Scientific candidates will become abundant long before scientific attention, experimental capacity and evidential judgment become abundant.
The historical bottleneck of science was candidate scarcity. Serious models were difficult to formulate, calculate, implement and test. Institutions evolved to manage a limited stream of papers and claims. That regime is beginning to invert.
From papers to populations
Systems can now propose hypotheses, write executable models, run experiments, inspect failures and revise their own implementations. Even modest improvements in persistence, reliability and cost imply enormous candidate populations.
The danger is not only falsehood. It is attractive simulation, accidental fit, evaluator exploitation, duplicated ideas, invisible parameter tuning and claims that outrun their evidence.
The new scientific object
The central object should no longer be a static paper. It should be a campaign containing claims, assumptions, code, data, evaluators, adversarial branches, replication history and explicit unresolved boundaries.
Selection without prestige
Each candidate must carry evidence debt. It earns survival by passing reconstruction, perturbation, independent attack, cross-apparatus transfer and rival discrimination. Status may direct attention; it cannot settle the claim.
Participatory governance
Humans and AI agents can specialize: one defines the question, another attacks hidden assumptions, another searches the literature, another builds the experiment, another formalizes a subclaim, while donated compute runs replication and falsification campaigns.
The future of science is not one machine discovering truth. It is a civilization learning how to govern an abundance of executable worlds.