned, and it prevents those who lost one from understanding which part of their situation is alterable. Understanding the difference is one of the more useful things a person can carry through a life. So let's be precise.
The first lottery is biological. It is the one the essay has been largely about: the joint distribution of traits, the modest but real correlations, the Gaussian bells. This lottery transmits through chromosomes. It is stochastic, which means noisy and imprecise. It regresses toward the mean, which means that exceptional values tend to dilute across generations - the children of two people with IQs of 135 will, on average, have IQs closer to 115. Genes are shuffled at each conception. Extreme values are eroded. The biological inheritance system is, over the long run, self-correcting. It tends, however slowly and incompletely, back toward the centre.
The second lottery is legal. It transmits through property law, through wills and trusts and the step-up in cost basis at death, through alumni networks and donor records and the quiet architecture of institutions that recognise their own. This lottery does not regress toward the mean. A trust fund does not mean-revert. A house does not become worth less when it is inherited - it appreciates. Capital earns returns. Jeff Bezos's parents wired $250,000 into the nascent Amazon operating account in 1995.11 That transmission crossed no chromosome. It was subject to no regression. It moved through a bank, subject only to the laws governing private transfers of capital - laws that were written by humans and can be rewritten by them. When Harvard considers legacy applicants at six to seven times the standard admission rate,12 it is not transmitting a trait. It is transmitting an institutional relationship - something that lives in a database, not a genome.
You tell them: if you were fortunate in both lotteries, you received two gifts from two entirely different systems. The biological gift came with regression built in - your children will be less exceptional than you, statistically, and their children less so again. This is how the Gaussian works. The legal gift came with compounding built in - it will grow unless something actively intervenes. These are not equivalent gifts. The first is self-limiting. The second is not. The humility appropriate to your IQ is about luck. The humility appropriate to your inheritance is about power. These require different responses.
You also tell them that these two systems do not merely coexist - they interact. The model in Section V makes this concrete, and the numbers are worth sitting with. A person at the 85th percentile of IQ with median starting wealth can expect, across a working lifetime, a household income somewhere around $130,000. A person at the 50th percentile of IQ - genuinely average, not disadvantaged - but with starting wealth at the 90th percentile can expect somewhere around $160,000. The more modest mind with the larger inheritance does better, in expectation, than the sharper mind starting from the middle. Push the wealth to the 99th percentile and the gap widens further. This is not an argument that intelligence is irrelevant. It is an argument that the income equation is multiplicative, not additive - that wealth acts as a multiplier on whatever traits a person brings, not merely as a bonus added on top. The same unit of conscientiousness produces more income when it operates on a base of capital, credentials, and network than when it operates in the absence of all three. The floor matters. The floor is not Gaussian.
Fortune is not a possession. It is a velocity - and its direction is set partly by which lottery you won, and partly by which clock is doing most of the work.
And so you tell them, if they are among the fortunate: something is owed. Not guilt - guilt is useless, it performs discomfort without producing change. Not conspicuous humility, which is its own kind of status game. What is owed is seriousness - the specific, unglamorous, long-horizon seriousness of someone who has been handed resources and actually knows it. Find the intersection of what you are capable of and what the world needs, and work there with everything you have. Not as a side project. As a commitment. The biological lottery gave you a head start that will dilute with time. The legal lottery gave you a head start that will compound. Compound interest on inherited advantage demands compound interest on purpose. That is the exchange.
And when they feel they have been dealt a short hand - and they will feel this, because everyone does - you give them two correctives, not one. The first is statistical. The second is historical. Both are more useful than sympathy.
The first is the horizontal corrective: they are almost certainly not as disadvantaged as they feel, relative to their contemporaries, once they account for all the dimensions they cannot easily see. The short hand rarely contains every card. There are forms of excellence - in reliability, in depth of feeling, in the willingness to remain present when the situation demands nothing and offers nothing - that do not load heavily on any general factor, that are not predicted by IQ, and that the world rewards more than it admits. The dimensionality of a human life exceeds what any correlation matrix can contain.
