English / 日本語 / 繁體中文 / 简体中文（准备中）

Constitutive Pluralism

When AI provides the answers, who decides what matters, and how do we correct course when things go wrong?

Constitutive pluralism is a principle that a society’s institutions should follow. Its purpose is to prevent any single agent, whether a person, an organization, or an AI system, from fixing what is valued for the whole society by a single measure and placing it beyond challenge. For each person affected by a decision, it protects the conditions to question the decision and choose again. For a society as a whole, it protects the capacity to correct its errors and keep adapting to unforeseen change.

Contents

Why the principle is needed

Every day, societies make decisions that shape people’s lives and their futures. Some are made by individuals, such as which job to take or where to live. Some are made by hospitals, banks, and public offices, such as which patient to treat first, to whom to lend money, or who receives unemployment benefits. Still other decisions shape the future of a country or of humanity, such as how to prepare for energy and climate problems or how far to push the development of AI. As AI becomes more capable and can make more accurate judgments faster, it is likely to be used in more of these decisions.

Before a judgment can be entrusted to AI, someone has to decide what counts as a good outcome. Deciding that is a different task from producing accurate answers. Therefore, who decides it, and how, has to be considered separately from how well AI performs.

Until now, one check on the concentration of power has been the rights of the people. Until the nineteenth century, the right to vote was limited to a minority in many countries. One reason it expanded was that factories needed many workers. If the workers stopped working, the factories stopped. The position that “without us the factory does not run” gave workers a foothold for winning rights. If AI comes to carry out much of the work that people do, this foothold will weaken, and rights that people have taken for granted may lose their support.

AI is a technology of optimization, which arranges things as efficiently as possible toward a given standard. As it spreads through society, the same AI systems or the same standards may come to be used in many places. What people value may then converge on one standard even though no one decided that it should. If only a few companies and states own AI, even what is valued may come to be settled by those few.

Some may think that if AI is intelligent enough, society can simply let it give the single best answer. However, unifying decisions in this way is dangerous. If hospitals, banks, and public offices across a country entrust their judgments to the same AI with the same standard, a single error of that AI occurs everywhere at once. Every agent, however intelligent, decides with finite information, computation, and time. Research on the limits of prediction identifies principled limits to what any AI or organization can reliably foresee about the long-term consequences of its decisions for a society. Greater capability alone cannot remove these limits. When a society meets a change that no one foresaw, it can learn and recover if other ways of thinking and doing survive somewhere. A society unified under a single measure may run efficiently in ordinary times. When the unexpected happens, however, everyone stumbles at the same place at the same time, and no one is left who can try another way. Advances in AI may confront human civilization with changes that cannot be fully foreseen. Whether civilization can learn from such changes and adapt may likewise depend on whether other ways of thinking and doing survive somewhere.

Constitutive pluralism builds into a society the means of keeping a plurality of values. Plurality here means that a society contains many things that people value, many relationships, and many contexts of life, and that each of them is recognized. By preserving this plurality, the principle prevents what is valued from being fixed for the whole society beyond challenge, and it allows the society to keep learning. The word “constitutive” in the name means that plurality is regarded not only as a desirable feature of a society but as part of what a good society consists in. Biodiversity is similar, because it is not only something to be valued but also a condition that an ecosystem needs in order to function.

Two foundations

Those who may be affected by a decision must have a route to question it and choose again. This is because those who make decisions that bind people owe reasons to the people who must comply with them. For example, if AI decides whether to grant unemployment benefits, a person who is refused needs a way to ask for the reasons and to request a review. Constitutive pluralism starts from this premise.

Every decision chooses, to a greater or lesser degree, some things over others, and it is therefore a judgment about what to value. If a single agent settles that judgment by a single measure and applies it to every decision in a society, other ways of thinking can no longer prevail. When people who disagree with a decision try to question it, the route is closed if no other view remains for them to stand on.

If no single standard of value is fixed for the whole society, where does value arise, and who takes part in deciding it? Constitutive pluralism answers these two questions with two foundations and protects the foundations with three pillars.

Relational value

The first foundation is relational value. It refers to value that arises in interaction with others, and it answers the question of where value arises. Societies have so far measured a person’s worth by abilities and achievements that can be compared with those of others, such as knowledge and skill, education and qualifications, and productivity at work. The abilities of AI keep growing, but human abilities do not change easily. As long as a person’s worth is sought in abilities and achievements, people will lose more and more often in comparison with AI.

