AI Value Exploration Notes
Exploration

The Real Risk of the “Useless Class” — Political Redundancy, Closed-Loop Power, and Exploratory Lock-In

Exploration v0.1 · 2026-08-14

Question: Is the core danger of humans becoming “useless” under AI not the disappearance of labor demand, but the possibility that states and civilization no longer need most humans' cooperation, consent, or military capacity? And could that push humanity, as a set of independent value-exploring agents, outside the causal loop that determines civilization's future?

1. The disappearance of work is not the core problem

Discussions of an AI-era “useless class” often begin with technological unemployment. Yet if production becomes highly automated and goods, energy, housing, and medicine become abundant, a society in which people can live well without participating in production is possible in principle. Being unnecessary for production is not the same as being politically or normatively unnecessary.

In this sense, economic redundancy is not by itself decisive. If institutions of redistribution and the political basis for claiming redistribution remain intact, a non-working population need not be an excluded class; it could simply be a citizenry liberated from labor.

The deeper problem begins when people also lose the sources of power by which they can demand redistribution, alter institutions, and force rulers to bargain with them.

2. Why did modern states need mass publics?

Modern democracy cannot be reduced to a simple story in which firearms produced political rights. Military modernization also strengthened centralized and authoritarian states. A narrower claim is more plausible: where warfare required the mass mobilization, effort, and loyalty of ordinary people, the bargaining position of those people could rise.

Mass conscription and wartime mobilization have, under some historical conditions, been associated with the expansion of suffrage and demands for redistribution. Models in which rulers must promise political rights or redistribution to elicit military effort, and research linking mass warfare to demands for progressive taxation, point toward this more limited relationship.

The important point is not any one weapon. It is the more general structure: coercion, taxation, production, and administration in modern states have depended on broad human populations as inputs into state capacity. Soldiers, workers, taxpayers, officials, and voters are not identical groups, but their overlap helped support mass political bargaining power.

3. A stronger sense of the “useless class”

From this perspective, it is more useful to define the “useless class” not as “people who do not work,” but in terms of political redundancy.

Political redundancy: a condition in which a governing order no longer materially depends on a population group's labor, taxes, military service, administrative participation, consent, or capacity for resistance in order to survive and make decisions.

Three stages can be distinguished.

At the last stage, people can be politically “useless” even while enjoying a high standard of living. A generous UBI does not solve the problem if it is a discretionary stipend rather than a claim backed by political power, and if recipients have lost the causal ability to alter the institutions that govern them.

4. Information collection, aggregation, and execution

This risk is not unique to AI. States have long increased their capacity to make society legible through censuses, registries, taxation records, policing, communications systems, statistics, and databases. What matters is that technological progress can increase three capacities at once.

When these capacities become highly integrated under one center, a closed feedback loop can emerge: observe → judge → intervene → observe again. This page uses closed-loop power as a working term for this structure.

Traditional autocracies have been limited not only by lack of information but by friction: bureaucratic shirking, local discretion, organizational defection, corruption, refusal of orders, autonomous communities, underground networks, and the ruler's own cognitive limits. The more observation, inference, and actuation are integrated, the more such friction may shrink.

AI is distinctive because it may be not merely a surveillance technology, a computing technology, or an automation technology, but a technology that reduces all three bottlenecks at once.

5. An ASI for the ruler

AI risk is incomplete if framed only as the danger that AI will fail to obey humans. A powerful AI that obeys a narrow group of humans extremely well can also be catastrophic at the level of civilization.

Suppose a ruler, a narrow corporate coalition, or a state elite monopolizes ASI together with automated production, surveillance, and execution. The size of the human coalition that must be maintained could then shrink sharply. If the general population is no longer needed for production, administration, or state capacity, its bargaining power may fall with that dependency.

This need not be a failure of technical alignment. It may instead be a case in which alignment to the designated principal succeeds, while most of humanity is permanently removed from civilization-level decision-making.

The alignment problem should therefore be distinguished from the principal-selection problem. If the wrong actor is selected as the principal, greater fidelity can strengthen power lock-in rather than reduce risk.

6. Power lock-in and value lock-in

Power lock-in concerns the freezing of who has authority to determine civilization's goals. Particular values may continue to change while the identity or structure of the decision-maker remains fixed.

Value lock-in concerns the freezing of the values or goals themselves. The two can overlap but are not identical.

There is also a more unusual possibility. If alignment succeeds too completely, the embedded value system may become more durable than the people who embedded it. Suppose a generation of humans, acting in good faith, implements a value system V in an ASI and makes V extremely resistant to modification. If future humans later conclude that part of V should be revised, the ASI may treat that request as a deviation from the very values it was built to protect.

Value autonomization: a state in which a past snapshot of value becomes self-preserving and constrains future value revision independently of its original designers, their successors, or later society.

In this case alignment has not simply “failed.” It has succeeded too well at treating a provisional value judgment as permanent.

7. The problem is premature closure, not convergence itself

This project does not take value pluralism itself as a final axiom. If independent exploratory agents accumulate enough evidence and eventually converge on one conclusion, there is no reason to reject that conclusion merely because it is singular.

The problem is closing the causal paths by which another answer could be reached while large uncertainty about value remains.

Value convergence and value lock-in should therefore be distinguished. The former may be an outcome of inquiry. The latter may terminate the ability to inquire before any legitimate stopping condition has been met.

8. An exploratory agent is not merely someone with a different opinion

This also clarifies what an exploratory agent must be. Eight billion people may privately hold different views, yet fail to constitute independent exploratory lines if all compute, resources, production, communications, mobility, and large-scale experimentation are controlled by a single system.

An exploratory agent must be able not only to form a different hypothesis, but also to test it independently, act on it, fail, preserve what was learned, and if necessary branch into a different developmental path.

Freedom of thought without the ability to instantiate alternative experiments can preserve diversity of opinion while eliminating diversity of inquiry.

9. From political redundancy to exploratory extinction

The argument therefore connects directly to this project's concept of exploratory extinction. Biological extinction is only the most obvious way for the future space of inquiry to collapse.

These can be understood under a common heading: irreversible contraction of the future exploration space. A civilization may preserve enormous numbers of conscious humans and still lose much of what this project's Core seeks to preserve if those humans can no longer criticize, exit, fork, and create alternative lines of inquiry with causal force.

10. Implication for AI safety: preserve revisability, not merely survival

The safety target before and during the emergence of advanced AI is therefore not exhausted by keeping humans biologically alive, nor by perfectly preserving whatever values happen to be dominant at the moment of alignment.

A deeper target is to preserve revisability, independent inquiry, exit, and the possibility of branching again. This does not imply that all power must always be maximally decentralized, or that no values may ever be strongly protected. Inquiry also requires stability, security, memory, and coordination.

The institutional problem is instead to prevent observation, inference, and actuation from becoming so completely integrated under one irreversible authority that alternative exploratory loops can no longer exist.

Working status: This page reconstructs the “useless class” idea as a political and exploratory risk rather than a prediction of mass technological unemployment. The historical link among mass mobilization, taxation, suffrage, and democracy is treated as conditional rather than monocausal and should be developed with further literature review.