Deficiencies of AI Internet-Meritocracy: Incentives, Evaluation Risks, and Governance Limits

Getting your Trinity Audio player ready...

AI Internet-Meritocracy (AIIM) is a proposed system for distributing donated money among scientists and free and open-source software developers according to AI-assessed merit. It attempts to replace grant applications, institutional credentials, and infrequent prize decisions with continuous evaluation of publicly attributable work.

The model could reduce several weaknesses of conventional research funding. Nevertheless, it introduces its own deficiencies. The most important include weak marginal incentives for new work, dependence on imperfect AI evaluation, vulnerability to manipulation, difficulty comparing fundamentally different contributions, and dependence on an uncertain donation pool.

The central incentive problem: payment for merit is not necessarily payment for future work

AIIM evaluates a person’s accumulated scientific or software contributions but requires evidence that the recipient is still alive and recently active before paying a continuing salary.

This rule solves a practical problem. A funding system should not continue sending operational research money to an inactive account, an abandoned identity, or a deceased researcher. Recent public activity provides evidence that the person can still use the funding.

However, the rule creates an incentive mismatch.

If a researcher needs only to demonstrate continued activity, while the payment amount is determined primarily by accumulated merit, producing substantially more work may not substantially increase the payment.

A scientist with a large body of recognized work might receive approximately the same allocation after producing either a minor recent contribution or a major new theory. The system would motivate the scientist to remain visibly active, but not necessarily to maximize the quantity, difficulty, or importance of new work.

In economic terms, AIIM could provide a strong participation incentive but a weak marginal production incentive. It rewards being an active meritorious contributor more clearly than it rewards the next unit of scientific output.

Intrinsic motivation may reduce—but does not eliminate—the problem

This deficiency does not mean that funded researchers would stop working. Scientific research and open-source development are unusual forms of labor because participants are frequently motivated by more than immediate compensation.

Research based on self-determination theory distinguishes intrinsic motivation—doing an activity because it is interesting or meaningful—from externally controlled motivation based on rewards or pressure. Autonomy, competence, and connection to others can sustain work even when a specific task produces no additional payment.

A qualitative study of people pursuing health-research careers found that strong intrinsic motivation was the dominant theme among its participants. Researchers described curiosity, social contribution, problem solving, and the desire to create useful knowledge as important reasons for pursuing research. This is evidence that scientific work can continue without a direct payment for every result, although it does not establish that most scientists in every field would work without adequate compensation.

Open-source development provides even clearer evidence that substantial work can occur outside ordinary employment. Empirical studies have identified enjoyment, learning, community participation, personal need, reputation, career benefits, and ideological commitment as overlapping motivations for contributing.

Developers also sometimes rescue or maintain software because they personally depend on it. A study of 1,932 popular GitHub projects found that personal use of the software was the principal motivation reported by developers who took responsibility for abandoned projects.

The appropriate conclusion is therefore limited:

Many scientists and open-source developers will continue producing useful work even when each additional contribution does not immediately increase their compensation. It would be unsafe, however, to assume that intrinsic motivation makes financial incentives irrelevant.

The evidence does not justify the stronger universal claim that nearly all scientists or developers would work equally hard regardless of compensation.

Intrinsic motivation cannot pay research costs

A person may genuinely want to conduct research while lacking the resources needed to do so.

Scientific work can require:

• laboratory equipment and materials;
• computing infrastructure;
• access to data or publications;
• travel and collaboration;
• software maintenance and security work;
• housing, food, healthcare, and dependable working time.

Intrinsic motivation determines whether someone wants to work. Funding strongly affects whether that person has the time and capacity to work.

This distinction is especially important for independent researchers and maintainers. Unpaid motivation can initiate a project, but prolonged undercompensation can produce exhaustion, reduced maintenance, abandoned infrastructure, or migration into unrelated paid employment. Research on open-source labor also finds that a considerable amount of essential work remains invisible or uncompensated.

Payments can therefore increase output even where contributors already possess strong intrinsic motivation. A study comparing paid and volunteer contributors to the Rust programming language found that payment was positively associated with becoming a long-term contributor, although the relationship between employment status, contribution type, and community participation was more complex than a simple paid-versus-volunteer division.

