How Prediction-Free Research Funding Could Support Unexpected Discoveries

Getting your Trinity Audio player ready...

Scientific discovery is inherently uncertain. Researchers can define a question, choose rigorous methods, and explain why an investigation matters—but they cannot reliably predict what nature, mathematics, or experimentation will reveal.

Prediction-free research funding addresses this mismatch by reducing the need to promise specific discoveries in advance. Instead, it allocates some funding according to completed work, verified milestones, emerging evidence, and demonstrated scientific usefulness.

This does not mean abandoning prospective grants. Laboratories still need equipment, materials, staff, and time before results exist. Rather, prediction-free funding adds mechanisms that can recognize value after it becomes observable, including retroactive rewards, continuous micro-funding, milestone payments, and support based on downstream research dependencies.

The result could be a more adaptive funding system—one capable of supporting discoveries that no applicant, reviewer, or funding agency could have described beforehand.

What Is Prediction-Free Research Funding?

Prediction-free research funding is a financing model that does not require every award to depend on a detailed forecast of future scientific results.

A conventional grant application usually asks researchers to predict:

• what they will discover;
• which hypotheses will be confirmed or rejected;
• what methods they will use;
• how long each stage will take;
• what impact the project will produce;
• why the proposed work is more promising than competing proposals.

These questions are reasonable for budgeting and risk management. However, they create a structural problem: the most fundable project may become the one that is easiest to describe, not the one most likely to change science.

Prediction-free mechanisms can instead ask:

What scientifically valuable work has already been produced, validated, reused, extended, or shown to be necessary?

Possible evidence includes published results, mathematical proofs, datasets, experimental replications, open-source scientific software, documented corrections, technical methods, and demonstrable contributions to later research.

Why Scientific Discovery Cannot Be Fully Predicted

Research proposals are predictions about future knowledge. Science itself is a process for discovering where those predictions are wrong, incomplete, or unexpectedly productive.

Researchers may begin with one objective and encounter:

• an anomalous experimental result;
• a mathematical structure unrelated to the original theorem;
• an unexpected biological mechanism;
• a reusable method developed as a side effect;
• a failed hypothesis that eliminates an entire class of explanations;
• a software tool that becomes more important than the project that created it.

Serendipity in science is not merely luck. It usually combines an unexpected observation with a researcher capable of recognizing its importance. Research on scientific serendipity has found that openness to unfamiliar information can contribute to more innovative work. One large-scale study examined approximately 2.4 million papers and found that scientists exposed to less familiar work could produce more innovative research when they were sufficiently open to such encounters.

Unexpected discoveries also depend on infrastructure. A recent analysis argues that apparently serendipitous breakthroughs often occur shortly after a new instrument, method, or research tool makes the relevant observation possible.

Funding systems therefore should not try only to identify the winning hypothesis. They should also maintain the people, tools, datasets, experiments, and intellectual diversity from which unplanned discoveries can emerge.

How Proposal-Based Funding Can Discourage Surprise

Traditional grants do not explicitly prohibit unexpected discoveries. The problem is subtler: proposal evaluation can reward characteristics that make surprise less likely.

Reviewers Prefer Legible Projects

Reviewers must compare applications under limited time and incomplete information. A project with familiar concepts, established methods, preliminary data, and a predictable research path is easier to evaluate than a genuinely unfamiliar idea.

That does not mean reviewers oppose innovation. It means the decision architecture can make uncertainty look like weakness.

A field experiment on scientific proposal evaluation found persistent concerns about conservatism and examined how communication among evaluators can influence the selection of unconventional projects.

Applicants Learn to Minimize Apparent Risk

Researchers understand that proposals must appear ambitious but feasible. They may therefore present uncertain research as a controlled sequence of expected outcomes.

This can encourage applicants to:

• emphasize predictable extensions of established work;
• conceal speculative branches;
• promise results that fit existing disciplinary categories;
• avoid questions without sufficient preliminary data;
• spend time constructing a persuasive narrative rather than investigating the problem.

The proposal may become a form of scientific marketing: sufficiently novel to sound important, but sufficiently conventional to reassure reviewers.

Feasibility Can Be Confused with Value

A highly feasible project can be useful, but feasibility is not equivalent to scientific importance.

Review criteria often emphasize whether the proposed work can be completed as described. Yet an unexpected discovery is important precisely because it was not contained in the original description.

Research on grant-review criteria has found that reviewers may give substantial weight to feasibility, rigor, applicant experience, and methodological clarity, while originality can be evaluated less consistently.

Prediction-Free Funding Does Not Mean Evidence-Free Funding

The phrase “prediction-free” should not be misunderstood. It does not mean distributing money without standards, evidence, or accountability.

A credible system would still evaluate:

• whether a result is genuine;
• whether methods are transparent;
• whether claims are supported;
• whether data and code are available where appropriate;
• whether other researchers can reproduce or build on the work;
• whether an output solves a scientific problem;
• whether a contribution remains useful over time.

