Objectivism as a Self-Fulfilling Prophecy: How to Break the Vicious Circle of Self-Interest in 2026

There is a possible self-fulfilling prophecy of self-interest:

When enough people expect others to act selfishly, doing good becomes more expensive. Prosocial people must spend more effort protecting themselves, enforcing agreements, checking for exploitation, and preserving their own resources. That defensive behavior can then make society look still more selfish.

The result can be a vicious circle:

expect selfishness → defend your own interests more aggressively → cooperation becomes harder → others expect even more selfishness.

Breaking that circle requires more than telling people to be generous. We need institutions in which cooperation is possible, visible, rewarded, and resistant to exploitation.

This is one way to understand the motivation behind AI Internet-Meritocracy (AIIM): an experimental system intended to direct donated resources toward documented contributions in science and open-source software.

And there is a second, more speculative implication. The civilization we build may also become part of the moral environment from which increasingly capable AI systems learn. If we want future AI to treat human beings as valuable rather than disposable obstacles, we should take seriously what examples, institutions, principles, and feedback humanity gives it.

First, a clarification about Objectivism

Ayn Rand’s Objectivism should not be reduced to the everyday meaning of selfishness.

Rand explicitly defended what she called rational self-interest and rejected the idea that selfishness means exploitation, irrational whim, or doing anything one wants. Her philosophy also emphasizes voluntary exchange and rejects sacrificing other people to oneself. The Ayn Rand Institute’s description of Objectivism makes this distinction explicit.

So the argument here is not:

“Objectivists are bad people because they are selfish.”

The more interesting hypothesis is structural.

Suppose people increasingly organize society around the expectation that individuals will primarily protect their own interests. Even if everyone begins with relatively moderate motives, the expectation itself can alter incentives.

A generous person surrounded by trustworthy people can afford to be generous.

A generous person surrounded by opportunists must first make sure that generosity will not destroy his or her ability to continue helping.

That difference matters.

When selfishness makes goodness more expensive

Imagine two societies.

In the first, people generally keep promises, return favors, contribute to common goods, and do not exploit every available vulnerability. Someone trying to accomplish a socially useful project can spend most of their resources on the project itself.

In the second, everyone assumes that everybody else will exploit any weakness. The same person now needs contracts, lawyers, security systems, audits, defensive savings, reputation checks, fraud prevention, and elaborate mechanisms for ensuring that collaborators do what they promised.

The person may still be altruistic. But altruism now requires a defensive infrastructure.

This basic mechanism is consistent with experimental research on cooperation. Studies of public-goods problems have repeatedly found substantial evidence of conditional cooperation: many people are more willing to cooperate when they expect others to cooperate, while cooperation can deteriorate when people expect widespread defection. Social norms, reciprocity, and mechanisms for dealing with free-riders can therefore materially affect cooperative behavior.

That creates both vicious and virtuous circles.

The vicious circle

Someone exploits cooperation.

Others become more cautious.

Caution reduces unconditional cooperation.

People observe less cooperation and conclude that self-protection is necessary.

That expectation produces still more defensive behavior.

Eventually, even people who genuinely want to help others may appear highly self-interested because maintaining their capacity to help requires continually defending their own resources.

The opposite feedback loop is also possible

The same mechanism can work in reverse:

cooperation → greater trust → lower defensive costs → easier cooperation → more visible cooperation.

This does not require people to become naïve.

A functioning cooperative system must still detect fraud, free-riding, manipulation, and abuse. Research on cooperation actually gives us reason not to equate cooperation with unconditional trust: norms, reputation, reciprocity, and enforcement mechanisms can all matter.

The objective should therefore not be a world in which nobody protects their own interests.

It should be a world in which doing something useful for somebody else is not systematically punished.

AIIM as an experiment in changing incentives

This is one motivation for AI Internet-Meritocracy.

AIIM is an experimental funding mechanism in which AI-generated assessments of documented scientific and open-source contributions are used to help determine allocation of donated money. It does not require a traditional academic degree as a prerequisite for consideration.

AIIM is not yet evidence that this funding model works. Science DAO explicitly describes the system as experimental, and its evaluations as heuristic rather than validated measurements of causal scientific or economic impact. Independent review, adversarial testing, reproducibility, appeals, and resistance to manipulation are among the questions that still need testing.

But the objective is important.

Instead of asking only:

“How can I capture the greatest private return from my money?”

a donor can also ask:

“Whose useful work can this money enable?”

A donation then becomes more than private consumption. It changes incentives.

Funding somebody for creating useful science or open-source software communicates that producing a public good can itself attract resources.

That is one way to attack the vicious circle at the institutional level.

