The Washington Consensus
The Washington Consensus is John Williamson's 1989 term for the set of market-liberalization policies that the IMF, World Bank, and US Treasury promoted as standard prescriptions for developing economies. The package included fiscal discipline, tax reform, interest rate liberalization, exchange rate competitiveness, trade liberalization, capital-account opening, privatization of state enterprises, deregulation, and secure property rights. The framework treated these policies as universal solutions applicable to any developing economy, regardless of institutional context or historical circumstance. Stiglitz's critique — developed in Globalization and Its Discontents and elaborated across subsequent works — demonstrated that the one-size-fits-all application produced worse outcomes than alternative policies attentive to local conditions, and that the framework's intellectual foundations did not survive empirical examination. The AI governance discourse is now reproducing the same structural pattern: universal frameworks developed by technocrats and industry representatives without substantive input from affected populations, producing predictable distributional consequences that populations absorb without recourse.
The Necessary Infrastructure Problem — Contrarian ^ Opus
There is a parallel reading that begins not with the Washington Consensus as failed ideology but with the material substrate that makes any economic transformation possible. The Washington Consensus policies, however flawed in application, were attempting to solve real problems: hyperinflation that destroyed savings, state enterprises that consumed resources without producing value, closed economies that trapped populations in poverty. The failures Stiglitz documents were often failures of sequencing and capacity, not of direction. Argentina's hyperinflation required fiscal discipline; the question was how to achieve it without destroying the social fabric. Post-Soviet enterprises required restructuring; the question was how to manage the transition without creating oligarchic capture.
The AI governance parallel reveals a deeper problem than Stiglitz's framework suggests. The Washington Consensus at least had mechanisms for course correction — elections could change governments, social movements could resist policies, alternative development models could be tried. AI governance lacks even these limited feedback loops. The compute infrastructure is owned by a handful of firms, the technical expertise concentrated in even fewer hands, the capital requirements so extreme that meaningful alternatives cannot emerge. Where the Washington Consensus could be reformed through political pressure and intellectual critique, AI governance is structurally immune to such corrections. The Lagos developer isn't constrained by bad policy choices that better governance could fix; they're constrained by the physical reality that the infrastructure they depend on is owned elsewhere, controlled elsewhere, and designed for purposes other than their flourishing. The problem isn't that we're repeating the Washington Consensus mistakes — it's that we lack even the limited tools that eventually corrected them.
In the AI Story
The Washington Consensus emerged from the intellectual convergence of monetarist macroeconomics, supply-side fiscal theory, and rational-expectations market efficiency — the intellectual framework that dominated economic policy in the Reagan and Thatcher era and was extended to international institutions through the 1980s. The policies were applied to Latin American debt-crisis countries in the 1980s, to post-Soviet transition economies in the 1990s, and to Asian crisis countries in the late 1990s. In each case, the outcomes were substantially worse than the consensus had predicted: lost decades of growth in Latin America, catastrophic transition in Russia, deepening of the Asian crisis by IMF-imposed austerity.
Stiglitz's critique identified three analytical failures. First, the framework assumed institutional prerequisites — competition, information symmetry, enforcement capacity, democratic accountability — that developing economies often lacked, producing predictable pathologies when market-based solutions were imposed without the institutional infrastructure that makes markets function. Second, the framework treated economic policy as technocratic, excluding democratic input from populations bearing the costs and producing backlash that undermined the political sustainability of the reforms. Third, the framework conflated American institutional arrangements with universal market requirements, imposing American-style institutional choices on societies whose development trajectories required different arrangements.
The AI governance parallel is structural. The contemporary AI governance framework — the EU AI Act, American executive orders, voluntary industry commitments, multilateral principles — treats AI policy as technocratic, designed by a combination of regulators and industry representatives with limited input from affected populations. The framework is applied universally across contexts that vary substantially in institutional capacity, economic structure, and cultural preference. The outcomes reproduce the Washington Consensus pattern: gains concentrated among the powerful, costs absorbed by the less-powerful, and governance structures that preserve rather than correct the concentration. The Lagos developer whose capability expanded but whose capture remained constrained by institutional infrastructure is the AI analog of the Indonesian worker whose economy was liberalized but whose welfare deteriorated.