The difference is temporal.

A predictive grant asks:

What valuable work do we believe this researcher will produce?

A prediction-free reward asks:

What valuable work has this researcher produced, and what does the available evidence now show about it?

The second question is not automatically easy. Scientific impact can take years to become visible, and fashionable work can accumulate attention without being foundational. Nevertheless, completed outputs provide evidence that does not exist when reviewers assess a proposal.

Mechanisms for Prediction-Free Research Funding

Prediction-free funding is not a single mechanism. It is a family of complementary approaches.

1. Retroactive Research Funding

Retroactive funding rewards scientific work after it has produced observable value.

A researcher might receive support after:

• publishing a significant result;
• releasing a dataset that other laboratories use;
• proving a theorem that resolves an established problem;
• building software that becomes research infrastructure;
• correcting a widely repeated scientific error;
• developing a method adopted across multiple projects.

Retroactive funding changes the role of scientific finance. Instead of only buying promised work, funders also reward demonstrated contribution.

This resembles a scientific prize, but it need not be limited to rare, prestigious awards. A continuous system could distribute many small and medium-sized rewards across the research ecosystem.

The AI Internet-Meritocracy model proposed by World Science DAO is an example. It aims to evaluate visible scientific and open-source contributions and allocate funding according to evidence of merit and use.

2. Continuous Funding Based on Accumulating Evidence

Scientific value rarely becomes visible at one moment. A paper may initially receive little attention and later become important when another field discovers its method.

A continuous funding system could update assessments as evidence accumulates:

• an independent replication is published;
• a proof is formally verified;
• a dataset receives sustained reuse;
• a software package becomes a dependency;
• a concept is applied in another discipline;
• later work reveals that an earlier result was foundational.

This approach avoids the binary structure of many grant competitions, where a project is funded or rejected based on one review round.

As discussed in the comparison between AIIM and Horizon Europe, continuous and retroactive payments could allow neglected work to become fundable when its usefulness eventually becomes visible.

3. Milestone Funding Without Predetermined Conclusions

Some research cannot begin without advance funding. Prediction-free principles can still reduce unnecessary forecasting through milestone payments.

The milestones should concern research actions and evidence, not predetermined conclusions.

Appropriate milestones might include:

• constructing an experimental apparatus;
• collecting a defined dataset;
• completing a blinded trial stage;
• publishing a protocol;
• releasing source code;
• testing a theorem under specified assumptions;
• submitting results to independent replication.

An inappropriate milestone would require the experiment to confirm the funder’s preferred hypothesis.

Researchers should be accountable for doing rigorous work—not for forcing reality to match the proposal.

4. Funding Tools and Research Environments

Prediction-free funding can support the conditions under which discoveries occur rather than requiring a precise forecast of the discovery itself.

Eligible outputs could include:

• microscopes and sensors;
• mathematical libraries;
• open laboratory automation;
• scientific databases;
• simulation frameworks;
• formal-proof systems;
• new measurement techniques;
• long-term observational infrastructure.

The NIH and NSF have created programs intended to support high-risk or potentially transformative research, showing institutional recognition that ordinary review mechanisms may not adequately serve unconventional work. The NIH High-Risk, High-Reward Research program includes awards designed for unusually creative investigators, while NSF initiatives have explicitly encouraged novel and potentially transformative research.

Prediction-free funding extends this logic: rather than merely trying to predict which high-risk proposal will win, it can reward useful capabilities and outputs as their value becomes evident.

5. Portfolio Funding

No funding system can reliably identify a single future breakthrough. A better approach is to finance a diversified portfolio of:

• conventional research;
• high-risk experiments;
• replication work;
• independent researchers;
• foundational theory;
• scientific infrastructure;
• retroactive rewards;
• early-stage exploratory work.

Most projects in a high-uncertainty portfolio will not become major breakthroughs. That is not necessarily failure. The purpose is to create enough independent paths for some projects to produce unusually valuable results.

The portfolio should be assessed at the system level rather than demanding that every funded project justify itself as a future revolution.

How Prediction-Free Funding Helps Unexpected Discoveries

It Allows Researchers to Follow Anomalies

A rigid grant may require researchers to remain within an approved work plan. Prediction-free or flexible funding gives them more freedom to investigate results that were not anticipated.

An anomaly may initially appear to be:

• an error;
• contamination;
• a computational defect;
• irrelevant noise;
• an observation outside the project’s scope.

Many such anomalies are indeed mistakes. But some are clues. Researchers need time and discretion to determine which is which.

It Rewards Useful Failure

A failed hypothesis can make a substantial contribution if the experiment is rigorous and the result is made available.

Negative results can:

• prevent duplication of unsuccessful work;
• constrain theories;
• expose defective assumptions;
• improve experimental design;
• redirect an entire research program.

Proposal-driven funding often encourages researchers to present success as confirmation of the original plan. Prediction-free assessment can instead reward the informational value of the outcome.