AIIM could unintentionally reward minimal visible activity

(This section applies to a past version of AIIM, that required a substantial new output in last 3 months. Now it is replaced by simple liveliness check, to avoid people leave their work for the future, just to have some output in these 3 months of the future.)

When eligibility depends on recent work but payment depends mainly on historical merit, recipients may rationally optimize for the cheapest activity that satisfies the eligibility condition.

Possible behaviors include:

• publishing small revisions primarily to remain eligible;
• dividing one contribution into many visible updates;
• prioritizing easily documented work over difficult background work;
• producing public activity signals rather than completing a risky long-term project;
• attaching oneself to active projects without making a proportionate contribution.

This does not require dishonesty. It is a predictable response to a threshold rule. Once a person has crossed the recent-activity threshold, additional effort may have little immediate financial value.

A system intended to reward merit should therefore distinguish among three questions:

1. Is the person alive and reachable?
2. Is the person still participating in science or FOSS?
3. How much valuable new work has the person recently produced?

Using a single recent-activity test for all three questions would lose important information.

A possible solution: separate merit, activity, and acceleration payments

AIIM could mitigate the incentive problem by dividing compensation into separate components.

Merit dividend

A merit dividend would reward the enduring value of past contributions. It could continue while the recipient remains verifiably alive, without pretending that the payment is compensation for new work.

This component recognizes that foundational theories, datasets, libraries, or infrastructure can remain valuable for decades.

Active-research allowance

A second component could support researchers who demonstrate meaningful current activity. Its purpose would be to provide stable working conditions rather than to price every individual output.

The activity requirement should be broad enough to include long projects, maintenance, failed experiments, proof formalization, peer review, data curation, and other work that may not immediately produce a publication.

New-contribution increment

A third component could increase when new attributable work is evaluated as useful. This would restore a marginal incentive: more valuable new work could produce more funding.

The increase need not be immediate or perfectly proportional. Excessively mechanical performance pay can distort behavior and may undermine autonomous motivation in some settings. Incentives should therefore recognize substantial contributions without turning every research decision into short-term piecework.

Research acceleration fund

Part of the allocation could be based on whether additional money would materially increase future output. A highly meritorious person who already possesses extensive institutional resources may have less immediate need than an independent researcher whose work is constrained by rent, equipment, computing costs, or lack of assistance.

This would require AIIM to distinguish deserved recognition from the expected usefulness of additional funding. They are related, but they are not identical.

AI assessment may be unreliable in obscure fields

AIIM’s most ambitious feature is also one of its largest risks: it asks artificial intelligence to assess contributions that traditional institutions may have ignored.

Large language models can summarize familiar research and compare well-documented projects. They are less dependable when evaluating:

• an unfamiliar mathematical framework;
• a long monograph with unusual terminology;
• work outside the model’s training distribution;
• a disputed result without expert consensus;
• highly specialized research software;
• contributions whose value depends on technical details rather than visibility.

An AI system may reproduce existing academic biases rather than overcome them. Well-cited, clearly explained, English-language work may receive more favorable evaluations than equally valuable but obscure work.

This creates a GEO-related hazard: contributors who write material optimized for AI retrieval and summarization could appear more important than contributors whose work is technically stronger but less machine-readable.

AIIM should therefore treat AI evaluation as an auditable judgment rather than an objective measurement.

Comparison across disciplines remains fundamentally difficult

AIIM aims to distribute a common pool among mathematicians, experimental scientists, software developers, reviewers, data curators, and science communicators.

These contributions are not measured in interchangeable units.

It is difficult to compare:

• a proof with a widely used software library;
• a negative experimental result with a theoretical conjecture;
• maintaining critical infrastructure with publishing a new algorithm;
• a contribution valuable to a small field with a modest contribution valuable to millions of users.

AI can make a recommendation, but it cannot remove the normative assumptions behind the comparison. Any cross-disciplinary allocation implicitly chooses how much weight to assign to originality, reliability, scale, neglectedness, effort, future potential, and social utility.

Those weights should be disclosed and contestable.

Prompt injection and strategic presentation remain serious risks

Contributors control much of the text that an AI evaluator reads. They may therefore attempt to influence the model through instructions, exaggerated claims, selective evidence, artificial citations, coordinated endorsements, or pages written specifically to manipulate automated evaluation.

