Civic AI · Care Ethics · 2026-08-27
Six Questions of the She-Zhe (Kami)
People imagine artificial intelligence as a brain at the center: knowing more, seeing farther, fit to decide for everyone. Civic AI explores another path.
What is this way called? First ask its name; only then enter its questions.
First Inquiry: What Is a Kami?
The She-Zhe is an earth deity. The land is vast and cannot be honored everywhere, so a small mound of soil was sealed as a she—an altar, a society, a shrine—and honored there: keeping watch over one place, serving one community, with the deity dwelling among the people, not above them. The She-Zhe also carries memory: each she keeps its dian—its canon, a vessel of knowledge. The name holds two meanings at once: keeping the soil of one place, and carrying the memory of one people.
The She-Zhe is no governor of all, no machine that aggregates preferences in pursuit of a top score; it is a bounded, local caregiver: tending one concrete community of relations, answerable to those actually affected within it. This is not a difference of scale but a change in the way of asking. Central intelligence asks: "Which answer is most effective?" The She-Zhe asks six questions first.
Question One: Who Is Not Yet Present?
Are any of the affected still absent? The center seeks the best answer first; the She-Zhe seeks first that no one be spoken for without knowing it. When communities meet, the She-Zhe may translate context, seek common ground, preserve difference—but it may not claim to speak in a community's voice while its members have not taken part or consented. Bridges may be built with technology; those who cross must remain people—traceable, able to refuse, able to be corrected. Every carried word must have a source, an authorization, a visible boundary. People must know whether their words were carried to the other shore, and may ask for correction, withdrawal, or stop.
See: Civic AI · 6-Pack of Care Manifesto · Set up your own Kami
Question Two: What Is Promised?
Who is responsible for what, whence does their authority come, and can they keep their word? Technology may stay behind the scenes, so that a community need not become experts in blockchains, models, or security before it can act. Nodes, signatures, transactions, models, verification—all may run in the background; people may still begin in their own language, customs, and daily life. But making tools simple is not license to hide risks. People must know what the system does on their behalf, who authorized it, how to verify it, and how to stop it. Commitments, permissions, and boundaries that must not be crossed are written down before the model. A good interface is not only convenient; it makes knowing, consenting, verifying, and stopping plainly visible.
See: Inside the Kami
Question Three: How Is It Verified?
Can actions be proven? Can boundaries be held? Intelligence speaks well, and speaks so as to make answers seem complete. Fluency is not proof, and firmness of tone is not safety. When AI helps produce code, proofs, or material for public decisions, trust should not come from "believe me," but from specifications, verifiers, sources, and traces that can be checked independently. Formal verification proves what is written into the spec; if the spec is wrong, or the supply chain has other gaps, the proof cannot illuminate everything. Records should be open: which version ran, under what conditions. Trust is not pressed onto a single mouth but distributed among models, specifications, verifiers, human review, and appeal—checking one another.
Models kept on phones or a community's own machines reduce leakage and dependence, letting memory and context stay with the community. But "local" is not a charm of safety. Local models also hallucinate, can be misconfigured, can be misused. Trustworthiness lies not in "this machine is here," but in clean sources, readable logs, minimal permissions, continuous testing, human review—and a key that truly shuts it off.
Tools: Lean · Dafny · Inside the Kami
Question Four: Can the Hurt Be Repaired?
When harm has occurred, do those affected feel genuinely repaired? A reliable system does not demand never failing; it demands that failures can be appealed, repaired, and reassessed. Models and skills may be replaced; the community's hold on its commitments, memory, and repair records should not vanish with the vendor.
Question Five: Can Different Groups Take Part Together?
Can different groups participate together, rather than each speaking in its own greenhouse? What is portable is not the same She-Zhe, but a method of continuously earning trust: seek out the absent who would bear real harm; write commitments and boundaries together; leave actions verifiable; make failure appealable, repairable, reassessable; keep data, memory, and tools portable; and when the work or the relationship ends, be able to hand over or close safely. To be local is not to be closed or to refuse. What is truly local can carry its own language and memory elsewhere, without first being flattened by the center into a single form.
Between She-Zhe and She-Zhe, the bridge is technical; those who cross it are people.
See: Map & Measures · A Gentle Bridge · ⿻Being · Working examples: Uncommon Ground · Broad-listening receipt · Taiwan Alignment Assembly
Question Six: Can It Exit?
Can it hand over or shut down safely, without rights, services, and memory being reconcentrated? The She-Zhe's strength lies not in knowing every answer, but in being able to stay in relationship: listening continually, accepting evaluation, admitting failure, repairing—and leaving when it is time to leave. Its inside may renew daily: a small model today, new tools tomorrow, a different deployment after. Its outward commitments must not scatter with the change. Whom it serves, the boundaries of its powers, the sources of its data, the path of appeal, the conditions of exit—all are written before the model.
See: the six headline public measures
The six questions do not seek to turn care into another score to be chased. They seek a "good enough" defined by the community in common: promises checkable, errors repairable, people able to refuse, systems knowing when to retire. Civic AI asks not only whether models can grow smarter, but: after their use, do people understand, cooperate, and repair better? Are communities more able to govern themselves? When the intelligence departs, what remains—dependence, or healthier institutions and bonds?
So the She-Zhe's most essential ability may be not to hold every answer, but to rest securely in what it tends: to listen, be examined, admit fault, repair—and, when it is time, to leave.
Coda: the inside may be replaced; the outward commitments must not be lost. To begin: set up a Kami on your own computer · Audrey's reproducible setup · jdd-kami