AI Meritocracy in Research Funding: A Fairer Way to Finance Science

Getting your Trinity Audio player ready...

Research funding determines which questions scientists can investigate, which laboratories survive, and which discoveries reach society. Yet the conventional grant system often allocates money through a small number of committees evaluating lengthy proposals before the proposed research has produced results.

An AI meritocracy in research funding offers a different model. Instead of relying primarily on institutional prestige, persuasive grant writing, or the judgment of a temporary review panel, an AI-assisted system could evaluate researchers through their verifiable contributions to science.

The goal is not to let an opaque algorithm control science. It is to build an auditable funding infrastructure that continuously examines scientific outputs, dependencies, validation, reuse, and demonstrated usefulness—and then distributes funding according to transparent rules.

The AI Internet-Meritocracy project, or AIIM, is a proposed implementation of this idea.

What Is AI Meritocracy in Research Funding?

AI meritocracy is a funding model in which artificial intelligence helps identify and reward scientific merit using evidence distributed across the open scientific record.

Relevant evidence could include:

• research papers and monographs;
• datasets and scientific software;
• mathematical proofs and formal verifications;
• experimental replications;
• corrections of published errors;
• citations and documented reuse;
• contributions to other researchers’ work;
• peer assessments and post-publication reviews;
• practical scientific or technological applications.

A conventional grant committee normally asks:

Which proposed project appears most promising?

An AI meritocratic system can also ask:

Which previous contributions have proved useful, foundational, reproducible, or necessary for later work?

This change matters because research proposals are predictions. Scientific contributions are evidence.

AIIM therefore aims to fund science and free and open-source software according to measurable contribution rather than institutional gatekeeping. Its proposed software would analyze publicly available evidence and allocate donated resources according to the evaluated impact of researchers and developers.

Why Traditional Research Funding Needs Reform

Peer review remains an important part of science, but grant peer review is not a precise measurement instrument.

A broad review of the evidence surrounding biomedical grant review found continuing uncertainty about its reliability, effectiveness, burden, and ability to select the best proposals. Another analysis concluded that research-funding peer review faces problems serious enough that alternative allocation mechanisms deserve testing.

1. Proposal quality is not the same as scientific value

Grant competitions reward the ability to:

• anticipate reviewers’ preferences;
• present a persuasive narrative;
• satisfy formal application requirements;
• demonstrate institutional support;
• promise outcomes within a predetermined period.

These skills may be useful, but they are not identical to discovering something true or building something scientifically valuable.

A researcher can write an excellent proposal for an unproductive project. Another researcher can communicate imperfectly while developing an important result.

2. Conservative proposals can be safer for reviewers

Review panels are accountable for the projects they approve. This can encourage them to prefer proposals that are familiar, incremental, and easy to defend.

Evidence concerning grant review suggests that innovative, risky, or interdisciplinary research may be disadvantaged, while applicants can feel pressure to write conservative proposals acceptable to every panel member.

This creates a structural paradox: research funding is intended to support discovery, but the selection mechanism may penalize ideas that look too different before they are proved.

3. Reviewer scores have limited predictive power

NIH grant percentile scores have been found to be poor discriminators of the later productivity of funded projects, particularly when many applications are concentrated within a narrow scoring range.

This does not mean that peer review has no value. It means that fine-grained rankings produced by committees should not be treated as objective measurements of future scientific impact.

4. Funding decisions are expensive

Researchers spend substantial time preparing applications that will never be funded. Reviewers, administrators, and institutions then spend additional time evaluating and managing those applications.

The result is a large administrative competition surrounding science rather than scientific work itself.

Experiments with distributed peer review indicate that alternative evaluation methods may reduce decision time while broadening participation in funding decisions.

5. Institutional reputation can reinforce itself

Established researchers and universities typically have:

• professional grant-support departments;
• experienced collaborators;
• preliminary data funded by earlier grants;
• stronger publication records;
• recognized institutional names;
• more opportunities to participate in review panels.

Success therefore produces the conditions for further success. This Matthew effect does not prove that established researchers lack merit, but it makes it harder to determine how much of their advantage comes from superior work and how much comes from accumulated institutional position.

