The AI Elephant

In the AI Story

The four segments correspond to distinct populations with distinct experiential realities. The excluded two to three billion, concentrated in sub-Saharan Africa and parts of South Asia, experience neither gains nor losses but differential acceleration: their absolute condition is unchanged while the populations above them pull away, compounding a gap that becomes progressively costlier to close. Their flatness is not safety; it is exclusion reproducing itself through institutional infrastructure they cannot access.

The valley is populated by the professional middle class whose implementation skills are being commoditized — paralegals, junior analysts, copywriters, customer service workers, mid-level developers. The mechanism is not mass unemployment but competitive compression: wage premiums erode as AI enables less-skilled workers to produce equivalent output. The absolute skills do not change; the relative scarcity that sustained the premium does. This is the population whose distributional experience will shape the politics of the coming decade.

The upper hump is the population Edo Segal describes most vividly in The Orange Pill — the builders who experience AI as genuine capability multiplication. Their gains are real and substantial. Their existence demonstrates that AI is not uniformly negative for labor. But the size of the hump relative to the valley, and the institutional mechanisms by which hump gains might be shared with valley populations, are the questions the hump's visibility tends to suppress.

The trunk is the concentration at the top: AI capital owners whose wealth has accumulated in ways that compress decades of normal appreciation into quarters. Trillions of dollars in market capitalization created in a few years, venture returns exceeding any previous cycle, the homoploutic elite whose positions are amplified simultaneously on both the capital and labor dimensions. This concentration is the distributional signature of the AI transition and the primary target of any institutional architecture that would moderate it.

Origin

The AI elephant is not an empirical curve drawn from retrospective data. It is a structural hypothesis derived from the characteristics of AI technology — capital-intensive, platform-concentrated, productivity-multiplying in ways that flow disproportionately through capital channels — combined with the institutional context in which AI is being deployed, a context in which the capital-labor split has been shifting toward capital for four decades and the institutional architecture for redirecting productivity gains is at its post-war nadir.

The sketch is testable. As household survey data and administrative records from the 2023–2028 period become available, the shape can be measured rather than hypothesized. But the measurement will arrive too late to inform the institutional choices that determine the shape. The distributional analysis must be prospective or it is not useful.

Key Ideas

Four segments, four populations. Excluded tail, compressed valley, augmented hump, concentrated trunk. Each corresponds to a distinct experiential reality that aggregate AI productivity statistics erase.

Concentration at the trunk is more extreme. AI's gains are flowing to a smaller group of firms and investors, overwhelmingly geographically concentrated, than globalization's gains. The Gini of AI wealth is higher than the Gini of globalization wealth.

The valley is deeper and faster-forming. Knowledge-task automation compresses wage premiums in months, not decades. The speed of competitive compression is qualitatively different from the speed of offshoring.

The timeline compresses institutional response. Two decades of globalization produced inadequate institutional responses. The AI transition's years-not-decades pace narrows the window for institutional construction to a point where the default — radical inequality — operates unchallenged.

The curve is not fixed. The shape is a political question, not a technological one. Institutional choices about taxation, education, labor law, and international coordination determine the depth of the valley and the flatness of the left tail.

Debates & Critiques

The critique most frequently raised is that AI's productivity gains may be exponential rather than linear, invalidating historical analogies. Milanovic's response cuts both ways: if gains are exponential, so are the distributional consequences of institutional failure. A more serious critique is that the hypothesis is currently unfalsifiable given data-infrastructure lags; the analytical response is that waiting for retrospective confirmation concedes the institutional window by default.