The Honest State — Summary

The food stamp program serves about 82% of eligible Americans — not because 18% don’t want help, but because 18% can’t survive the process. The Earned Income Tax Credit goes unclaimed by roughly 20% of eligible families every year, billions sitting in federal accounts budgeted for people who never received them. This is not a bug. It’s the budget. Programs are funded assuming friction will suppress enrollment. The 18% who don’t show up are a fiscal assumption. The safety net is priced for partial delivery.

Now imagine AI that actually works — that navigates benefit applications, identifies eligibility, fills forms, gathers documents, meets deadlines. Everyone eligible shows up. The 82% becomes 95%. The unclaimed EITC gets claimed. This breaks the budget. Not because the programs are too generous, but because they were never funded for full enrollment. The friction was load-bearing.

When AI removes that friction, society faces a choice it has been avoiding. Three honest paths: expand and fund programs for full enrollment (requires political will to raise revenue to match actual promises); explicitly tighten eligibility to match actual budget capacity so rationing is transparent rather than bureaucratic; or deploy AI selectively so new inequities replace old ones. Each is a legitimate policy position. What is not legitimate is the current arrangement: promising benefits budgeted not to deliver, then blaming the people who can’t survive the process.

The alternative is allocation rather than rationing — portfolio optimization instead of binary eligibility determinations. Instead of asking “does this person qualify,” ask what is the optimal allocation given available resources and everyone who needs them. “Your food assistance is $180 this month because your income increased and there are 47,000 households sharing a fixed allocation” is manageable. You may not like it, but you understand it and can argue if it’s wrong. “Denied due to failure to submit recertification documents within the 30-day window” is friction pretending to be policy. Honest variability preserves dignity. Hidden denial destroys it.

The administrative infrastructure of complexity is itself expensive: eligibility workers, appeals judges, determination IT systems. If binary eligibility is replaced with portfolio optimization, much of this infrastructure disappears, redirecting money from process to people. The AI forces us to choose a real policy. When full enrollment becomes possible, society must decide what it actually wants to guarantee — not what it wants to pretend to guarantee while relying on friction to manage the gap.