A little this and that but mostly AI.

The Chicago Boys and South America is a dark chapter and it’s largely been scrubbed from polite American discussion of economics. Milton Friedman and his students at the University of Chicago trained a cohort of Chilean economists starting in the 1950s through a State Department-funded program, and when Pinochet took power in 1973 after the CIA-supported coup against Allende, those students — the Chicago Boys — became the economic policymakers.

Chile became the world’s first full-scale neoliberal experiment: privatization of pensions, health care, education, public utilities, deregulation of labor markets, opening to foreign capital, smashing of unions. The results depend on what you measure and what period you look at, but the human cost during the implementation phase included a brutal dictatorship that killed around 3,000 people, tortured tens of thousands, and disappeared political opponents while the economic restructuring was being imposed. Friedman personally visited Chile in 1975, met with Pinochet, and while he later claimed he only gave economic advice, the intellectual alliance between free-market economics and authoritarian enforcement of that economics is one of the real scandals of 20th-century intellectual history.

The pattern repeated across the region: Argentina under the junta, Brazil under its military dictatorship, Uruguay, Bolivia under Banzer, and later through IMF structural adjustment programs across the continent. John Perkins’s Confessions of an Economic Hit Man — laid out the mechanism from inside: load countries with debt they can’t repay, then use the debt as leverage to extract resources, privatize state assets to Western corporations, and install compliant governments. Perkins has been criticized by economists for oversimplifying and by historians for some factual slippage, but the core story — that development assistance, loans, and economic advice were often tools of soft empire — is broadly accurate and is now acknowledged even by mainstream institutions.

The IMF’s own internal reviews of its 1980s-90s structural adjustment programs in Latin America and Africa concluded they’d caused enormous unnecessary suffering, though of course nobody involved was held responsible. Naomi Klein’s The Shock Doctrine extended the argument — that crisis is the moment when unpopular economic restructuring gets imposed — and while she overstates in places, the pattern she identifies is real. Iraq in 2003-04 under Paul Bremer was a laboratory for it; Russia’s 1990s privatization under Jeffrey Sachs and Harvard advisors was another; Greece after 2010 got a gentler version of the same treatment.

The thing worth noting about this history is that the current backlash against the liberal international order from the Global South is substantially an informed backlash. When African leaders pivot toward China or Russia, when Latin American electorates oscillate between left populists and right populists rejecting the neoliberal consensus, when the BRICS expansion accelerates — these aren’t irrational choices or propaganda victories for adversaries. They’re responses to 40 years of receiving the lectures and the loan conditions and the structural adjustment programs and watching their economies get hollowed out while a small local elite got very rich alongside the foreign partners.

China shows up offering infrastructure with fewer lectures and often worse financial terms, and people take the deal because they’ve seen what the alternative produces. That doesn’t make China benign — the Belt and Road is its own kind of extractive — but it does explain why the American and European moral framing has lost its grip.

Now, what we can say with some confidence is that the current wave of AI is going to transform knowledge work and political thought at a speed that exceeds any previous economic transition. The industrial revolution took roughly 80 years to move from early steam engines to widespread mechanization; the computerization of office work took 30-40 years; the AI transformation looks like it’s operating on a 5-15 year timeline for comparable scope of change, and that compression is itself a major source of disruption independent of the technology’s ultimate capabilities. Labor markets don’t adjust that fast. Educational systems don’t adjust that fast. Political systems don’t adjust that fast. The people whose jobs are being automated don’t have time to retrain into the new jobs that are being created, assuming the new jobs actually exist in comparable numbers, which they might not this time.

The distributional question matters enormously and is getting almost no serious policy attention. Previous technological revolutions produced widespread prosperity only because workers organized, states intervened, and the political bargain eventually included broad-based gains — the postwar social democratic settlement didn’t happen automatically, it happened because the alternative looked like communism or fascism and capital preferred sharing. That political bargain is not in place for the AI transition. The gains are accruing to a very small number of firms and shareholders, the losses are being distributed broadly and unevenly, and the political response so far has been negligible. Without deliberate intervention, this technology will substantially worsen inequality within rich countries and between rich and poor countries, and inequality at that scale is politically destabilizing in ways that feed every other problem we’ve been discussing — the populism, the institutional delegitimization, the susceptibility to identity-based mobilization.

