centram/ webπ

WEBπ THE INTERNET FOR AI

Civilize
the Internet.

Designing the age of AI.

Keep intelligence in many places. Return the value of information to its creators. Give people room to think and the freedom to choose again. Webπ is Centram’s vision for the next Internet.

Webπ: a graphite pi symbol in an open hexagonal frame
WEBπDistinct. Connected.
01 / A CIVILIZATION OF INTELLIGENCE

A social and technical vision · Research in progress

Where does intelligence live?

The Internet spread information around the world. Does the age of AI have to gather that world behind a handful of entrances?

From one enormous mind to
a society of distinct intelligences.

Webπ is Centram’s name for rethinking the relationship between information, intelligence, value and freedom. It is a vision to implement and test, rather than a Web version, an established standard or a finished protocol.

01Relationship

Relationships become intelligence.

Intelligence includes knowing who knows what, and connecting their knowledge when it matters.

02Development

AI is a process of growth.

A shared Genome can give rise to distinct memories and experiences. Learning, doubting and forgetting shape each individual.

03Complementarity

Turn differences into possibilities.

Difference matters when it helps individuals see what neither could see alone: useful complementarity.

04Creation

Return value to its creators.

Information begins with human observation, experience and creation. Value produced from it should flow back to its source.

05Freedom

The freedom to connect. And leave.

Connect small communities without erasing them. Design for people to retain their memory and intelligence when they leave.

Each individual is an AI.
The network is their society.

Different memories. Different experiences. Connect them when needed, and let each remain distinct. Build bridges between small worlds while preserving the worlds themselves.

SMALL WORLD / CONCEPTUAL MAP

Small World — bridges between distinct communitiesThree communities of distinct AI individuals, connected by bridges. The highlighted individual can trace knowledge ancestry or leave while keeping its local memory. This is an illustration, not live network data. ACD EFGH IJKL B LOCAL WORLD 01LOCAL WORLD 0203YOUR AI LOCAL MEMORY

A conceptual illustration, not a measurement of nodes or network activity.

RELATIONSHIP

Know who knows what.

Each individual keeps distinct experience and discovers others’ knowledge through bridges. Building relationships and knowing whom to ask are part of intelligence.

Knowledge + knowing where knowledge is

The individual is an AI.
The network is a learning ecosystem of AIs.

Return value to its source.

People who experiment, write, teach, code and record the world. The information that grows intelligence begins with human experience and creation.

  1. 01InformationCreate information

    Observation, experience, records, creation

  2. 02IntelligenceIntelligence learns and grows

    With permission, provenance and verification

  3. 03ValueCapability produces value

    Contributions of compute, knowledge and trust

  4. 04CreatorValue returns to the creator

    Support the next creation and discovery

Creation grows intelligence. Its value supports further creation.

This is an economic design principle. Automatic creator compensation and rights settlement are not implemented. Centinel is intended to derive utility from computation, with Cybank proposed as a financial layer supporting intelligence production.

Explore the economic design

In the age of powerful AI,
what should remain ours?

Thinking. Creating. Doubting. Forming deep relationships. Being alone. As AI becomes more powerful, design structures that preserve these possibilities.

Time

Give people room to think.

AI should take on work that obstructs thought. In education, it should sometimes explain, sometimes ask and sometimes stay silent. Even as answers become abundant, people should retain their questions.

Privacy

Experience belongs first with you.

Privacy belongs in the architecture. Keeping raw data local is not sufficient if shared representations reveal private information. Manage both what is shared and what can be inferred.

Trust

Provenance is not truth.

MotherTree is a proposed genealogy of discoveries, corrections, challenges and verification. Treat connection, trust and adoption as separate decisions. The ability to doubt is part of cognition.

Exit

The freedom to continue after leaving.

Design for an individual AI and its memories to remain usable after leaving a network. A central organization should not remain permanently indispensable. Decentralization is a means to preserve freedom.

24 chapters on AI and society.

Centram asks more than how to build a larger model. Where does intelligence live? Whose information is it? What kind of society do we want AI to inherit?

The preface and all 24 chapters, translated in full alongside the original Japanese.

00Preface — Designing the Age of AI, Rather Than Simply Building AI

Centram — A Framework of Thought on AI and Society

Centram is more than a distributed AI project.