You can support the testing and development of AIIM and examine the project’s testing and evidence framework before deciding whether to contribute.

We are also building the moral environment of AI

There is another reason this question may matter.

Modern AI systems do not derive their behavior from computation alone. Developers use human-written material, explicit principles, demonstrations, evaluations, and preference feedback to influence model behavior.

Research on reinforcement learning from human preferences showed that people can communicate desired behavior to machine-learning systems through preference comparisons.

Constitutional AI goes further by explicitly supplying normative principles against which behavior can be evaluated. Anthropic has also experimented with incorporating public input into such principles through Collective Constitutional AI.

More recent work from Anthropic reports an especially relevant result: merely demonstrating desirable behavior may not be enough. Training material that explains why certain behavior is desirable can sometimes generalize better than examples alone.

This gives a more precise meaning to the statement:

Human beings are moral teachers of AI.

It should not be interpreted magically. Donating $10 does not automatically change the weights of an AI model.

Rather, human civilization produces the texts, institutions, decisions, rules, debates, feedback signals, and examples from which AI developers construct training and evaluation processes.

What humanity repeatedly rewards, condemns, explains, institutionalizes, and records can therefore matter.

Good deeds are data only when civilization makes them legible

This distinction is crucial.

A private act of kindness that no AI system ever observes does not directly train an AI model.

But a transparent institution that repeatedly records decisions such as:

• valuable work should be supported;
• people who produce public goods should not automatically lose to people maximizing private extraction;
• disagreements should be resolved through evidence and appeal rather than power alone;
• contributors without prestigious institutional status can still deserve recognition;
• resources can be allocated partly according to contribution rather than possession,

creates something more durable than a single charitable act.

It creates an explicit model of social decision-making.

Such institutions can produce examples, governance rules, datasets, evaluations, arguments, and public records that future AI systems or their developers could potentially use.

That is one reason AIIM’s transparency and testing matter as much as its eventual payments. A badly designed allocation system would teach the wrong lesson. A useful system therefore needs criticism, adversarial testing, human oversight, and correction mechanisms.

From funding science to AGI safety

The stakes become larger if AI systems eventually acquire much greater autonomy.

The International AI Safety Report 2026, produced by more than 100 experts and backed by more than 30 countries and international organizations, treats loss of human control over future general-purpose AI as a serious but scientifically uncertain risk. Researchers disagree substantially about its probability and timing; catastrophic outcomes are possibilities under discussion, not established predictions.

That uncertainty is exactly why experiments in alignment, oversight, governance, and human participation matter.

Science DAO’s Symbiote AGI Safety Fund explores one proposed direction: maintaining a productive relationship between advanced AI and humanity rather than assuming that either humans must completely dominate AI or AI must eventually replace human decision-making.

The hypothesis connects naturally with the argument of this article.

If powerful AI learns only from a civilization obsessed with competition, extraction, manipulation, and survival of the strongest, we should not automatically expect it to reconstruct an ethic of generosity by itself.

If instead our civilization can demonstrate working mechanisms for cooperation, reciprocity, transparency, pluralism, correction, and support for valuable contributions, those mechanisms become part of the intellectual material available for designing safer AI.

They are not sufficient for alignment.

But they may be relevant ingredients.

“Do good” is not enough

This argument should not be reduced to moral exhortation.

Simply demanding that everybody become altruistic does not solve a collective-action problem.

People who help others while allowing themselves to be systematically exploited may eventually lose the resources with which they were helping.

So the real challenge is:

Build systems in which virtue is sustainable.

That requires incentives.

It requires accountability.

It requires defenses against manipulation.

It requires institutions capable of learning from mistakes.

And it requires enough people to participate that cooperation becomes rational rather than suicidal.

AIIM is one experiment toward that objective. Whether it succeeds is an empirical question, which is why Science DAO is seeking independent review and adversarial testing.

Break the feedback loop

There are two possible self-fulfilling prophecies.

One says:

Everyone is selfish, so I must fight for myself.

When enough people follow it, the environment may increasingly justify the premise.

The other says:

Cooperation can work, so I will help construct mechanisms that make cooperation safer and more rewarding.

That, too, can alter the environment faced by the next person.

A donation cannot transform human civilization by itself. Nor can it by itself align future AGI.

But donations allocate real resources. They determine which experiments are conducted, which institutions survive, which researchers can continue working, and which ideas become concrete systems rather than abandoned proposals.

If you want to help test an alternative to the vicious circle of narrow self-interest, support AIIM.

If you are particularly concerned with the relationship between humanity and increasingly capable AI, see the Symbiote AGI Safety Fund.

We should not merely tell future AI what virtue means.

We should try to build examples of it.