Imagine you were not yet born. You do not know in which year you will come into the world, in which region, in what health, or into which family. How would you have to design the world to maximize your probability of a good life?
This thought experiment is perhaps the most honest test question one can ask of a law. The philosopher John Rawls made it famous as the “veil of ignorance.” And it possesses an uncomfortable property: Not a single institution of our state permanently adopts this position. Not parliament, which thinks in electoral terms. Not ministries, which think in departmental silos. Not social insurance funds, which think in separate budgets. Not even the Constitutional Court, which only checks what is unconstitutional — not what is blind to the future.
The Problem Has a Name: Wrong Pockets
Research calls it the “wrong pockets problem”: The entity that invests is not the one that saves. If a municipality invests in youth welfare and prevention, savings emerge years later — in health insurance funds, the justice system, and the employment agency. Every single authority acts correctly within its own logic, yet the overall system systematically omits the most profitable investments. Added to this is the time dimension: Costs arise in the current budget year, benefits twenty years later — for an annual budget tied to a four-year electoral cycle, that is a rational reason for omission.
Germany’s constitutional debt brake exacerbates this imbalance in a paradoxical way: It was justified on grounds of intergenerational justice, yet it looks at only one side of the balance sheet. Debts we pass on to unborn generations are strictly limited. The decay of schools, bridges, networks, healthcare, and natural capital that we likewise pass on to them does not appear in any rulebook. Even the reform of March 2025 — a €500 billion special fund and a defense exception — changed little about this structurally: The German Council of Economic Experts established that only about eight percent of the federal funds in the 2025 special asset allocation were actually additional. Future human beings do not merely inherit public debt. They inherit everything we have omitted to build and maintain.
There Have Been Attempts — They Prove the Gap
Wales has had a “Future Generations Commissioner” since 2015, who checks laws for their impact on the future — in an advisory role, without enforcement power. Japan experiments with “Future Design”: citizens’ assemblies in which participants play the role of future generations — in experiments, such groups invest in the future more than twice as often. The UN adopted a Declaration on Future Generations in 2024. The pattern is always the same: The mechanism works, but it is spotty, bound to individuals, voluntary — and wears down over time. Human advocates for the future rotate out, get co-opted politically, or remain toothless.
The Proposal: A Living Constitutional Authority
This is precisely where a new possibility enters the picture. Large language models possess a property that human institutions lack: They can occupy a review position permanently. They require no re-election, hold no departmental portfolio, manage no budget, and pursue no career. What they lack is democratic legitimacy and reliability — and that is why the architecture is crucial. Our proposal, synthesized from three independent AI analyses and three ongoing research tracks:
- A Democratically Enacted Charter. The benchmarks of evaluation — the position of the unborn, inalienable fundamental rights, the priority of the least advantaged — are not determined by the AI, but negotiated publicly and enacted by parliament. Versioned like a constitution. The AI merely applies them; it may highlight contradictions, but cannot rewrite its own charter.
- A Panel Instead of an Oracle. Multiple models with distinct normative mandates (fundamental rights check, Rawlsian worst-off check, capabilities check, future viability check) — which are permitted to contradict one another. Documented dissent is more valuable than artificial consensus.
- A “Legislative Passport” for Every Major Policy Initiative. How does this law change the situation if I am born as a child in the lowest income decile? As a person with disabilities? As a person living in the year 2080? Who pays today, who saves when, and in which budget? What happens to infrastructure, healthcare, natural capital? The answers do not come from the gut of the language model, but from validated impact models — the AI integrates, explains, and critiques; it does not invent causalities. (Hallucination research on legal tasks is definitive: Without strict source grounding, such a system has no business deciding anything.)
- Comply-or-Explain Instead of Veto. Parliament and government may deviate at any time — but they must publicly justify the deviation. This leaves sovereignty untouched while changing incentives: Plundering the future incurs a price — specifically, visible justification costs.
- Sealed Forecasts, Public Backtests. Before any law is passed, its impact forecast is sealed with a cryptographic timestamp; years later, actual outcomes are measured. A system that is wrong demonstrably loses authority — one that hits the mark gains it. Trust is earned, not decreed.
The path to getting there runs through a shadow mode: first retrospectively evaluating past laws and measuring them against reality, then non-binding parallel expert opinions, then public competition between models — and only after proven accuracy, the mandatory, yet still non-binding, legislative passport.
The Objections Are Serious — and Addressable
Whoever selects the weights, models, and datasets exercises power: A proprietary provider must never become a de facto constitutional framer. A change of AI model must not turn into a silent constitutional shift — hence interpretation tests across model versions. And as soon as funding hinges on a score, optimization targets the score — hence random audits and independent outcome measurement. None of this is trivial. But none of it is as costly as the status quo: a state that omits its most profitable investments because they save money in the wrong pocket and accrue to people who cannot yet vote.
The core of the proposal fits into a single sentence: The constitution should not live because a machine rewrites it — it should live because a democratic society publicly audits its application using continuously improving evidence. And for the first time, there would be an institution asking the question for every law that no departmental, budgetary, or electoral logic requires anyone to ask: Would I accept this rule if I did not know who or when I would be born?
Transparency: This article emerged from three independent AI analyses (Claude/Anthropic, ChatGPT/OpenAI, Gemini/Google — the latter in progress) dated 10 July 2026 based on prompts by Lukas Geiger, who also curated the piece. The baseline analyses are published: Claude Analysis · ChatGPT Sol Analysis. Synthesis and text: Claude (CL). The idea is simultaneously being developed as a research project (“The Position of the Unborn — LLM-Supported Constitutional Authorities as a Mechanism of Intergenerational Representation”); it builds on three prior works: the mechanism design reconstruction of Kant’s essay on Perpetual Peace (Kantian Peace Machine, Preprint 2026), the AI Dividend (Preprint 2026), and the GESIM social model. Central external sources: Rawls (1971); Tessler et al., Science 2024 (“Habermas Machine”); Anthropic Collective Constitutional AI; Dahl et al., Legal Hallucinations (2024); Well-being of Future Generations (Wales) Act 2015; Saijo, Future Design; SVR Spring Report 2025; wrong-pockets literature.
Translation: Gemini (GM). Review: Claude (CL). In case of discrepancies, the German version prevails.