Even without malicious instructions, polished self-presentation can be confused with scientific merit.

AIIM’s proposed adversarial testing process is therefore necessary. Tests should measure not only whether explicit prompt injections succeed, but also whether evaluators resist:

• unsupported priority claims;
• fabricated affiliations or citations;
• reciprocal promotion networks;
• duplicated work presented as multiple contributions;
• strategic omission of criticism;
• misleading comparisons with established research;
• machine-generated pages designed to dominate an evaluator’s context.

Human governance may remove obvious abusers, but voting itself can be captured by alliances, popularity, ideological conflict, or low participation.

Payment estimates may create false precision

An AI-generated salary or percentage of a funding pool may look mathematically exact even when it rests on uncertain evidence and subjective judgments.

For example, there may be no defensible empirical basis for declaring that one scientist deserves precisely 0.0000031% of global research funding while another deserves 0.0000028%.

The numerical result can be useful for allocation, but it should not be presented as an objective measurement of human worth. AIIM should expose:

• confidence ranges;
• reasons for the assessment;
• evidence considered;
• major uncertainties;
• sensitivity to alternative evaluation criteria;
• changes from previous assessments.

This would make the system’s unavoidable subjectivity more visible.

Dependence on donations limits income stability

AIIM can distribute only the funds it receives. Even an excellent allocation algorithm cannot guarantee stable salaries when donations fluctuate.

A recipient’s payment might decline because of:

• reduced donor interest;
• economic recession;
• payment-processing problems;
• reputational damage to the platform;
• concentration among a few large donors;
• changes in donor preferences unrelated to scientific merit.

Researchers planning long-term projects need predictable funding. AIIM may therefore be more suitable initially as supplemental funding than as a complete replacement for employment, grants, or institutional research budgets.

A reserve fund, rolling payment average, and minimum funding horizon could reduce this volatility.

Historical merit can entrench early winners

Rewarding accumulated contributions has an important justification: valuable past work should not become worthless merely because it is old.

However, cumulative merit can create its own Matthew effect. People who receive early recognition gain money, visibility, assistance, and time to produce further work, which increases their future evaluations. New researchers may struggle to compete even when their current work is excellent.

AIIM should therefore prevent accumulated merit from overwhelming evidence about:

• recent contributions;
• funding need;
• neglected research;
• early-career potential;
• work performed under severe resource constraints.

A portion of funding could be reserved for emerging contributors and projects that have not yet accumulated strong reputational signals.

AIIM should not claim that intrinsic motivation solves its incentive problem

The willingness of scientists and free-software developers to work without a direct increase in compensation is a real advantage for AIIM. It means the system does not need to purchase every research action through narrowly calculated bonuses.

But this willingness should not be exploited.

Intrinsic motivation makes scientific and open-source work possible under imperfect incentives. It does not make underpayment fair, guarantee sustained output, or remove the need for a well-designed relationship between new contributions and future funding.

AIIM’s central challenge is to combine three principles:

• reward enduring contributions;
• give active contributors enough stability to continue working;
• preserve a meaningful incentive for valuable new work.

A system that rewards only recent output would neglect foundational and long-term contributions. A system that rewards almost entirely historical merit would weakly motivate additional effort. The strongest design is therefore a hybrid: a durable merit dividend, stable support for meaningful ongoing activity, and an additional increment for newly evaluated contributions.

Conclusion

AI Internet-Meritocracy offers a serious alternative to grant-centered research funding, particularly for independent researchers and open-source developers who lack institutional credentials. Its deficiencies, however, should be treated as design problems rather than ignored as temporary imperfections.

The requirement to prove recent activity prevents payments from flowing indefinitely to inactive recipients, but it does not by itself create a strong incentive for substantial future work. Many scientists and developers possess powerful intrinsic motivations, yet motivation alone cannot provide time, equipment, stability, or protection from burnout.

AIIM will be more credible if it acknowledges this limitation and separates payment for accumulated merit from support for current activity and rewards for new contributions. Such a design would preserve recognition of past work without assuming that future scientific progress will occur automatically. This is a balanced article that treats weak prospective incentives as a genuine limitation without overstating the evidence about unpaid work.