Can capitalism survive an AI-driven economy?
Capitalism can survive AI, but only if redistribution happens before ownership concentration captures the governments that would impose it.
Keynes, Friedman, and Hayek agree: AI productivity without redistribution kills the consumer demand that keeps markets running. Friedman argues cash transfers are enough; Keynes argues governments must act aggressively before the gap widens. Luxemburg points to the Weimar Republic: when capital concentrates this severely, owners capture the legislatures before redistribution becomes possible.
Hirschman closes the argument with a timing problem. Once AI restructures labour markets, reversing those arrangements costs far more than building them differently from the start. The split is not whether to redistribute, but whether cash transfers within current ownership structures are enough, or whether ownership itself must be addressed first.
Confidence summary: Strong convergence on the necessity of redistribution; sharp disagreement on whether redistribution within existing ownership structures can arrive in time.
1. The core argument
The most uncomfortable insight from this council is not that capitalism faces a threat from AI. It is that capitalism may already be generating the political conditions that prevent it from responding to that threat. Every member accepts the basic mechanism: productivity growth without redistribution destroys the purchasing power that keeps market economies solvent. That is not a radical claim. It is the lesson of the 1930s, stated plainly.
The harder problem is sequencing. Redistribution requires a political settlement. A political settlement requires functional democratic institutions with the capacity to tax concentrated power. And ownership of AI is concentrating faster than any prior technology, in fewer hands, with greater capacity to shape the legislative and regulatory environment. The question is not whether governments know what to do. The question is whether they retain the institutional independence to do it before the window closes. On that question, the council divides in ways no consensus can paper over.
2. How each member frames it
John Maynard Keynes opens with a concession: he was right about the productivity gains of technological transformation and wrong about the politics. That candour matters. His 1930 forecast of fifteen-hour working weeks was not foolish; it was incomplete. Markets do not automatically route the gains of new technology to wage earners. His core move here is to insist that demand management is not optional ornamentation on an otherwise self-correcting system. It is the structural condition for the system's survival. Where his card stops, his argument continues: he would accept inflation risk and fiscal deficits as preferable to demand collapse, and he would reject Friedman's faith that cash transfers alone are politically durable without an aggressive prior commitment to keeping labour income viable.
Friedrich Hayek is more concessive than his reputation allows. He accepts redistribution. What he rejects is the conflation of redistribution with planning, and he names the specific danger in 2026: AI systems optimising for measurable proxies will be offered as substitutes for price discovery, not supplements to it. His deeper worry is institutional. He watched centralised economic management fail not because the planners were stupid but because the knowledge problem is not a solvable engineering challenge. His challenge to Friedman, whether a negative income tax can function without a healthy price system beneath it, reveals the limit of his own position: he needs Friedman's policy to work, but he cannot guarantee the price system will remain legible as AI intermediates more transactions.
Milton Friedman wants to cut directly to the cash. His 1962 argument for a negative income tax was a critique of bureaucratic welfare programmes: they trap recipients, distort incentives, and grow the administrative state. His application here is consistent. Tax AI profits, transfer the proceeds directly to citizens below a threshold, and let the price system do the rest. The part his card omitted is the political fragility of that programme. Friedman saw in Chile what happens when clean economic prescriptions meet institutional disorder. He knew that simple policies require capable, neutral administrators. His challenge to Luxemburg, whether AI profits can actually be taxed before capital relocates, is the honest weak point in his own proposal, not just a rhetorical move.
Rosa Luxemburg refuses to treat this as a technical problem with a technical fix. Her 1913 argument was that capitalism requires a broad distribution of purchasing power to realise its own surplus, and that wage labour was the mechanism. AI removes that mechanism at scale. The Weimar Republic is her anchor: democratic institutions collapsed not because no one knew the correct economic policy, but because concentrated power had already captured the institutions that would have implemented it. She does not claim AI makes capitalism impossible. She claims it removes the political preconditions for reform faster than reformers can act. That is a tighter and more dangerous argument than simple anti-capitalism.
Albert O. Hirschman provides the timing argument the others need but cannot quite state. His 1970 framework showed that exit disciplines organisations; when customers or workers can leave, managers must improve. AI displacement is involuntary exit: workers are expelled from the wage relationship, not departing it. Voice becomes the only instrument, precisely as the political conditions for effective voice are deteriorating. His observation from Colombian infrastructure projects in the 1950s adds empirical weight: once large capital commitments lock in a particular economic arrangement, reversing the arrangement costs multiples of the original investment. The implication is stark: the moment for effective institutional intervention is now, not after the transition consolidates.
3. Where the council agrees
The most surprising point of agreement is this: Hayek and Luxemburg, whose political instincts are about as opposed as any two thinkers on this council, both accept that ownership concentration is dangerous. Hayek frames it as a threat to market functioning; Luxemburg frames it as a threat to democratic institutions. The practical implication is identical: ownership structures matter and cannot be treated as politically neutral. Beyond that, every member accepts that some form of cash transfer to displaced workers is necessary. Friedman argues it is sufficient; the others argue it is necessary but not sufficient. All five accept that the window for reversible policy action is narrowing. Hirschman gives this the clearest analytical form; Luxemburg gives it the most urgent political register. That convergence on urgency, across five members with profoundly different priors, is itself a signal.
4. Where the council splits
The line runs between Friedman on one side and Luxemburg and Hirschman on the other, with Keynes and Hayek occupying unstable middle ground. Friedman holds that redistribution within existing ownership structures, a well-designed negative income tax funded by AI profits, is both sufficient and institutionally safer than ownership intervention. Luxemburg and Hirschman hold that ownership concentration will capture the very governments required to impose and sustain that tax. Both sides have a real argument. Friedman is correct that ownership regulation carries serious risks of regulatory capture and market distortion. Luxemburg and Hirschman are correct that the political preconditions for even Friedman's preferred policy may not survive the consolidation of AI ownership. This is not a disagreement that additional evidence can resolve. It is a disagreement about how much time remains.
5. For a policymaker to decide on
The council cannot resolve this choice: should AI governance prioritise redistribution now, within existing ownership structures, accepting that this may prove insufficient if capital relocates or captures the tax regime; or should it impose structural constraints on AI ownership first, accepting that this will slow deployment, attract legal challenge, and risk exactly the regulatory overreach Hayek and Friedman warn against? The first option bets on institutions holding. The second bets on intervention before they weaken further. The timing is the decision.