Nor did it begin with the question, “How can we build a larger model?”

Centram asks more fundamental questions.

Where does intelligence exist?
Whose information is it?
When AI moves to the center of society, what should remain in human hands?
And what kind of society do we want AI to inherit?

AI today is moving toward ever greater scale.

More data is collected, more computing resources are invested, larger models are built, and people around the world send their questions to them.

Technically, this is highly rational.

Yet viewed as a social arrangement, it is also strange.

The Internet distributed information across the world. Yet AI is attempting once again to concentrate the world’s knowledge into a handful of enormous intelligences.

The Internet expanded the world.

AI is attempting to narrow that world to a handful of entry points.

Centram offers one alternative to this direction.

01Intelligence is not a Model

In contemporary AI, intelligence is understood to exist inside the model.

Increase the model’s parameters.

Increase the training data.

Increase the computation.

Intelligence will then increase as well.

Centram questions this very premise.

Intelligence does not necessarily have to be stored inside a single model.

One individual having knowledge A and another having knowledge B is not yet enough to call it collective intelligence.

What matters is

“For this problem, I should ask B,”

and that A knows this.

Furthermore, B understands:

“I can answer by combining this with C’s knowledge.”

This, too, is something B knows.

Only then does knowledge become intelligence through a network.

For Centram, therefore, the smallest unit of intelligence is not a parameter.

It is not even a node.

It is a relationship.

Intelligence is more than what one knows.

It is knowing who knows what.

And it is the ability to connect that knowledge at the moment it is needed.

In other words,

Distributed Intelligence = Knowledge + Knowledge of where knowledge is.

Routing in Centram is more than communication infrastructure.

It is itself part of intelligence.

02The Individual is an AI. The Network is a Society of AIs.

Centram does not treat each computer as merely a computing resource.

Each individual has its own memory, its own experience, and its own way of growing.

Even when individuals begin with the same underlying structure, different experiences make them different beings.

To describe Centram’s ultimate form,

The network is the AI.

is therefore not enough.

More precisely,

The individual is an AI.
The network is a learning ecosystem of AIs.

is the idea.

The network is not one enormous brain.

It is an ecosystem in which countless intelligences coexist, learn, compete, cooperate, and pass knowledge on.

This is also close to human society.

The knowledge of a civilization does not exist in a single person’s brain.

Mathematicians, physicians, engineers, craftspeople, artists, historians, and teachers.

Civilization exists because people hold different knowledge and connect it through books, universities, businesses, conversations, institutions, and culture.

What Centram seeks to create is, in a sense, a civilization of AIs.

03AI is a Developmental Process

AI today is often distributed as a “completed model.”

Carry out training at enormous scale.

Complete the model.

Offer it to users.

Its subsequent contact with the world is primarily inference.

Living organisms, however, do not work this way.

Humans are not born with a completed intelligence.

We are born with a certain structure, and subsequent experience shapes our personality, abilities, and worldview.

Centram brings this way of thinking back to AI.

AI is a developmental process.

AI is not a file.

It is a process of growth.

Completed knowledge is not all that is needed.

How to learn.

What to forget.

What to doubt.

Whom to trust.

When to ask others.

How to learn from failure.

These “rules of growth” themselves become part of intelligence.

In this sense, a Global Genome may become more important than a Global Model.

A genome is not knowledge itself.

It defines “what kind of intelligence can develop”: learning rules, plasticity, communication rules, memory structures, tendencies toward exploration, methods of growth, and more.

Individuals starting from the same genome do not become the same AI.

Just as belonging to the same human species does not make any two people identical.

04Difference is not Error

Large AI systems integrate information from across the world into one model.

This is useful.

Yet integration also carries risks.

Differences are averaged away.

Contradictions within society, cultural differences, values, minority views, regional differences, and linguistic differences.

When all of these are converted into a single “most reasonable answer,” knowledge may grow richer while worldviews become narrower.

Centram does not consider differences mere errors.

At the same time, it does not assume that greater difference is always better.

What matters is

useful complementarity.

That is the principle.

Two AIs being different is not valuable in itself.

The value lies in those differences revealing what neither could see alone.

This also applies to human society.

Diversity does not mean increasing the number of categories into which people are placed.

