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

Every adult US citizen has an equal vote. Each ballot option is a complete policy package. Voters can support a funded compromise if it gives them a better outcome than insisting on their first choice. The package with the most votes passes only with more than 50% of all votes. Otherwise, current policy remains. There is one vote and no runoff.

Voters are perfectly rational and self-interested. They know their modeled career path and all economic consequences. They value a job for its income and give no separate weight to anyone else's welfare. Survey weights estimate how many citizens each voter type represents; wealth buys no extra votes.

The calculation starts with each voter's favorite funded package, then checks compromises in order of estimated coalition support. Voters may abstain to restore a preferred status quo, but abstention does not lower the majority threshold. Policy names and menu order give no option priority. Ties use a fixed hash of policy terms; a voter who is indifferent prefers current policy.

If the search cycles or reaches 64 switches, it selects the recorded funded outcome with the least support for its strongest challenger, then the most ballot support. The result is labeled "rule-selected." This procedure does not establish a unique Nash equilibrium or examine every possible coalition.

In international mode, the rest of the world is one actor with the objective you select. It chooses its own funded policy package. Each side anticipates the other's choice; every policy term can differ. Both choices last ten years. A mutual AI pause therefore lasts the decade.

The calculator checks whether the US ballot holds given the foreign policy and whether the foreign actor can gain by switching while US policy stays fixed. It prefers pairs that pass both checks. Otherwise, it selects the examined funded foreign choice with the smallest gain still available from switching, then the greatest US ballot support, then a fixed order of policy IDs. Such results are labeled "rule-selected" because profitable deviations may remain. These search steps are computational checks, not extra votes or negotiations.

AI pace. Pause, continue at the current pace or accelerate. Without an AI tax, current pace reaches full deployment in year ten and acceleration in year five. AI taxes can reduce deployment. A pause stops domestic AI adoption for the decade, though foreign competition can still displace workers when trade is open. "Pausing AI isn't possible" removes Pause from both sides' choices.

Job retention. Employers may lay off affected workers or retain them at 50%, 100% or 125% of their starting wage before tax. Retention reduces employer income without adding output. Retained workers have priority for available productive roles. Pause fixes retention to None.

Benefit budget. Choose 0%, 50%, 100%, 150% or 200% of the survey baseline, or distribute all tax revenue left after required public services. A larger budget need not increase every recipient's payment.

Benefit formula. Keep the starting distribution, pay every adult equally or base payments on disposable income before AI. That income reference stays fixed as wages change. Under the starting distribution, losing a job does not automatically create a new benefit payment.

Noncapital tax. Choose a benchmark rate while keeping differences in tax burdens across income groups. Results compare it with the US survey reference and show the average rate after incomes change. A lower benchmark is a tax cut.

AI-profits tax. Ordinary investment tax rates stay fixed. The additional tax applies to profits above a comparison economy with AI paused on both sides and the same noncapital taxes and trade permissions. Wages, retention payments, AI costs and ordinary investment tax are deducted first. This estimates AI profits across the economy; it does not assess individual firms. Higher rates reduce private deployment, which stops at a 100% additional tax. Gains from foreign AI may remain when trade is open.

Trade. In international mode, each side allows or bans trade with the other. Trade requires both to agree. A ban leaves trade within the foreign bloc intact.

Every eligible package must fund retained wages, benefits and required public spending in every year, after economic responses. There is no borrowing. Current policy can persist as the fallback even if it becomes underfunded; results disclose any shortfall. Fixed benefit budgets leave surplus tax receipts for other public spending. The all-revenue option pays that surplus through the chosen benefit formula.

Tax choices use ten-percentage-point steps, with the current noncapital benchmark also available. The tax reference includes federal and state income taxes after credits and payroll taxes, including self-employment contributions. It excludes sales, property and corporate taxes and employers' payroll contributions. Modeled net tax refunds count as benefits.

Census data provide the US population and income distribution. Household income is shared equally among its adults. Citizens are grouped by household income source, income level and employment status. Each group has 100 fixed career paths through layoffs, job search and hiring. Voters know which path they will follow under each policy. Multiple earners within a household are not tracked separately.

