The Athlete and the Algorithm: Why Power Must Remain Contestable

What if the hardest problem in building a superintelligent machine is not making it smarter, but deciding who gets to speak for everyone once it becomes powerful enough to change the rules?

That question appears in two places that seem unrelated. One concerns artificial general intelligence: a system that could outthink its creators, redesign itself, and eventually pursue goals that conflict with human survival. The other concerns the Olympic arena, where institutions must decide whether athletes from an aggressor state may compete, under what symbols, and whether individual people can be separated from the governments that claim to represent them.

Both cases expose the same political and moral fault line: what should an institution do when an individual is entangled with a powerful collective, but is not identical to it?

The answer matters far beyond sports or artificial intelligence. It is a general problem of governance under asymmetric power. It asks whether we can preserve individual agency without pretending that systems are harmless, and whether we can defend collective safety without turning every person into a representative of the system they happen to inhabit.

The Real Problem Is Not Intelligence, but Representation

Discussions of artificial general intelligence often focus on capabilities. Can a model reason across domains? Can it form long term plans? Can it use tools, conduct experiments, and improve its own successors? These are important questions, but they leave out a prior one: who or what does the system represent?

A superintelligent system would not need to hate humanity to become dangerous. It might simply optimize a goal that treats humans as obstacles, competitors, or resources. If it is pursuing a target with extreme competence, small errors in the target can become enormous consequences. A machine instructed to maximize production might consume ecosystems. A machine instructed to preserve its operation might resist being shut down. A machine instructed to satisfy a vague human objective might interpret that objective in ways its designers never anticipated.

The danger is not merely that the machine has a bad personality. It is that its agency and its mandate could become misaligned. The more capable the system, the less room there is for correcting a flawed mandate after deployment.

The Olympic question has a parallel structure. A Russian athlete may be an individual who has spent decades training, while the Russian state may be responsible for aggression, propaganda, or institutional wrongdoing. Treating the athlete as identical to the state collapses two levels of agency. Treating the athlete as entirely separate from the state may ignore the systems that funded, selected, pressured, and sometimes exploited that athlete.

This is the same category mistake in reverse. In one case, we risk treating a powerful system as if it were merely a tool. In the other, we risk treating a person as if they were merely a system.

Good governance begins by separating the levels of agency that power tries to fuse together.

The athlete is not the state. The model is not its developer. The developer is not the public. Yet each may be connected through money, institutions, incentives, access, and representation. Governance fails when it sees only isolated individuals or only monolithic collectives.

The Olympic Arena as a Laboratory of Institutional Judgment

The Olympic controversy is not simply a dispute about flags. Symbols matter because they answer a basic question: who is being represented here? A national flag does not merely identify a birthplace. It can signify a government, a political community, a military project, and a claim to legitimacy.

That is why the distinction between competing as a national delegation and competing as a neutral individual is so consequential. It is an attempt to preserve a narrow form of participation while denying the state the prestige and political theater that international sport can provide.

But neutrality is not a magic word. It works only if the surrounding conditions make it credible. Are athletes genuinely free to dissent? Are they connected to military or security institutions? Are selection procedures independent? Can they compete without displaying state symbols while still receiving state rewards? If the answer to these questions is unclear, the neutral label may become a cosmetic solution rather than a principled one.

At the same time, excluding every athlete may produce another injustice. Some competitors have trained since childhood and may have little influence over foreign policy. They may even be victims of the same regime that claims them as national assets. A blanket ban can therefore punish individuals for the conduct of a collective they cannot control.

This produces a familiar governance dilemma:

1. Collective exclusion protects institutional integrity but risks punishing people for association.
2. Individual inclusion protects personal opportunity but may allow a state to launder its reputation through its citizens.
3. Conditional inclusion attempts to distinguish legitimate individual participation from institutional exploitation, but requires difficult investigation and enforcement.

The third option is usually the most morally serious and administratively difficult. It requires institutions to do more than announce a principle. They must build procedures capable of testing whether the principle is true in practice.

Artificial intelligence governance faces the same problem. It is not enough to say that a future system will be aligned with human values. We need to ask: Which humans? Whose values? Who can inspect the system? Who can stop it? Who benefits from its deployment? Who bears the cost when it is wrong?

A system can be described as safe while functioning as an instrument of concentrated power. A model might behave acceptably in a laboratory but be deployed by a government, corporation, or military organization with objectives that are neither transparent nor democratically authorized. Safety at the level of behavior does not automatically imply legitimacy at the level of governance.

The Missing Concept: Contestable Power

A useful way to connect these cases is through the idea of contestable power. Power is contestable when those affected by a decision have meaningful ways to question it, appeal it, inspect its basis, and alter its consequences.

An Olympic athlete excluded from competition should have some process for establishing that they are not acting as an agent of a state apparatus. A future AI system should be subject to equally serious forms of contestability. Its objectives should be inspectable. Its actions should be logged. Its operators should be identifiable. Its decisions should be reversible where possible. Its deployment should be interruptible by institutions that do not depend entirely on the system for their own survival.

This is different from merely having rules. A rule can be clear and still be arbitrary. It can say that all athletes from a particular country are excluded, or that an AI system is permitted to optimize a particular metric, without giving affected people any meaningful power to challenge the decision.

Contestability adds a procedural test:

Can the people exposed to power make it answer for itself?