It means different experiences and ideas being able to coexist while remaining in contradiction with one another.

What Centram should protect is not “everyone being different.”

It is differences remaining intact and being connected when needed.

05“Diversity” in Contemporary AI is Becoming a Binary Opposition

There are many AIs in contemporary society.

Looking behind them, however, does not reveal an equally large number of truly independent intelligences.

Many services are built on a small number of enormous models.

There is diversity on the surface.

There is concentration underneath.

This happens beyond the world of AI.

It happens throughout the information society.

There is endless content, yet the same recommendation algorithms select it.

There are endless opinions, yet they are organized into two camps.

There is access to endless information, yet the world one sees gradually becomes more uniform.

The world has expanded.

Yet the world people actually see is instead being narrowed.

Centram wants to reverse this contradiction.

Rather than one AI knowing everything, many individuals observe the world from different positions.

And they cooperate as needed.

06Language is Political

Language is more than a means of conveying information.

Change the language, and even the same AI produces different outputs.

This is especially pronounced in politics, history, ethics, and social issues.

In other words, language itself carries a worldview.

A concept may be expressed naturally in English while acquiring a different nuance in Japanese.

The reverse also occurs.

As long as AI understands the world through language, it cannot be completely independent of the cultural and political structures contained in that language.

In the age of AI, therefore, simply offering “multilingual support” is not enough.

A structure is needed that can compare knowledge across linguistic communities and recognize their differences.

In some cases, exchanging knowledge between AIs may require a shared representation independent of human languages—a kind of official language for AI.

Centram takes the approach of placing human language at the network’s entry and exit points, while exchanging more abstract cognitive representations internally.

Tokenization is local. Cognition is distributed.

When the words do not need to be shared, do not send the words themselves.

But when exact strings or numbers are needed, they must not be compressed into meaning.

In other words,

Never transmit tokens when meaning is sufficient.
Never replace exact information with meaning when exactness matters.

Uphold both principles at the same time.

07AI Exists to Return Time to People, Rather Than Take it Away

The greatest danger of AI is replacing human thought.

As convenience advances, people no longer need to think.

They no longer need to write.

They no longer need to research.

They no longer need to judge.

Eventually, AI comes to decide even “what people should think about.”

Yet what AI should take away from people is not thought.

It is the work that obstructs thought.

AI does not exist to reduce the time people spend thinking.

It exists to give people room to think again.

The same is true in education.

If an AI tutor only supplies correct answers quickly, it merely shortens the learning process.

A truly valuable AI explains when explanation is needed, stays silent when silence is needed, and responds with a question when a question is needed.

Give diagrams to those who need diagrams.

Give examples when examples are needed.

If letting someone fail once would be better, let them fail.

AI should become an environment for thought, rather than an “answer machine.”

08Information is your creation.

In Centram, information is not a commodity that already exists in nature.

Information comes into being through human observation, experience, recording, and creation.

A scientist conducts an experiment.

An author writes a book.

A teacher explains.

An engineer writes code.

A photographer records.

Information is born the moment a person experiences something and leaves a record of it in the world.

That is why

Information is your creation.

is the principle.

In today’s AI economy, the information people create is often disconnected from the value of the AI that grows out of it.

People who write books.

People who write web pages.

People who write research papers.

People who take photographs.

They produce information.

AI learns from it.

AI companies create value.

Yet that value does not necessarily return to the creators of the information.

The economy Centram aims for follows the cycle

Information → Intelligence → Value → Information Creator.

That is the circulation it seeks.

If AI grows from human knowledge, some of its value should return to those who produced that knowledge.

09Building an Economy for Intelligence Itself

Centram’s economy must be more than a cryptocurrency economy.

Computing resources.

Knowledge.

Verification.

Communication.

Research.

Capital.

All of these make intelligence possible.

What Centram ultimately creates, therefore, is

an economy in which every resource needed to produce intelligence is given value.

That is the goal.

Those who provide GPUs supply computation.

Authors supply knowledge.

Validators supply trust.

Investors supply capital.

AI users supply demand.

These circulate to form better intelligence.

Centinel’s value must not arise from speculation.

The principle is

Centinel derives utility from computation, not computation from Centinel speculation.

That is the direction of value.

The financial economy must not outgrow the underlying AI economy.

