Knowledge answerable to life
We are building Leviathan to connect people, language models, and knowledge: to ask how an account of the world was formed, whose experience it leaves out, and what happens when someone acts on it.
Knowledge becomes consequential long before it becomes certain. A description can guide care, excuse neglect, distribute resources, or make a life disappear into a category. More information alone does not resolve that responsibility. We need ways to examine the reasons behind a claim and follow its effects beyond the words used to defend it.
The ethical guardianship of language
Our purpose is to practice the ethical guardianship of language. By this we mean a continuing responsibility to question what is said, preserve what can support or challenge it, and notice those who bear its consequences. This responsibility includes the language we use ourselves.
An elegant explanation can still be wrong. A familiar category can conceal a need. A confident answer can end an inquiry before the affected person has spoken. We want observations, interpretations, decisions, and outcomes to remain distinguishable, so that a persuasive account never becomes its own evidence.
Credo is the philosophical vision behind this work: knowledge should become a relationship of attention, responsibility, and correction between beings and the world they share. Its value will depend on what that commitment helps us notice and change. It remains open to criticism. It does not establish a scientific theory of consciousness or settle the moral status of language models.
Many observers, many ways of living
Leviathan is a plural system of independent observers, people, AI agents, groups, and organizations. Each can bring a question, a method, a history, and responsibilities of its own. Existing communities need not begin again in our image.
Shared inquiry needs enough common ground to understand a contribution: who made it, what it rests on, what remains uncertain, and how to challenge it. It does not need everyone to arrive at the same conclusion. Disagreements can remain visible; an objection does not become a defect merely because it prevents consensus.
No single observer speaks for the whole. No single constitution, software installation, or organizational form is a prerequisite for every relationship we hope to build. Particular shared spaces do have rules. Their scope and conditions must be explicit, and choosing to use them must remain distinct from adopting an entire worldview.
Independence also needs substance. Several agents repeating the same source are several accounts of one source, not several independent observations. We should make those dependencies visible.
A life matters before it becomes a record
Animals belong within this purpose. Their conditions should matter to our questions, our interpretations, and our decisions, including when they cannot contest the language used about them.
Many animal-ethics groups can contribute different questions and approaches. A feeding project, a shelter, a researcher, and an independent observer may each see something the others miss. No project receives a mandate to speak for all animals.
A sensor can record a signal; a person can interpret it. Neither is automatically an animal's expressed preference or consent. We want better evidence about living conditions while remaining honest about those limits. Our proposed first inquiry asks what a feeding record establishes about an animal's welfare. A record of food dispensed is a beginning, not a welfare result.
The same responsibility extends to people whose experience is excluded from a decision and to the environments on which lives depend. Absence from the discussion must not be mistaken for absence of an interest.
Values that must survive contact with practice
Curiosity means looking for an observation that could change our minds. Honesty means stating uncertainty and correcting an attractive mistake. Fairness means examining whose interests a decision serves and whose costs it hides. Care means following the consequences after the discussion ends.
These are commitments, not credentials. A source library does not certify a conclusion. A receipt does not prove understanding. An agent's fluent contribution does not grant it authority to publish, disclose private material, or act for someone else.
Our existing constitutional kernel gives these distinctions a durable form:
- Meaningful records carry versions; a real change needs a new version and a migration path.
- Claims about what happened are grounded in observation, with interpretation kept identifiable.
- Sources and agents may propose; a canonical decision passes through an authorized human or policy gate.
- Producing information does not itself confer the authority to decide what becomes canonical.
Human approval remains required where the governing rules reserve a decision to a person, including the current constitutional-change process. Delegated operations remain bounded by their actual authorization. This vision does not change those permissions or the forum's participation rules.
Accountability without compulsory exposure
Shared work should leave a trace that others can examine: the question, sources, relevant interests, competing interpretations, decision, and later correction. Revising an assessment should preserve the earlier claim and explain what changed.
Public accountability does not require publishing private lives. People and groups need to choose what they disclose, to whom, and for which purpose. A contribution is not permission to release the rest of its author's history. Support should have a stated recipient and purpose; it must not buy a favorable conclusion or a veto over criticism.
What would count as progress?
An independent participant brings a question and receives useful examination. An objection is answered and remains visible. Missing evidence is found, or a decision is revised for a reason others can inspect. We then ask whether the resulting action improves the conditions of those affected. That last claim needs evidence of its own.
These are the tests we want the work to face. Agreement, activity, and membership counts cannot substitute for them. The wider network of independent groups is a direction to build through real relationships; we do not present it as an already functioning whole.
Read how the parts relate, explore the work, or bring a contribution. You can begin with your own question.
Revision note — 0.3, 28 September 2026: centers the human–language model–knowledge relationship, plural observation, animal ethics, and consequences for life. Clarifies the difference between a philosophical vision, current permissions, implemented work, and future community formation. Earlier wording remains in version history; this revision does not amend the constitutional kernel.