A 30-second reading
The Grounding Protocol is a light, revisable practice for inquiries in which several forms of knowing meet and something consequential may follow. It does not remove uncertainty or produce a view from nowhere. It keeps difference, authority, limits and consequence visible long enough for one responsible next movement to become possible.
Knowing remains under the care of what it touches. No centre can complete the field. Every closure must preserve a path of return.
Why grounding?
People, institutions, specialists, documents, AI systems and living places can each disclose something real. None contains the whole inquiry. Human knowledge is embodied and situated, but not infallible. Institutions preserve memory and coordinate responsibility, but can harden partial categories into reality. AI can compare, structure and articulate across large symbolic fields, but fluency does not create lived access, consent or legitimate mandate.
Grounding is therefore not a permanent identity possessed by a specially reliable person. It is a practice visible in behaviour: locating a question, separating observation from inference, naming power and absence, allowing unlike evidence to remain unlike, choosing a proportionate movement and returning to what the movement did.
Full, fallible and differently situated
A person may know a place through a lifetime of work. A hydrologist may know a water system through instruments and comparison. A legal record may establish a formal fact. An AI model may expose a pattern no individual readily saw. Each contribution can be full in its own mode without becoming final for the whole inquiry.
The positions are not equal or interchangeable. They differ in embodiment, history, power, capability, exposure to consequence and right to decide. A person whose livelihood or habitat is at stake is not merely one data point among many. Specialist knowledge does not become sovereign because it travels easily. AI output does not gain authority through speed or polish.
Care is not control
Care here is not kindness added after the analysis, nor permission to manage other people for their own good. It is attention, restraint, maintenance, reciprocity, accountability and repair. A caring inquiry asks not only whether a claim is plausible, but what relation it enters, who carries its burden and what must remain possible if it is wrong.
Care without distributed authority can become paternalism. The protocol therefore treats knowing as polycentric: several legitimate centres may coexist and correct one another without being fused into one commanding view.
Response without representation
More-than-human life need not be translated into a human voice to matter. Soil, water, plants, animals, seasons and material limits may enable, resist, redirect or refuse what humans propose. Their response may be encountered, measured or inferred; it must not be romanticised as speech or consent.
Humans remain responsible for interpretation. To observe that a wetland is changing is not to claim to speak for the wetland. To infer ecological stress is not to receive consent from a species. The distinction keeps living response present without turning it into human ventriloquism.
Eight movements
- Locate the living question. Name the context, stakes, initiator, possible consequence and prohibited use.
- Locate the knowing positions. Name contributors, affected parties, authority, absences and knowledge that should remain protected.
- Separate observation, inference and claim. Distinguish what was noticed, inferred, claimed, proposed, disputed and remains unknown.
- Make the encounter prismatic. Bring unlike readings into relation without forcing premature consensus.
- Establish authority, consent and limits. State who may interpret, decide, pause or stop—and who carries the consequence.
- Choose the smallest responsible movement. Prefer a proportionate, reversible step with explicit burden and stop conditions.
- Return to consequence. Notice what changed across living systems, stewardship and relations, and governance and coordination.
- Record correction and reopen. Preserve what changed, significant disagreement and what could revise the temporary closure.
The cycle may be entered at different points. None of its movements should disappear silently. Sometimes the grounded result is to narrow the question, wait, protect an absence or stop.
AI as rapid snapshots, not absolute knowing
A language model produces finite responses from a changing and partly inaccessible computational process. A response can be treated as a rapid analytical snapshot: useful, coherent and revisable. It should not be treated as absolute knowing because it arrives quickly or in complete sentences.
The protocol does not claim to settle what AI ultimately is. Operational humility works in both directions: neither projected personhood nor reductive certainty should substitute for evidence. What matters here is that present AI systems do not share the human participant’s body, exposure or situated responsibility.
The anti-short-circuit principle
AI assistance becomes depleting when it removes the human learning through which judgement would otherwise mature. AI should help a person notice, compare, articulate and test—not make every ambiguity disappear before the person has encountered it.
Sometimes the useful answer is a distinction, a question, a visible uncertainty, a request for local contact or a bounded experiment rather than a finished synthesis. The measure is not maximal output. It is whether the collaboration increases the participants’ capacity to perceive, judge, correct and act without making them dependent on the system’s fluency.
A minimum record
A light use of the protocol can remain on one page: the living question and boundary; knowing positions and protected absences; observations, inferences, claims and unknowns; agreements and conflicts; mandate, consent and AI role; one bounded movement with stop conditions; and a return date recording consequence and correction.
The record is a memory surface, not proof of truth. More writing does not repair missing authority, consent or knowledge.
What this protocol does not claim
- Grounding is not validation; a grounded process may still be wrong.
- It is not consensus; legitimate disagreement may remain.
- It is not representation; ecological response does not authorise humans to speak for other life.
- It is not extraction; access to knowledge creates no right to capture or publish it.
- It is not AI verification; model agreement is not independent evidence.
- It is not universal method; a situated application may translate, limit, supplement or refuse it.
A question to carry
What would help your present inquiry remain open to correction without leaving you unable to make one responsible next movement?
You can stop here. The central practice is complete: no participant contains the whole field, and every temporary closure must remain answerable to what it touches.
Go deeper
- The Grounding Protocol candidate v0.4 — full HTML
- The Grounding Protocol candidate v0.4 — PDF
- Knowing From the Ground — Short Read
- Protocol Habitat — Short Read
- The Correction Loop — Short Read
You can stop here, return to Café Sofia, or continue into the full Papers library where depth, qualification and disagreement remain visible.