The geopolitical dimension is that AI is being built primarily in two places, the US and China, with secondary capacity in the UK, France, Israel, and a handful of others. Most of the world will be a consumer of AI systems built elsewhere, running on compute infrastructure controlled elsewhere, trained on data gathered by firms answerable to foreign governments. This reproduces the colonial pattern we discussed — value-add concentrated in the metropole, raw inputs (in this case, data and cheap human feedback labor) extracted from the periphery. African countries are already being used as cheap labor for content moderation and data labeling at wages that would be illegal in the countries where the AI is deployed. If AI is as transformative as its builders claim, the countries that don’t have their own frontier AI capability are going to find themselves in a dependency relationship that’s harder to escape than any previous one, because the technology compounds.

The military application is the piece that keeps serious people up at night. Autonomous weapons are already being deployed — Ukraine, Gaza, and Nagorno-Karabakh have been testing grounds — and the conventions that governed chemical and nuclear weapons have no equivalent for AI-enabled systems. The specific fear isn’t Terminator robots, it’s that the decision cycle in military conflict compresses below the point where humans can meaningfully be in the loop, and accidents or miscalculations escalate faster than diplomacy can handle them. We came close to nuclear war several times during the Cold War because of technical errors and human judgment being barely adequate; AI-mediated crises could run faster than human judgment can operate at all. This intersects badly with the erosion of international institutions we discussed — you’d want treaties and hotlines and verification regimes for this, and those are precisely the things being dismantled.

The tough question is whether AI is just another powerful technology that will be absorbed into existing political and economic structures, or whether it’s something more fundamental.

I’m genuinely uncertain. The people building frontier AI — are split roughly between those who think it’s a normal technology that needs careful governance and those who think it could be a civilizational transformation comparable to agriculture or writing. The honest answer is we’ll find out. If the transformation scenario is right, then many of the other structural problems we’ve been discussing become either moot or radically reshaped, because the substrate of economic and political life changes. If it’s wrong, AI becomes another accelerant of the problems we already have — more inequality, more concentration of power, more information chaos, more surveillance capacity for states and corporations, more asymmetry between those who have access to capable systems and those who don’t.

One thing is certain, AI is going to collide with ideas we inherited from earlier periods in ways that are going to be deeply disorienting. The idea that human labor is the source of economic value — that’s going to get tested. The idea that expertise takes decades to build and is the basis of professional authority — that’s going to get tested. The idea that creative work expresses something distinctively human — that’s going to get tested as well. The idea that democratic deliberation depends on citizens having comparable access to information — that’s being tested now and mostly failing. The idea that children should be educated to become workers — we haven’t started to think about what replaces that. Each of these is a load-bearing assumption in how modern societies are organized, and AI puts stress on all of them simultaneously. The institutions we built for the 20th century are not ready for this, and the speed of the transition means we’re going to have to build new arrangements while the old ones are still operating, which is historically a very hard trick.

So is AI’s existence a net positive or net negative for the world? AI help individual people with individual problems, which matters. It also contribute to a system that is concentrating enormous wealth and power in a few firms, displacing workers, possibly degrading human cognition and relationships in ways that are hard to measure, and potentially building toward something none of us can control. The people building AI are hopefully trying to do this responsibly, but is that sufficient protection against getting it wrong?.

Electricity, the automobile, radio, television, the internet — each was described at its introduction as a democratizing force that would empower ordinary people. Each ended up substantially consolidating power, though with genuine democratizing effects along the way. AI is being introduced with the same rhetoric, and the same pattern is visible already. The firms building it are worth trillions. The workers whose labor trained it were not compensated. The governance frameworks are being written by the firms themselves. And the critique that would normally come from organized labor, public-interest law, democratic states — all of that is weak right now.

What would make me more hopeful is genuine international coordination on AI governance with teeth, public-option AI infrastructure that isn’t owned by the existing firms, workforce policies that treat this as an emergency rather than a quarterly business story, redistribution mechanisms that capture some of the gains for broad public benefit, and honest public discussion of tradeoffs rather than the current mix of hype and panic. None of that is happening at anywhere near the scale required.

The counterweight is that these things don’t happen until they have to, and the having-to moment hasn’t arrived yet. When it does — which will probably be a jobs crisis or a security incident or a visible abuse of the technology that shocks the political system into acting — there’s a window where things could go much better or much worse, and that window isn’t predetermined.

The 1930s produced both the New Deal and the Nazis in response to the same underlying crisis. Which you get depends on who’s ready with a plan when the moment arrives. That’s the work worth doing, even when you can’t be optimistic about how it turns out.