When AI does the work,
how would rational voters respond?

Every adult US citizen has an equal vote. Survey weights represent their shares of the population; the model does not use a fixed number of artificial voters. Job retention, benefit amounts, the benefit formula, non-capital taxes and capital taxes are separate decisions. A result is stable when no change to one decision is preferred by more than 50% of the electorate. A package that changes several decisions together may still win.

Each voter is perfectly rational and self-interested. They understand all modeled consequences and maximize their own expected household-income utility over ten years, including the effects on investment and production. They place no separate value on anyone else’s welfare. Ties do not count as support for a change. A person keeps their vote when they lose a job. Job retention has no separate value beyond its income.

In international mode, the rest of the world acts as one decision-maker. It chooses its own complete policy, including AI deployment, to maximize the objective you select. The outcome without a treaty requires both US ballot stability and a foreign best response. The foreign policy itself does not require a US vote.

The search checks candidate policies, then verifies a selected result against every permitted change to one US decision and every foreign policy in the menu. Failure to find a stable result is not proof that none exists. Voting details show any remaining challenge. This verifies the outcome without a treaty, which is the fallback if negotiations fail. Policies are fixed for ten years; repeated elections and renegotiation are outside this model.

In international mode, treaty negotiation follows the verified outcome without an agreement. The foreign actor chooses its preferred offer from those a US majority would accept. Each voter compares the entire treaty with the fallback using their own expected income utility. A strict majority and a strict foreign gain sign the treaty automatically. The proposal rule gives the foreign actor bargaining power; rationality alone does not select it.

The offer menu includes every common policy package for both sides; every US package with the foreign fallback unchanged; every foreign package with the US fallback unchanged; and pairs where each side changes one fallback decision. Either deployment target can change. Both sides can have different policies. The calculator checks every offer in this menu, but not every possible asymmetric pair of packages. No accepted offer means no treaty in this menu, not that all agreements are impossible.

Commitments are binding for ten years and add no output or money by themselves. There are no cross-border side payments. A signed package is ratified as a whole, so its individual terms need not survive separate votes or unilateral foreign changes. The separate-ballot results remain available for the fallback without an agreement.

1. Job retention. Employers can lay off workers or retain obsolete roles at 50%, 100% or 125% of their starting wage before tax. These roles add no output. The employer bears the cost, which reduces capital income. Finding new productive work is a separate transition.

2. Benefit amounts. Government spending on the modeled benefits can be 0%, 50%, 100%, 150% or 200% of its survey baseline. The result states the proposed budget change separately from the amount actually funded. Higher spending does not mean every recipient receives more.

3. The benefit formula. Keep the current mix, pay every adult an equal amount, or link payments to prior income. “Prior income” means observed disposable income before AI, not lifetime earnings. The current mix preserves the starting distribution of benefits; losing a job does not automatically create a new unemployment-benefit payment.

4. Non-capital taxes. The result compares the selected tax benchmark with the survey-based US reference and also shows the actual average after incomes change. A lower rate is a tax reduction. The calculation retains differences in tax burdens across income groups.

5. Capital taxes. The same comparison applies to capital income. These are modeled average burdens, not statutory brackets or official separate rates for “workers” and “capital owners.” A person can receive both labor and capital income.

The tax reference covers modeled federal and state income taxes after credits, plus payroll taxes including self-employment contributions. It excludes sales and property taxes, corporate taxes and employers’ contributions for employees. Modeled net tax refunds count as benefits; this does not capture the full value of every refundable credit. A policy must fund the fixed reference commitment to other public spending to enter the vote. Benefits and retained wages may still fall short of their targets; voters compare actual payments. Surplus receipts fund non-transfer public spending rather than an additional dividend.

The US population and income distribution come from the Census survey. Household income is pooled and divided among adult household members. The model groups citizens by that household’s main income source, income level and their own employment status. Main sources include work, capital, public benefits and pensions. Benefit receipt and employment overlap. These group labels do not determine how anyone votes; their own projected income does.

International defaults use relative GDP and bilateral trade exposure. The foreign actor uses the same income distribution as a simplifying assumption; it is not a survey of foreign voters. AI gains, displacement, return to work and responses to taxes remain adjustable assumptions. Zero foreign frontier capability leaves the foreign economy and its imports in place.

GDP growth and employer gains are separate. The default assumes 5% annual real GDP growth at full AI exposure. During rollout, the potential growth rate is 5% multiplied by that year’s exposure and compounds each year. There is no separate background-growth assumption. The foreign economy has its own growth setting. Taxes and investment responses can change realized growth.

The employer-gain parameter divides that output. At a 50% gain, an automated role with a $100,000 prior wage gives the employer a $150,000 income claim. Keeping the employee at $100,000 leaves $50,000 before other adjustments. To keep total income consistent with the separate GDP assumption, all gross income claims are scaled to the available output before costs, taxes and retained wages are paid. This allocation rule can change workers’ and owners’ shares without changing the GDP path for a fixed policy. Changing those shares can change the policy voters choose.

The selected deployment target is reached by year ten. At full deployment, 100% obsolete roles and no return to productive work means every worker is affected. Taxes and required employer pay can delay rollout within those ten years and reduce productive effort. They can also erode existing capital capacity through an assumed 5% annual renewal requirement. This rate is a modeling choice, not an estimate from the cited data.

This is a comparison within a policy menu, not a full model of the world economy. Observed household investment income is not total company profit or wealth. The employer-funding limit and production responses remain modeling assumptions. It does not solve market prices, demand, debt, firms’ investment decisions or individual labor supply. It omits lobbying, monetary policy, AI safety and military outcomes. Changing the policy menu or voting procedure can change the result.

The pirate puzzle asks who can win enough votes to divide a fixed treasure. Automation changes how much is produced and who receives it. This simulator keeps the voting question and adds production, employment and government payments.