It Can Recover Neglected Research

A discovery may be ignored because it appears in an obscure journal, comes from an independent researcher, crosses disciplinary boundaries, or lacks fashionable terminology.

A system that continuously analyzes scientific dependencies could identify when later research begins relying on such work.

This is particularly important in mathematics and fundamental theory, where practical impact may be delayed for decades. The value of a definition, theorem, or abstraction may become visible only after an unexpected connection is established.

It Reduces Dependence on Grant-Writing Skill

Writing a strong proposal is not the same activity as producing strong science.

Researchers working outside major universities may lack:

• grant offices;
• professional editing;
• institutional reputation;
• preliminary funding;
• established reviewer networks;
• time to prepare repeated applications.

Funding based partly on visible outputs can provide another route to recognition. The Science DAO model for independent researchers explores how transparent, decentralized mechanisms could widen access beyond conventional institutions.

It Supports Contributions That Do Not Fit Grant Categories

Unexpected discoveries often cross administrative boundaries. A mathematical technique may affect biology; a physics instrument may produce archaeological evidence; a software library may become essential to climate research.

Proposal calls are commonly divided by field, program, institution, and policy objective. Output-based systems can evaluate relationships after they emerge rather than requiring researchers to predict the correct category beforehand.

Can Artificial Intelligence Identify Unexpected Scientific Value?

AI could help prediction-free funding systems process more evidence than a conventional committee can examine.

An AI-assisted evaluation system might map:

• citations and their context;
• theorem dependencies;
• dataset reuse;
• software dependencies;
• replication outcomes;
• corrections and retractions;
• methodological adoption;
• cross-disciplinary applications;
• expert assessments;
• the historical development of a research problem.

However, AI should not be treated as an oracle.

A language model can repeat popular errors, favor well-documented fields, misread technical novelty, or confuse attention with importance. Citation counts can also be manipulated and may reflect criticism rather than dependence.

AI therefore should support a plural evaluation process involving transparent criteria, expert review, open challenges, appeals, audit trails, and adversarial testing. The purpose is not to replace one unaccountable committee with one unaccountable model.

The broader AI meritocracy approach to research funding describes how automated analysis could be combined with human governance, scientific review, provenance tracking, and fraud controls.

Limitations of Prediction-Free Funding

Prediction-free funding cannot replace all prospective research finance.

Expensive Research Must Be Funded Before Results Exist

Particle accelerators, clinical studies, space missions, and advanced laboratories cannot operate on retroactive rewards alone. They require long-term commitments and substantial initial capital.

Impact Can Take Too Long to Appear

Some fundamental research becomes influential only decades later. A purely impact-based system could underfund young ideas while waiting for external validation.

Early-stage baseline funding therefore remains necessary.

Popularity Is Not Scientific Merit

Citation counts, downloads, media coverage, and community votes can all favor fashionable fields. Evaluation must distinguish visibility from genuine scientific dependence.

Researchers Cannot Bear Unlimited Financial Risk

A system that pays only after success would privilege wealthy researchers who can finance themselves. Prediction-free funding must not become unpaid speculative labor.

A fair model should combine advance support with retroactive rewards.

Metrics Can Be Manipulated

Researchers may create citation rings, artificial software dependencies, coordinated endorsements, or superficial replications.

Any automated funding system needs identity controls, anomaly detection, conflict-of-interest disclosure, independent audits, and meaningful penalties for fraud.

A Hybrid Model Is More Realistic

The strongest funding architecture is likely to combine several mechanisms:

Research need	Suitable mechanism
Equipment and laboratory setup	Prospective grants
Clearly defined intermediate work	Milestone payments
Early exploratory research	Microgrants or lotteries among qualified proposals
Completed scientific contribution	Retroactive rewards
Long-term reuse or dependency	Continuous impact-based funding
Major verified breakthrough	Scientific prizes
Replication and correction	Dedicated verification funding
Essential shared infrastructure	Sustained institutional or public-goods funding

This model does not require a conflict between grants and retroactive rewards. Each solves a different problem.

Prospective grants provide capacity. Milestones provide accountability. Retroactive funding rewards demonstrated value. Continuous funding recognizes contributions whose importance develops over time.

From Predicting Winners to Maintaining Discovery Capacity

The central mistake in research funding is assuming that scientific administration can consistently identify future discoveries in advance.

Expert judgment can evaluate rigor, plausibility, ethics, and competence. It cannot eliminate fundamental uncertainty.

A more resilient system would therefore ask fewer researchers to pretend that discovery is predictable. It would finance a broad portfolio, preserve room for exploration, and reserve part of its budget for rewarding value after the evidence appears.

Prediction-free research funding does not attempt to predict which idea will change the world. It ensures that when an unexpected discovery does emerge, the people, tools, and prior contributions that made it possible can still receive support.

That shift—from funding promises alone to funding both capacity and demonstrated contribution—could make science more open to surprise, more tolerant of productive failure, and better able to recognize discoveries that were invisible at the proposal stage.