The framework's decline is instructive for what comes next. By the mid-2000s, even the IMF had begun acknowledging that the Washington Consensus had overreached, and the post-2008 policy discourse has largely abandoned the framework's most confident prescriptions. The replacement has not been a coherent alternative but a more pragmatic, context-sensitive approach that Stiglitz's work helped enable. The AI governance discourse is still in its Washington Consensus phase — confident universal prescriptions, technocratic development, industry capture, predictable distributional consequences. The transition to a more pragmatic alternative will require the accumulation of documented failures that the Washington Consensus critics spent two decades compiling.
Origin
John Williamson coined the term in a 1989 paper for the Institute for International Economics, intending it to describe the convergence of opinion among Washington-based economic institutions rather than to prescribe a universal policy framework. The term escaped Williamson's control, becoming shorthand for aggressive market-liberalization policy and the intellectual framework supporting it. Stiglitz's critique, developed through his World Bank tenure and subsequent academic work, was the most influential internal dissent within the economic establishment.
Key Ideas
Universal prescription without institutional context. The framework assumed market-based solutions would work anywhere, ignoring the institutional prerequisites that markets require.
Technocratic governance without democratic input. Policies were developed by institutions accountable to wealthy-country finance ministries rather than to the populations bearing the costs.
Predictable distributional consequences. Gains concentrated among the already-advantaged, costs flowed to populations lacking political power to resist them.
Intellectual framework failure. The theoretical foundations — rational expectations, market efficiency, institutional universality — did not survive empirical examination.
AI governance parallel. The current AI policy discourse reproduces the same structural pattern, suggesting similar outcomes absent significant reform.
Debates & Critiques
Defenders of the Washington Consensus argued that its failures reflected implementation problems rather than framework flaws, and that the alternative — interventionist policies — had its own record of failure. Stiglitz's response acknowledged the mixed record of alternatives while arguing that the Washington Consensus framework was systematically worse in the cases where its failures were documented, and that the alternative — pragmatic, context-sensitive policy attentive to institutional infrastructure — had been foreclosed by ideological commitments rather than empirical failure.
Appears in the Orange Pill Cycle
Scales of Capture and Correction — Arbitrator ^ Opus
The right frame depends entirely on which scale we examine. At the level of intellectual genealogy, Stiglitz's parallel holds completely (100%): both the Washington Consensus and current AI governance share the fatal combination of universal prescription, technocratic development, and predictable distributional consequences. The structural similarities are not superficial — they reflect the same underlying dynamic where technical expertise becomes the justification for excluding democratic input, producing governance that serves the powerful rather than the affected.
At the level of mechanisms for change, the contrarian view dominates (80%). The Washington Consensus operated through governments that remained nominally accountable to their populations, even when that accountability was attenuated by debt obligations and institutional pressure. AI governance operates through private infrastructure where democratic mechanisms have no purchase. The IMF could be pressured to change its prescriptions; OpenAI and Google face no comparable pressure. The intellectual critique that Stiglitz pioneered worked because policymakers ultimately needed some legitimacy that empirical failure could erode. AI companies need only market dominance.
The synthesis requires recognizing that we're dealing with two different types of capture operating simultaneously. The Washington Consensus represented ideological capture — bad ideas enforced through institutional power but ultimately vulnerable to better ideas backed by evidence. AI governance represents infrastructural capture — control over the means of computation that no amount of intellectual critique can dislodge. The Lagos developer faces both problems: governance frameworks that ignore their context (the Washington Consensus parallel) and infrastructure dependencies that no governance framework can address (the deeper material constraint). Understanding this dual capture suggests that fixing AI governance requires not just better policy frameworks but alternative infrastructures — a challenge that makes Stiglitz's reform of the Washington Consensus look modest by comparison.
Further reading
- Stiglitz, J. (2002). Globalization and Its Discontents.
- Williamson, J. (1990). Latin American Adjustment: How Much Has Happened?
- Chang, H-J. (2002). Kicking Away the Ladder.
- Rodrik, D. (2007). One Economics, Many Recipes.
- Serra, N. & Stiglitz, J. (2008). The Washington Consensus Reconsidered.