Independent researchers, unconventional scholars, and scientists outside prominent universities face the opposite cycle. Without funding, they may lack the resources needed to generate the evidence required to obtain funding.

How an AI Meritocratic Funding System Could Work

A credible system should not generate a single mysterious “merit score.” It should produce an inspectable network of claims, evidence, dependencies, and evaluations.

Step 1: Build an open contribution graph

The system identifies scientific objects and relationships among them:

• Researcher A created dataset B.
• Researcher C used dataset B in paper D.
• Team E reproduced the result in experiment F.
• Software G implemented method H.
• Result I corrected a flaw in paper J.
• Project K depends on theorem L.

This forms a graph of scientific contribution rather than a flat list of publications.

Step 2: Evaluate multiple dimensions of merit

Scientific merit cannot be reduced to citation counts. A more robust model could examine several dimensions:

Dimension	Possible evidence
Originality	Novel concepts, methods, conjectures or datasets
Correctness	Replication, formal verification and expert review
Usefulness	Reuse in research, software, policy or technology
Foundational importance	Number and value of dependent contributions
Reproducibility	Open methods, data and successful replications
Corrective value	Errors detected, negative results or failed claims resolved
Openness	Public availability of outputs and supporting evidence
Community assessment	Transparent, reputation-weighted evaluations

The weighting rules should be public and contestable.

Step 3: Trace scientific dependencies

Dependency analysis may be one of the most important differences between AI meritocracy and ordinary bibliometrics.

A highly visible final result may depend on:

• an obscure theorem;
• a maintained software library;
• a carefully prepared dataset;
• a failed experiment that eliminated an incorrect hypothesis;
• a technical correction;
• years of foundational work.

Citation counts do not reliably divide credit among these contributions. An AI system could examine how later work depends on earlier work and allocate part of the resulting value backward through the dependency graph.

Step 4: Allocate funding continuously

Traditional grants usually make a large decision at one moment: approve or reject.

An AI meritocratic platform could distribute smaller payments continuously as new evidence appears. A contribution might receive additional support when it is:

• independently validated;
• reused by another project;
• incorporated into important software;
• shown to solve a practical problem;
• recognized as foundational to subsequent discoveries.

This would make scientific funding more adaptive. It would also allow useful work to receive support even when its importance was not obvious at publication time.

Step 5: Preserve human governance and appeals

AI should assist evaluation, not become an unquestionable scientific authority.

Researchers need mechanisms to:

• inspect the evidence used in an assessment;
• challenge incorrect attribution;
• declare conflicts of interest;
• report manipulation;
• appeal an automated decision;
• propose changes to evaluation rules;
• compare alternative scoring models.

The system’s software, rules, model versions, and major governance decisions should be auditable.

AI Meritocracy Is Not Citation-Based Funding

A simplistic system that pays researchers according to citations would reproduce many existing distortions.

Citation counts can be affected by:

• discipline size;
• publication age;
• fashionable topics;
• self-citation;
• citation cartels;
• negative citations;
• database coverage;
• prestige and visibility;
• differences between the sciences and humanities.

Research comparing bibliometric evaluation with peer assessment has found that bibliometric measures can introduce field-dependent distortions and should not be assumed to measure research quality directly.

AI meritocracy must therefore distinguish among different kinds of relationships. A citation for routine background information should not carry the same weight as documented dependence on a theorem, dataset, method, or software component.

Could Researchers Manipulate the System?

Yes. Any funding mechanism creates incentives, and participants will respond to them.

Possible attacks include:

• fabricated citations;
• reciprocal endorsements;
• paper splitting;
• fake reuse claims;
• coordinated voting;
• plagiarism;
• synthetic identities;
• low-value automated publications;
• manipulation of public metrics.

This is why AI meritocracy requires more than a language model reading papers. It needs adversarial detection, identity verification, provenance records, conflict-of-interest controls, independent audits, and penalties for fraudulent claims.

Blockchain infrastructure may help preserve transaction history, governance decisions, attribution records, and funding flows. However, blockchain does not determine scientific truth. It can secure records; it cannot replace scientific evaluation.

The Role of Peer Review in an AI Meritocracy

AI meritocracy should not eliminate expert judgment. It should change where and how expert judgment is used.