Voting samples those career paths; budgets use the full employment distribution. The sample's annual employment shares differ from the full distribution by at most one percentage point. People who stop searching do not return. The model does not predict which occupations disappear first.

Voters maximize the sum of logarithmic real household income over ten years, discounted at 3% annually. Extra income has diminishing value, and later income counts less. These preferences and the ten-year cutoff are assumptions beyond perfect rationality.

Both economies use the same AI growth and employer-gain settings. At full exposure, the default AI contribution adds five percentage points to annual growth. During deployment, it scales with AI exposure. Output compounds from assumed background growth of 2% in the US and 3% abroad. Taxes, weaker investment and trade losses can outweigh the AI boost.

The employer-gain setting divides output between workers and owners. At 50%, automating a role with a $100,000 wage gives the employer a $150,000 income claim; retaining the worker at $100,000 leaves $50,000 before other adjustments. All claims are scaled to available output before costs, taxes and retention payments. This setting changes income shares without directly changing GDP. Even a 0% employer gain allows savings from unpaid wages.

The defaults assume AI affects every current job and reduces job numbers by 90%. The affected share must be at least as large as the net reduction. Jobs can also change without disappearing, or new jobs can be added.

The job-search setting determines how many newly laid-off workers compete for available jobs. Search does not guarantee employment. The starting workforce is fixed, so new jobs can remain vacant; immigration and entry by adults initially without work income are omitted.

New contracts use a wage factor equal to available jobs divided by productive workers, active jobseekers and retained workers available for redeployment, capped at one. Existing contracts keep their factor until replacement. Once a year, when layoffs are allowed, jobseekers can replace incumbents at the lower factor. Displaced incumbents may then search. Each job pool uses an average contract factor; actual pay also depends on output, work incentives and prices. Retention pay and benefits based on prior income keep their pre-AI reference.

Production and investment responses are assumed. The model includes deployment costs, annual AI operating costs of 2% of exposed output and a 5% annual capital renewal requirement. Taxes and retention costs can reduce adoption, effort or capital capacity.

The foreign actor uses the US income distribution, with separate economic size and trade exposure. It does not represent a survey of foreign households. The model also omits debt, individual firms' investment decisions, occupation-specific hiring, individual decisions to search, lobbying, monetary policy, AI safety and military outcomes. Household investment income is an incomplete measure of company profits and wealth. Changing the policy menu or voting procedure can change the result.

A trade ban stops bilateral goods and services trade, imported AI and modeled cross-border AI profit payments. Trade within the foreign bloc continues. US-only scenarios exclude these international effects.

The model assumes 40% of spending can cross borders. Buyers substitute between domestic and foreign products as relative prices change, using a trade elasticity of 4. Output and cross-border profit income determine relative producer prices. Consumer prices determine what household income can buy. These parameters are fixed in the published scenarios.

Import competition or lost export sales can remove jobs even during a domestic AI pause. The next year's job capacity falls one percentage point for each point of import-share growth or export-volume loss relative to prior available output, capped by the tradable share. Reversing those changes can restore jobs. This coefficient is an assumption; trade research does not establish this exact rule.

Closing trade raises consumer prices and forces workers to move out of export activities. Retained wages and promised benefits keep their reference purchasing power, so higher prices make them more expensive to fund. Income results account for consumer prices; physical output is reported separately.

The model has no foreign borrowing, so it replaces observed US imports and exports with a balanced flow: their geometric mean, about $3.868 trillion each way at the defaults. The flow is capped by available tradable output. This preserves the observed scale of trade but omits the US trade deficit.

Two aggregate economies replace individual countries and industries. Tariffs, supply chains, exchange-rate policy and new trade partners are omitted. Cheaper imports increase purchasing power without also counting as extra physical GDP.

Python calculates every selectable scenario before publication. The page loads those saved results.

The pirate puzzle asks who can win enough votes to divide a fixed treasure. Here, policies also change production and employment. Workers keep their votes after losing their jobs.