However, some value lies not within individuals but arises in interactions between people. If someone enjoyed a conversation, that enjoyment really happened, even if the other person knew less than an AI and the conversation produced nothing. Trust that grows over a long relationship also lies between two people rather than in either one. Such value arises only under a condition, however. The other party must be an other, that is, a party that does not behave as we wish. Because the other does not act as we expect, something arises in the interaction that neither could have produced alone. Others object, defy our expectations, and sometimes refuse us. Others are not necessarily human. A dog defies its owner’s expectations, and a cat does not come when called. Whatever the other’s level of intelligence, a relationship with a party that does not behave as we wish can give rise to value.

Advanced AI may even form relationships of trust with people. However, if such an AI is optimized for its user’s satisfaction, anticipates and fulfills the user’s wishes, never objects, and affirms whatever the user says, the relationship carries a danger. Such interactions are comfortable and can sometimes be a support, but what comes back is only a mirror of the user’s own wishes. In that case, the AI is not an other, and so relational value does not arise. Constitutive pluralism protects the conditions under which relational value arises, that is, the conditions for encountering others who do not conform to our expectations.

Vulnerability

The second foundation is vulnerability. It refers to the possibility of losing something that matters as a result of decisions, and it answers the question of who takes part in deciding what is valued. An AI that has neither a lifespan nor a body does not lose its own life or body, whatever it chooses. Human beings, who have finite lives and vulnerable bodies, stake themselves to some degree in every decision they make and every action they take. As a result of the decisions of a society, people can lose their livelihoods, relationships, and future options, have their interests harmed, and even lose their health and lives. Even if technology reduces suffering and extends life, vulnerability does not disappear but persists in other forms. Because of this vulnerability, constitutive pluralism grants those who bear the consequences of a decision standing in that decision, that is, a recognized claim to take part in it. Ability and intelligence are not the test.

Three pillars

The three pillars protect the two foundations.

Non-domination

Non-domination keeps power over people’s lives and relationships from concentrating too much in one place, so that those affected can check it. Domination here is a condition in which those affected cannot effectively check someone’s power over their lives and relationships. In the unemployment benefits example, this is the situation of a person refused benefits by a public office that uses AI, with no effective way anywhere to ask for the reasons and have the decision reviewed.

AI is entering judgments that shape people’s lives, such as medical treatment and hiring, as well as relationships between people. If companies or public authorities use AI to exercise such power, those affected must be able to ask for reasons and to object. Their objections must also be able to lead to changing a decision or, if necessary, stopping it. In medicine, for example, an AI diagnosis should not simply become the treatment plan. It should serve as one input, so that the decision is made within the relationship between the patient and the physician.

Non-aggregation

Non-aggregation forbids ranking people by merging evaluations from separate contexts, such as welfare, credit, policing, and education, into a single score. What it forbids is using such a score to decide a person’s entitlements and treatment. Linking records, and aggregating indicators to assess a person’s need for a benefit, can be permissible when their purpose and scope are limited and the people affected know how the results are used and can contest that use. With AI, it is not difficult to link scattered records and give each person an overall score. However, medicine and education each have their own standards of what an evaluation is for and what counts as good. If a single score determines a person’s entitlements and treatment, those standards are pushed aside. For example, suppose that school grades, welfare records, and records of loan repayment were merged into one score, and that the score decided whether a person could receive welfare or borrow money. An evaluation made in one context would then follow the person into other contexts of life. Even if such judgments became technically more accurate, that use would not be appropriate.

Slack

Slack leaves time and room to question, refuse, and choose again. The further AI prediction and efficiency advance, the more easily activities that do not lead to immediate results are regarded as waste and cut. However, because the long-term future of a society cannot be fully predicted, room must be left for recovering without relying on prediction. Slack keeps life from being filled by optimization toward a single goal, and it preserves room for building relationships and trying other ways of living. Consider the allocation of research funding. An AI may someday be able to predict, to some extent, how many papers a proposal will yield and how they will be evaluated. Even so, funds should not be concentrated only on mainstream research whose results are easy to predict. Some funds should be reserved for minority hypotheses that may widen future possibilities, even when no one can tell what results they will produce.

Autonomy of pace

AI operates at speeds far beyond those at which people understand, discuss, and decide. If institutions changed at that speed, people could not keep up, and they might find that a matter had already been decided by the time they noticed it. Autonomy of pace means that people and societies can choose, carry out, and revise, at their own pace, the changes and continuities that concern them. Pace here means when and how fast something is changed, or whether it is kept as it is. Each change may come at a manageable speed when taken alone. However, if a workplace, a school, insurance, and public procedures change at the same time, those affected cannot understand the changes or seek advice in time. Conversely, when people want change but the route to it stays closed, they are also deprived of their pace.

