The Plutocratic Bias

In the AI Story

The mechanism is familiar from behavioral economics: the anecdotal generalization. Stories of a developer in Lagos who built a successful product with AI tools, an engineer in Trivandrum who expanded her capabilities across domains, a non-technical founder who prototyped an application over a weekend. Each story is true. Each describes genuine expansion of individual capability. The inference from individual cases to distributional claims — the move from 'these individuals gained' to 'the technology is broadly beneficial' — is precisely the reasoning distributional analysis exists to challenge.

The anecdotal generalization is persuasive because it is vivid. A story about a Lagos developer is more emotionally compelling than a Gini coefficient. But the story is systematically selected from the positive tail of the distribution. The developer who succeeded with AI tools is visible, vocal, available for citation. The developer who tried and failed is invisible. The worker displaced by AI is not posting triumphant threads on social media. The resulting selection bias produces a discourse systematically skewed toward the positive end of the distribution.

The bias has direct policy consequences. When the most influential voices are disproportionately the voices of gainers, the policy prescriptions that emerge are tailored to the interests of the gaining population. The dominant AI policy prescriptions — accelerate capability development, invest in AI-complementary education, reduce regulatory barriers, build attentional ecology for AI-augmented workers — are all valuable. They are also prescriptions that primarily benefit populations already positioned to capture gains. They address the upper hump. They do not address the structural dynamics producing the valley: the capital-labor split, competitive compression, the geographic concentration.

The prescriptions distributional analysis identifies as most critical — progressive taxation of AI-derived capital gains, strengthened worker bargaining power, international coordination on digital taxation, mandatory profit-sharing in AI-augmented firms — are conspicuously absent from mainstream AI discourse. They are absent not because they are analytically unsound but because they are politically uncongenial to the populations dominating the conversation. The founder does not advocate for higher capital gains taxes on her equity. The investor does not advocate for profit-sharing that reduces returns. The executive does not advocate for stronger unions in her industry. The absence is structural, not conspiratorial.

Origin

The concept extends a long tradition in political economy recognizing that economic analysis is never independent of the analyst's position in the distribution it describes. Marx's critique of political economy, Gramsci's analysis of hegemonic consensus, more recent work in the sociology of economic knowledge — all converge on the recognition that the positions from which analysis is produced shape the analysis. Milanovic's specific contribution is documenting the correlation empirically and identifying the selection mechanisms — who gets published, whose stories get amplified, which policy prescriptions get taken seriously — through which the bias operates.

Key Ideas

Position shapes perception. The technology looks like progress from the trunk of the distribution because it is progress there. Generalizing the view from the trunk to the valley is the core failure mode of the AI discourse.

Anecdotal generalization. Individual success stories are selected from the positive tail and amplified through media channels dominated by gainers, producing a systematically skewed aggregate picture.

Policy follows discourse. The policy prescriptions that dominate — individual adaptation, educational reform, regulatory minimalism — are tailored to the interests of the populations dominating the discourse. Distributional policies are conspicuously absent.

Not conspiracy, structure. The bias operates through sincere belief, genuine experience, and rational self-interest. Individual actors do not need to coordinate; the structural correlation between position and advocacy produces the pattern.

The corrective is distributional measurement. Independent empirical measurement of the distribution — drawn from outside the communities whose experience dominates the discourse — is the only mechanism that can counter the bias. The elephant curve performed this function for globalization; the AI transition awaits its equivalent.

Debates & Critiques

Defenders of the current discourse argue that the distinction between gainers and losers is overstated, that broad access to AI tools means the technology is genuinely democratizing, and that distributional concerns are pessimism dressed as analysis. Milanovic's response is that access and capture are different variables (see Geography of Value Capture), that the empirical record of previous transitions is consistent, and that waiting for distributional data to arrive retrospectively is a form of advocacy for the current trajectory — the trajectory that favors those already gaining.