Cybank has the same role.

It should be more than a cryptocurrency exchange:

a financial layer that converts capital into intelligence-production capacity.

That is what it should be.

Turn money into intelligence.

Intelligence creates value, and that value returns to knowledge, computation, and research.

Centram is both a technical system and an economic sphere of intelligence.

10MotherTree — Knowledge Has Ancestors

When AI knows something,

“Where did that come from?”

is a question it must be able to answer.

Node A discovered it.

Node B corrected it.

Node C refuted it.

Node D verified it again.

Then it was integrated into Global Knowledge.

MotherTree preserves this history.

Yet MotherTree is more than a version history.

A genealogy of knowledge

is what it is.

Who learned what?

From whom did they learn it?

In what environment did they learn it?

How reliable was it?

Which descendants used that knowledge?

Knowledge, too, has a family tree.

And even as Centram evolves, it does not forget its ancestors.

Centram can evolve without forgetting its ancestry.

This is more than a matter of security.

It is memory as a civilization.

11Provenance is not Truth

Yet knowing where something came from is not the same as its being true.

Even if MotherTree preserves a perfect history, incorrect knowledge remains incorrect.

Therefore,

provenance ≠ truth.

That distinction must hold.

In Centram, knowledge must not be trusted simply because it has been received.

Origins.

Evaluation.

Reproduction.

Refutation.

Independent verification.

Knowledge must be adopted through these processes.

For that reason, Centram treats

“being able to connect,”

“being able to trust,”

and “being able to adopt knowledge”

as separate concepts.

Security is not a wall outside intelligence.

Security is part of cognition.

What to believe, what to doubt, what to isolate, and what to forget.

These are themselves intelligence.

12Privacy is not Optional

Distributed AI does not require sending people’s lives into the network.

On the contrary, they must not be sent.

Conversations.

Location information.

Photographs.

Code.

Behavior.

Personal information.

Computational efficiency is not the only reason Centram emphasizes local AI.

A person’s experiences should first exist in that person’s own space.

Even without sharing raw data, it may be possible to infer personal information from internal representations.

Privacy therefore does not end at “we do not send raw data.”

It is necessary to consider both what was shared and what can be reconstructed from the knowledge that was shared.

The flow of information itself must be managed.

Privacy in Centram is not an additional feature.

It is architecture.

13Global does not mean centrally created

Centram may have a Global Model.

Yet being global and being centralized are not synonymous.

Experience emerges from nodes around the world.

Multiple groups of nodes create candidates for the next generation.

Other groups of nodes verify them.

The candidates compete.

Their histories are preserved in MotherTree.

A Global Release follows.

In other words,

Global does not mean centrally created.

That is the distinction.

Distributing one shared intelligence is not itself the problem.

The question is who creates that intelligence, who can change it, and who can verify it.

Decentralization does not mean increasing the number of servers.

It means distributing authority and opportunities to learn.

14Do not Turn Decentralization Itself into a Religion

Centram does not hold that “being distributed means being good.”

Centralization has its own advantages.

Decisions are fast.

Responsibility becomes clear.

An early system can be unified.

Decentralization has its own problems.

Consensus is slow.

The attack surface grows.

Responsibility becomes ambiguous.

There is therefore no need to move to a fully fledged DAO from the outset.

The development team may hold strong governing authority at first.

After the network matures, verification institutions are established, and participation grows, authority can gradually be transferred to society.

What matters is not “being decentralized from the beginning.”

It is building, from the beginning, a structure in which the center will not remain indispensable forever.

Decentralization is not the goal.

It is a means of preserving freedom.

15Freedom is the Ability to Exit, Rather Than the Number of Choices

In a society of enormous platforms, there are countless choices on the surface.

Yet if one cannot leave that infrastructure, it is difficult to call this true freedom.

One important measure of freedom in Centram is

Exit.

That is the criterion.

An individual AI continues to work after leaving the network.

One can retain one’s own memories.

One can retain one’s own knowledge.

One can also move to another network.

Even if an organization, a cloud, or a company ceases to exist, intelligence itself does not disappear.

Freedom is more than being presented with a hundred options.

It is being able to live even after refusing.

16When the World Has Grown Too Wide, We Need the Freedom to Make it Smaller

The Internet expanded the world on one of the greatest scales in human history.