The second corrective runs through time. There have been, by reasonable estimate, around a hundred billion human beings who lived before anyone now alive. Almost all of them toiled against cold, hunger, disease, and the casual violence of political systems that treated most people as instruments rather than ends. A peasant in fourteenth-century France had no access to antibiotics, to education, to the compound interest on centuries of accumulated knowledge that any literate person in the modern world inherits for free. A woman in almost any century before this one had her legal personhood extinguished upon marriage. A child in an eighteenth-century textile mill worked fourteen hours a day in the dark. The short stick of today - in most of the modern world, for most people - is longer than the long stick of most of human history. And critically: part of what made the present possible was that some of those hundred billion people pushed, slowly and painfully and often without reward, against the legal structures that governed their time. The short-hand position is not merely a starting point. It is also, for those who can bear it, a vocation.
Even a modest position in today's distribution places you in the outermost tail of the distribution across all of human time. That is not a reason to be satisfied with injustice. It is a reason to be, underneath everything else, quietly astonished - and to do something worthy of the astonishment.
This is not an argument for complacency. It is an argument for a specific, grounded kind of resilience: the kind that draws its strength not from comparison with neighbours but from perspective across centuries, and not from fatalism about trait distributions but from clarity about which of the two systems is alterable. The biological lottery is what it is. The legal architecture that governs the second lottery was built by humans and can be rebuilt. That distinction is one of the most important things the mathematics of this essay points toward, and it is conspicuously absent from most public conversations about inequality, which tend to collapse the two systems into a single undifferentiated concept of "privilege" - losing, in that collapse, precisely the information that would tell you where to push.
There is one more thing to tell them, and it is the most uncomfortable. The person they choose to build a life with is not only an intimate decision. It is a mathematical one. Under strong assortative mating - the tendency of people to pair with those similar to themselves in education, intelligence, and socioeconomic position - the two inheritance systems fuse. The biological channel and the legal channel begin to flow in the same direction, toward the same children, in the same families. The Gaussian biological world, which is self-correcting, and the power-law legal world, which is not, are locked together. Their children inherit both the traits and the compounding wealth. The children of the opposing pairing - high trait, low wealth meeting low trait, low wealth - inherit the regression and the debt simultaneously. This is the mechanism by which dynasties form in a nominally meritocratic society: not through overt hereditary privilege, but through the quiet mathematics of who pairs with whom, repeated across a few generations, amplified by the non-regression of the legal channel. They should know this. Not as a constraint on how they love, but as a clear-eyed understanding of what their choices do, downstream, to the probability distributions of people who do not yet exist.
There is a third thing to tell them, and it is the most time-sensitive. The conversion mechanism between the first lottery and the second — the bridge that made the credential era possible — is being dismantled in real time. This is not a background condition they will grow up into. It is a transition happening now, during the period when they are forming their understanding of how the world works and what is possible within it. The decisions made in the next five to ten years — about what to build, about which skills to develop, about how to position at the intersection of domain expertise and AI capability — will determine whether they arrive at the other side of the transition with capital that can compound or without it. This is not alarmism. It is arithmetic. The window is in the model.
And tell them — this is the part none of the generation before theirs was told clearly — that the economic value of cognitive ability is being restructured in real time, and that the restructuring creates a specific, time-limited opportunity before it closes permanently. The world they are entering will reward human intelligence less reliably than the world their parents entered. But right now, during the transition, there is a premium for people who understand both a domain and the tools that are remaking it. That premium is not gated by inheritance. It is gated by speed and seriousness. The inheritance tax you advocate for, the step-up in basis you push to repeal, the dynasty trust laws you vote to prohibit — these are the political levers that determine what the world looks like after the window closes. But while the window is open, the most consequential thing is to be on the right side of it. Both matter. The window and the law. The personal and the political. One is a five-year clock. The other is a generational one. Both are running.
You tell them, finally, that the right response to all of this is not bitterness - which is as useless as envy - and not despair, which mistakes a description of structure for a sentence of fate. The structures are real. The mathematics is real. And the person struggling across multiple dimensions simultaneously is likely experiencing not just correlated trait disadvantages but the active compounding of a legal starting position that works against them at its own rate, independently of anything they do. Debt has an interest rate. Opportunity cost is real. Institutional illegibility - the condition of not looking or sounding like the people who control access - is a real tax levied on real effort, every day, with no exemptions for talent. This is not a personal failing. It is a mathematical structure. Compassion, in a world with this structure, is not sentiment. It is the rational response to a correct model.