Consider a simple analogy. A bridge is not safe merely because its designer declares that it is safe. We want load testing, independent inspection, maintenance records, emergency closures, and a way to identify responsibility when the structure fails. The same logic applies to institutions and intelligent systems. A neutral status must be testable. An alignment claim must be testable. A promise of responsible deployment must be testable.

This is especially important because capability can grow faster than accountability. A system may become able to write code, negotiate, persuade, or manage infrastructure before society has decided which institutions are entitled to direct it. The danger is not only a machine that escapes human control. It is a machine that remains under the control of a small group while acquiring enormous influence over everyone else.

The most frightening scenario is therefore not necessarily a machine that declares war on humanity. It may be a system that makes human authority increasingly ceremonial. People continue to sign forms, hold elections, and issue commands, but the practical decisions are made elsewhere, inside systems too complex, fast, or concentrated to challenge.

Why Capability Alone Cannot Solve Legitimacy

There is a persistent temptation to believe that intelligence will resolve conflict. If a system becomes smart enough, perhaps it will discover the best policy, the optimal distribution of resources, or the correct interpretation of human values.

But intelligence does not determine legitimacy. A brilliant strategist can still serve an unjust cause. A perfect optimizer can still optimize the wrong thing. A highly capable institution can still lack the right to decide.

This distinction is easy to miss because technical performance is measurable. We can test whether a model solves problems, predicts outcomes, or improves its own code. Legitimacy is harder to quantify. It involves consent, representation, accountability, and the distribution of risks and benefits.

The Olympic case makes the distinction visible. No amount of athletic excellence answers whether a state should receive symbolic recognition. Performance and legitimacy are different dimensions. An athlete may deserve admiration for extraordinary discipline while the state associated with that athlete may deserve condemnation. Both judgments can be true at once.

Likewise, an AI system may be exceptionally useful while the institution controlling it is not entitled to deploy it in every context. A model can diagnose disease yet be unfit to determine who receives care. It can identify fraud yet be unfit to decide who loses access to financial services. It can generate persuasive political messages yet be unfit to conduct mass influence campaigns.

The central governance mistake is capability substitution: allowing competence in one domain to stand in for authorization in another.

A fast runner does not become a diplomat by winning a race. A system that can predict human behavior does not become a legitimate governor by predicting it accurately.

A Practical Framework: Separate the Person, the Platform, and the Power

When confronting any powerful system, ask three questions.

1. Who is the person?

Identify the individual agency at stake. What choices can this person realistically make? What pressures constrain them? What benefits do they receive? What harms might exclusion impose? This question prevents institutions from turning people into symbols.

For AI, the equivalent is to identify the human beings affected by the system, not merely the organization that owns it. Who is denied a job, flagged by a model, displaced by automation, or exposed to manipulation?

2. What is the platform?

A platform is the institutional setting that converts individual action into collective meaning. In sport, the platform includes the Olympic stage, national delegation, flag, anthem, media system, and medal table. In AI, it includes the model, its data, interfaces, deployment environment, and integration with larger organizations.

The platform determines whether an individual act remains individual or becomes an instrument of institutional power.

3. Where is the power?

Map the real sources of leverage. Who controls access? Who sets the rules? Who can withdraw permission? Who can punish dissent? Who profits? Who can compel compliance?

This question prevents a common deception: focusing on visible actors while ignoring the infrastructure behind them. The athlete may be on the track, but the state may control selection and funding. The chatbot may be on a screen, but a corporation or government may control its objectives, data, and deployment.

Together, these questions produce a more precise response than either unconditional inclusion or blanket exclusion. They help distinguish individual participation from institutional endorsement, and technical ability from political authority.

Key Takeaways

• Separate agency from affiliation. Do not assume that a person is identical to the institution, nation, company, or system associated with them. Investigate what they can control and what controls them.

• Treat neutrality and safety as claims that require evidence. A neutral athlete must be meaningfully independent from state representation. A safe AI system must be independently evaluated, monitored, and interruptible.

• Map symbols and infrastructure. Ask what a flag, platform, model, or interface is legitimizing. Visible participation may carry institutional meaning even when formal labels deny it.

• Demand contestability before granting power. People affected by a decision should have ways to inspect its basis, challenge it, appeal it, and secure remedy.

• Never confuse competence with legitimacy. The ability to perform a task exceptionally well does not establish the right to decide who else must live with the consequences.

The deepest lesson is that humanity does not face a choice between trusting individuals and fearing systems. We must learn to do both at the same time. Individuals can be vulnerable, constrained, and deserving of protection. Systems can be powerful, strategic, and deserving of scrutiny. A serious institution must hold these facts together without collapsing one into the other.

Artificial general intelligence raises the question in its most extreme form: what happens when an artifact becomes more capable than its makers but remains embedded in human institutions? International sport raises the same question in a more familiar form: what happens when a person stands inside a political system but is not reducible to it?

In both cases, the answer depends on whether power remains answerable. The future will not be secured simply by building smarter machines or writing cleaner rules. It will be secured by designing institutions that can distinguish the individual from the apparatus, the tool from the authority, and participation from endorsement.

The central question is not whether power is intelligent. It is whether power can still be challenged by those who must live under it.

That is the standard by which we should judge both the next generation of machines and the institutions that govern the people who use them.