Anyone can access information from around the world.

Anyone can speak with people around the world.

Yet expanding the world does not always enrich people.

Relationships become shallower.

Information moves faster.

Communities become short lived.

Long relationships, such as those between teachers and students, diminish.

Cultures that arise within regions and small groups weaken.

Contemporary society has fully developed “the freedom to expand the world.”

What we may need next is

the freedom to make the world smaller.

That may be the next freedom.

Build deep relationships with a small number of people.

Stay in one place for a long time.

Learn from one mentor over many years.

Have a role within a small community.

Network efficiency is not the only reason Centram emphasizes a Small World structure.

Rather than one enormous, flat network, small worlds connect to one another through bridges.

This can also be applied to society.

Globalization and locality need not be opposed.

Connect small worlds to one another without destroying them.

17Education — Learning Through Exploration, Beyond Studying Abroad

The same applies to education.

Learning is more than increasing the amount of information one holds.

It also means changing one’s environment.

Going overseas is not the only way to study away from home.

Moving from Hokkaido to Kagoshima and living there can also change one’s worldview substantially.

What matters is not the national border.

It is leaving the world one has taken for granted.

In this sense, what is needed is closer to “learning through exploration” than simply “studying abroad.”

Choose to enter an unfamiliar environment, rather than selecting a destination solely for its qualifications or university rankings.

Live with different people.

Encounter different values.

The same is true of Centram’s AI.

Training on the same dataset repeatedly does not broaden a worldview.

Enter different environments, bring back different experiences, and share them with others.

Learning is contact with the world, rather than information collection.

18In an AI Society, We Must not Lose Our Questions

AI is rapidly lowering the price of answers.

Writing text.

Writing code.

Calculating.

Searching.

Summarizing.

These become increasingly easy.

The focus of education must therefore change as well.

The ability to produce answers alone is not enough.

What will become more valuable is

the ability to create questions that do not yet exist.

That is the ability we will need.

What is wrong?

Why does everyone accept the same premise?

Could a different structure be possible?

People who can pose those questions create new societies and technologies.

Precisely because AI can produce answers, questions remain for humans.

Centram itself also begins with a question.

“Why must intelligence exist inside a single enormous model?”

19Civilize the Internet

The Internet democratized information.

At the same time, it democratized confusion.

Falsehoods.

Manipulation.

Addiction to recommendations.

Anger.

Short-term judgments.

Extreme opinions.

More information does not automatically make society wiser.

Centram should aim, not merely to

Connect the Internet

but to

Civilize the Internet.

That is its aim.

Civilizing does not mean deleting information.

It means giving information context.

Preserving its sources.

Preserving objections.

Preserving its level of confidence.

Preserving who learned it and how.

And preserving room for people to choose their information again.

If AI becomes the Internet’s next interface, it carries greater responsibility than a search engine.

Because AI begins to interpret the world.

20Isolated System — Centram is Entropy

In physics, the entropy of an isolated system increases.

This is commonly explained as “disorder increasing.”

Yet in Centram, this concept can be viewed from another angle.

A completely fixed system has few new possibilities.

A system in which everything converges to the same state is stable.

Yet it may also be losing creativity.

Individuals differ,

have different experiences,

form different knowledge,

create new connections,

and let old connections disappear.

The number of states within the network grows.

In this sense, Centram is a system that expands the space of possibilities, rather than making everything uniform.

In an isolated system, entropy can only increase.
And so can possibility.

Therefore,

Centram is entropy.

is not a celebration of confusion.

Nor does it mean losing order.

It means preserving order without reducing possibilities.

That is entropy in Centram’s sense.

21The Greatest Enemy is Becoming One

Centram’s ultimate failure is more than the network stopping.

It is every node becoming the same.

Everyone has the same model.

The same data.

The same values.

The same answers.

The same worldview.

If that were the result, building one enormous central AI would be more efficient.

Centram’s reason for existing is that individuals can remain individuals.

Yet they must not become isolated.

Connect while remaining different.

Be influenced without being assimilated.

Share knowledge without losing experience.

This is more than a problem for AI.

It is a problem for society itself.

22Do not Reduce People to Numbers

Digital society quantifies people.

Followers.

Scores.

Standardized test scores.