And then - this last part matters - you tell them that no matrix captures everything. That the two inheritance systems are real and their multiplication is real and the intensification under assortative mating is real and the policy implications are real. And that the things which make a life worth living do not appear in any of it. The quality of attention you bring to another person. The texture of loyalty over time. The capacity to hold difficulty without breaking, to find meaning in ordinary days, to be genuinely glad that other people exist. These do not covary with height. They are not predicted by IQ. They are not in the Gaussian, they are not in the power law, they load on nothing in the factor analysis, they are not heritable in any meaningful sense, and they are not transmitted by either chromosome or property law.
They are, nonetheless, where most of life is actually lived. And they are available - more or less equally, more or less freely - to everyone who chooses to cultivate them.
The bell curves are real. The correlations are real. The multiplication is real. The two clocks are real. The fusion under assortative mating is real and it is happening now. The AI disruption to the IQ coefficient is real and it is already underway.
None of this determines what kind of person you decide to be. But it does determine what the honest conversation looks like — with yourself, and eventually with your children. You now have a model. It has coefficients, a threshold, a window, and a closing date. You can roughly estimate your own position on each variable. You can run the simulations with your own starting point. You can look at the IQ→income β slider set to 0.10 and ask yourself what your situation looks like in that world — and whether you have five years to build something before it arrives.
I would not wait.
The hardest thing the mathematics in this essay teaches is not that the world is structured — that much is obvious — but that it is structured by two systems with different logics and different susceptibilities to change, and that a third force is now altering the weights on a specific and observable timeline. The biological lottery is self-correcting. The legal lottery is not. Artificial intelligence is severing the wire that connected the first to income while accelerating the second through capital returns. The world your children will inherit is one where the bell curves are intact but their economic significance is diminished, and where the power law is ascendant and accelerating. This is not a speculation. It is a prediction with a quarterly data feed. Watch the numbers.
The work of a just society, in that world, is not to flatten the Gaussian — which biology will partially do on its own — but to rewrite the legal architecture governing capital transmission before the AI transition makes the concentration self-reinforcing enough to write the laws itself. The targets are specific. They are findable. They are words in documents — the step-up in basis, the dynasty trust, the estate threshold, the carried interest treatment. These were written by humans who understood exactly what they were doing. They can be rewritten by humans who understand it better. If the current stories don’t work, you can write new ones. That is not optimism. That is just an accurate reading of what law is.
And the work of an individual life, right now, in this specific window, is to understand the transition well enough to act before it closes. The bridge is being pulled up. The legal clock is accelerating. The compounding tier is locking in. But the window is open. The integration of AI into every domain of human expertise is still immature. Domain knowledge combined with AI fluency is still scarce and still commands a price. The people who build at that intersection in the next five to ten years are the last cohort for whom cognitive ability alone — without dynastic capital behind it — can generate the starting position from which the legal clock can compound. After that window closes, the floor is the floor and the ceiling is inherited.
You have a model. You have a window. You have the levers.
Do something with them that is worth the knowing.
- Spearman, C. (1904). "General Intelligence," Objectively Determined and Measured. American Journal of Psychology. See also: Wikipedia: g factor (psychometrics).
- Carroll, J.B. (1993). Human Cognitive Abilities: A Survey of Factor-Analytic Studies. Cambridge University Press. Wikipedia summary.
- Keller, M.C. et al. (2013). The Genetic Correlation Between Height and IQ: Shared Genes or Assortative Mating? PLOS Genetics 9(4). Full text.
- Gottfredson, L.S. (1997). Why g Matters: The Complexity of Everyday Life. Intelligence 24, 79–132. See also: Wikipedia: Cognitive epidemiology.