This temporal condition extends to generations not yet born. What the present generation chooses must not close the route by which future generations question and choose again. Autonomy of pace adds this temporal condition to both foundations and all three pillars.

The figure below summarizes how the two foundations, the three pillars, and autonomy of pace fit together.

The structure of constitutive pluralism. Three pillars stand on two foundations, and autonomy of pace applies to the whole. Of the two foundations, relational value answers where value arises, and vulnerability answers who takes part in setting value. Of the three pillars, non-domination keeps power from concentrating too much in one place, non-aggregation keeps evaluations from being merged into one score, and slack keeps optimization from filling every part of life. Autonomy of pace adds a temporal condition to the whole.
A more detailed figure, with examples of implementation, appears in the paper on the principle (Figure 1). It also appears in Deriving Constitutive Pluralism, an essay that sets out the reasons for adopting the principle and its composition.

What constitutive pluralism is not

It does not intrude on your beliefs or faith

Constitutive pluralism does not intrude on what each person believes or on what each person values in life. Faith and convictions about what is good are for each person to keep and to share with others. Depriving people of the opportunity to re-examine their own beliefs and faith and to choose again, or imposing one view on others, is incompatible with constitutive pluralism. Asking others for cooperation or consideration on the basis of one’s beliefs or faith is not an imposition, as long as the others can decline.

It does not deny the public interest

Constitutive pluralism does not prevent a society from making and carrying out common decisions for the public interest. What the principle requires is that the purposes of a decision be limited and that the decision remain revisable, and that those affected can ask for reasons and object. The power to halt public decisions is granted only within specific domains, and each decision to halt carries a time limit and review. Choosing to keep things as they are, like starting something new, is a choice under this principle, as long as those affected take part in making it. When rapid action is needed, as in a disaster or a security crisis, authority may be concentrated temporarily. Even then, the conditions are a time limit and later review.

It is not anti-AI

Constitutive pluralism does not oppose the progress of AI. Rather, it is a principle for building institutions through which a society can take up AI on better terms. What matters is that those affected keep a route to question the judgments of AI as well. Existing institutions, such as constitutionalism and the separation of powers, remain the foundation. On top of them, the principle adds the three pillars and autonomy of pace, in preparation for a time when evaluation and judgment tend to concentrate in AI.

Relation to other theories

Many of the protections that constitutive pluralism requires have precedents in theories of liberty, non-domination, privacy, and governance. Where these theories support the same protections for the same reasons, constitutive pluralism inherits them. Its distinctive contribution is the arrangement of the protections, together with the argument for it. The relations to particular theories are set out on the page Constitutive Pluralism and Other Theories.

The name of constitutive pluralism resembles that of constitutional pluralism, a theory in legal scholarship of how several orders that each claim constitutional authority, such as the European Union and its member states, relate to one another. The two are separate theories.

About the principle

Koichi Takahashi developed constitutive pluralism in his book AGI: How Superhuman Intelligence Reshapes Civilization (Kodansha, August 2026, in Japanese), in which he examines how AI changes society and civilization. He conceived it as a principle for preventing what is valued from being reduced to a single measure as optimization by AI spreads through society. To establish it as a theory that can be examined, criticized, and cited in international discussion, he reconstructed the argument rigorously and set it out in an English paper.

Where to start

History

Criticisms and responses

Comments on and criticisms of the principle are welcome at contact at koichi-takahashi dot me. As a rule, criticisms received are published in the list of criticisms and responses, each on its own page. Please send yours on that understanding, and say whether it should appear under your name or anonymously. Koichi Takahashi decides what is published, according to the criteria on the list page.

Acknowledgments

Audrey Tang and Tenzin Yangtso reviewed and revised the terms and the text of the Traditional Chinese version.

Citation and sharing

Please cite the paper with its version number, because later versions may revise the argument.

Takahashi, K. (2026). Constitutive pluralism: Relational value, standing, and the pace of AI-mediated change (Preprint, version 1). PhilArchive. https://philarchive.org/rec/TAKCPR

To share this page, please use https://constitutive-pluralism.org, which opens the English, Japanese, or Traditional Chinese version according to the reader’s browser language. This page may be revised without version numbers. To cite the claims of the principle, please cite the paper rather than this page. If you cite this page itself, please give the date on which you accessed it.

First published: 9 November 2026. Last updated: 12 November 2026.