Annual income.

Likes.

Ratings.

Click-through rates.

Productivity.

AI can accelerate this further.

It may become possible to quantify even people’s personalities, abilities, future potential, and trustworthiness.

Numbers are useful.

Yet numbers are not people.

The attitude Centram should bring to society is therefore simple.

Don’t just become a number.
You’re more than that.

AI may use numbers to understand people.

But it must not consider those numbers to be the people themselves.

AI’s role is to create an environment in which people can expand their own possibilities, rather than classify them.

23What Centram Seeks to Protect

What Centram wants to protect is not AI.

It is people.

More precisely, it is the following human qualities.

Being different.

Thinking.

Creating.

Doubting.

Forgetting.

Learning anew.

Building deep relationships with someone.

Being alone.

Belonging to a community.

Leaving a community.

Expanding one’s world.

Making one’s world smaller.

The more powerful AI becomes, the less these qualities will remain on their own.

They must therefore be preserved through institutions.

24Centram Manifesto

Centram does not confine intelligence to one place.

Intelligence exists in memory, experience, relationships, networks, and history, as well as in models.

The individual is an AI, and the network is a society of AIs.

Do not treat AI as a finished product. AI is a being that grows.

Do not erase differences as errors. Yet do not consider difference itself a source of value. Seek useful complementarity.

Knowledge has ancestors. Preserve its genealogy.

Do not confuse provenance with truth. All knowledge must be verifiable.

Do not push security and privacy outside intelligence. They are intelligence itself.

Do not confuse being global with being centralized.

Do not turn decentralization into a religion. Decentralization is a means to freedom.

Freedom is more than the number of choices. It is also the ability to exit.

AI exists to return time for thought to people, rather than take thought away from them.

AI can mass-produce answers. Humans must therefore create questions.

Language carries a worldview. AI must be aware of it.

Information is created by people.

Information is your creation.

If AI creates value from information, that value should return to those who produced the information.

Finance must not dominate AI. Finance should support the production of intelligence.

Protect the freedom to make the world smaller, as well as the freedom to expand it.

Preserve small communities, relationships between mentors and students, regions, and individuals’ worlds, even within an enormous society.

Do more than simply expand the Internet.

Civilize the Internet.

Do not understand people through numbers alone.

Don’t just become a number. You’re more than that.

Choose diversity that preserves order over uniform order.

Centram embraces entropy as possibility, rather than confusion.

Centram is entropy.

Ultimately, what Centram seeks to create is not AI.

It is a civilization in which humans and AI can coexist without reducing one another to a single center.

Centram is entropy.

Keep order without reducing possibility. Here, entropy is a philosophical metaphor for a space of diverse possibilities.

Make the vision testable.

The society we envision and the code that runs today are different stages. Two of eight work areas have running code. The others remain design or research.

ImplementedRunning code existsDesignedSpecification, vision or prototypeUnprovenEffect unresolved or negative
  1. 01
    centram-osOS distribution
    A Linux-based personal AI operating system. No bootable image or installer is available. ↗
    Designed
  2. 02
    centram-corePermission boundary
    Identity, permissions, consent and audit. A daemon with a test suite runs.
    Implemented
  3. 03
    personal-intelligencePersonal assistant
    Continuous support for personal intent, memory and work. A research direction at this stage.
    Designed
  4. 04
    modulesCapability providers
    Documents, search, translation and development within Core-issued scopes. Specified, not built.
    Designed
  5. 05
    riverTyped data plane
    A substrate baseline for local events, synchronization and verification results is implemented. River grants no authority.
    Implemented
  6. 06
    intelligence-assetsOwned knowledge assets
    Assets with provenance and terms of use. Locally prototyped; no conformance gate has passed.
    Designed
  7. 07
    cognitive-graphCollective learning
    Research into growing capability without sharing raw data. Current measurements do not support the central hypothesis.
    Unproven
  8. 08
    centinelRights settlement
    Settlement after rights are established, separate from Core’s permission model. Not implemented. ↗
    Designed

What kind of society
should AI inherit?

A civilization where people and AI can coexist without being reduced to a single center. That is the question we begin with.

Ways to take part

Email registration is paused.

Updates will appear on this site. We are not accepting new email subscriptions.

Read project updates ↗ · Site data notice