- Kanazawa, S. (2011). Intelligence and physical attractiveness. Intelligence 39(1), 7–14. PDF. Note: the large UK effect (r = 0.38) is likely inflated by non-independent rating; the same teacher assessed both traits. Kanazawa is a highly controversial researcher; his broader body of work has been widely criticised for methodological and ethical reasons. See Wikipedia.
- Mitchem, D.G. et al. (2015). No Relationship Between Intelligence and Facial Attractiveness in a Large, Genetically Informative Sample. Evolution and Human Behavior 36(3), 240–247. Full text (PMC).
- Border, R. et al. (2022). Cross-trait assortative mating is widespread and inflates genetic correlation estimates. Science 378(6621). Full text.
- Assortative mating meta-analysis across 22 traits: Yengo et al. (2022), bioRxiv preprint. Highest AM found for educational attainment, IQ, and political values.
- Greenwood, J. et al. (2014). Marry Your Like: Assortative Mating and Income Inequality. American Economic Review 104(5). Wikipedia: Assortative mating. Finds that if 2005 mating patterns had prevailed in 1960, the Gini coefficient would have been substantially lower. AM has intensified dramatically over the intervening decades, driven by women entering the workforce and education becoming the primary sorting mechanism.
- Chetty, R. et al. (2014). Where is the Land of Opportunity? The Geography of Intergenerational Mobility in the United States. Quarterly Journal of Economics 129(4). Opportunity Insights. The intergenerational wealth elasticity used to calibrate the wealth coefficient in the model (≈ 0.45) derives from this and related work.
- Acemoglu, D. & Restrepo, P. (2018). Artificial Intelligence, Automation and Work. NBER Working Paper 24196. NBER. Empirical analysis of automation's effect on employment and wages - finds that robots reduce both employment and wages of workers in affected commuting zones. AI extends this dynamic to knowledge work.
- Autor, D. (2024). Applying AI: The Next Frontier in Labor Market Impacts. MIT Economics. Surveys the emerging evidence on AI's occupational impacts, noting that unlike previous automation which hollowed out middle-skill routine work, AI disproportionately targets high-skill cognitive tasks - lawyers, radiologists, coders, analysts - while leaving physical and care-based work relatively untouched in the near term.
- Piketty, T. (2014). Capital in the Twenty-First Century. Belknap/Harvard. The core argument - that the return on capital (r) systematically exceeds the growth rate of the economy (g) when conditions allow - is the macroeconomic context for why the legal inheritance channel does not self-correct: capital compounds faster than income grows, concentrating wealth upward absent intervention. See also: Wikipedia summary.
- Fagereng, A., Mogstad, M. & Rønning, M. (2021). Marriage, Assortative Mating and Wealth Inequality. NBER Working Paper 29903. Using Norwegian administrative data, finds pre-marriage wealth rank correlations of 0.45–0.50 between spouses, independently of education — suggesting wealth has always been a sorting signal, partially concealed by the credential that proxied for it.
- Erola, J. & Kilpi-Jakonen, E. (2021). The role of partnering and assortative mating for income inequality: The case of Finland, 1991–2014. Acta Sociologica 64(3). Documents that changes in partnering patterns directly mediate income inequality growth even in a high-redistribution Nordic context, confirming the AM→inequality feedback loop is structural rather than institutional.
- Cajner, T. et al. (2021). The K-Shaped Recovery. US Bureau of Labor Statistics Research Paper. Documents the post-COVID bifurcation pattern in which upper-income households recovered and accelerated while lower-income households stagnated or fell — now considered a structural feature of AI-era labour markets rather than a cyclical anomaly.
- International Federation of Robotics (2024). World Robotics 2024: Global Robot Demand in Factories Doubles Over 10 Years. IFR Press Release. Reports 542,000 new industrial robots installed globally in 2024, worldwide operational stock exceeding 4.66 million units, and robot density in manufacturing doubling from 74 to 162 units per 10,000 employees between 2016 and 2023.
- Robotics and Automation News (2025). Venture capital and private equity in robotics: Where is the smart money going? Reports robotics startups raising over $6 billion in the first seven months of 2025, exceeding the total for all of 2024, with concentration in AI-powered physical automation backed by Nvidia, Amazon, Sequoia, and Andreessen Horowitz.
Comments