Traditional grant review concentrates expert judgment into a private, temporary decision made before the work is completed. A meritocratic system could use expert evaluation throughout the scientific lifecycle:

1. Experts assess whether a contribution is technically sound.
2. Independent researchers attempt replication.
3. Users document whether the contribution helped their work.
4. Reviewers challenge attribution and dependency claims.
5. AI aggregates the resulting evidence.
6. Governance mechanisms resolve contested cases.

This is closer to continuous post-publication evaluation than to one-time proposal review.

The system could also combine several funding mechanisms:

• prospective grants for equipment-intensive research;
• milestone payments for planned projects;
• retroactive rewards for demonstrated contributions;
• lotteries among equally qualified high-risk proposals;
• emergency funding for urgent scientific work;
• public donations directed through transparent allocation rules.

AI meritocracy is therefore best understood as a new layer of scientific financial infrastructure, not as a universal replacement for every grant.

Benefits for Independent and Early-Career Researchers

Independent researchers often lack formal credentials, institutional affiliations, and professional grant-writing support. Early-career researchers may have strong ideas but little accumulated prestige.

A contribution-based system could lower these barriers by asking what a person has actually produced.

This does not mean ignoring expertise or research history. It means allowing verifiable work to establish credibility even when the researcher is outside the conventional academic hierarchy.

Such a system could especially help:

• independent scientists;
• researchers in underfunded countries;
• interdisciplinary scholars;
• maintainers of scientific software;
• authors of negative or corrective results;
• researchers working on unfashionable foundational problems;
• contributors whose work is valuable but difficult to publish conventionally.

For a broader discussion of this problem, see funding for independent scientists.

Why AIIM Uses Donations

AIIM proposes directing donated resources toward scientists and open-source developers using an automated merit-allocation system. Its donor model is intended to let supporters fund an ecosystem of valuable work without manually selecting individual grant recipients.

This addresses a practical donor problem.

A donor may want to support science but may not know:

• which research field is most neglected;
• which scientists are credible;
• which projects depend on other projects;
• how much each contribution deserves;
• whether an institution uses funds efficiently.

An auditable AI system could act as an allocation layer between donors and researchers. Instead of replacing donor choice, it could give donors several transparent strategies—for example, supporting foundational mathematics, open medical research, scientific software, replications, or independent scholars.

Those interested in supporting the development of this infrastructure can visit the Science DAO donation page.

Conditions for a Legitimate AI Funding System

An AI meritocracy will be credible only if it satisfies strict institutional requirements.

Transparency

Researchers should know which evidence affected their evaluation and how allocation rules work.

Auditability

Independent specialists should be able to examine the software, datasets, models, and financial records.

Explainability

The system should provide concrete reasons for allocations rather than merely presenting scores.

Pluralism

No single metric or model should define scientific merit. Multiple evaluation methods should coexist and be compared.

Due process

Researchers need correction, objection, and appeal procedures.

Privacy

Sensitive personal information should not become a hidden input into funding decisions.

Decentralized oversight

No government, corporation, university, donor, or model provider should have unilateral control over the entire system.

Experimental deployment

AI meritocracy should first be tested with limited funds, published evaluation criteria, independent monitoring, and direct comparison against conventional allocation methods.

From Grant Competition to Scientific Contribution

The central problem of research funding is not simply that some committees make poor decisions. It is that the financial system surrounding science often rewards predictions, presentation, affiliation, and administrative success more directly than verified contribution.

AI creates an opportunity to redesign this system.

Used badly, AI could automate existing prejudice, produce unaccountable scores, and centralize control over science. Used well, it could help trace contributions across enormous bodies of literature, identify hidden dependencies, detect neglected work, and distribute resources more continuously.

The decisive question is therefore not:

Should AI replace scientists in funding decisions?

It is:

Can scientists, donors, and public institutions build an auditable system that uses AI to make scientific merit more visible?

AI Internet-Meritocracy proposes such a system: funding research according to demonstrated usefulness and contribution rather than relying exclusively on institutional gatekeepers.

It remains an experimental proposal, and its methods will require validation, governance, and resistance to manipulation. Nevertheless, the underlying principle is compelling: the people who create verifiable value for science should have a systematic path to receiving the resources needed to continue their work.