ai.rizom.brain.post

{
  "body": "Every institution has an official form and a shadow life. The official form is recorded in policies, minutes, role descriptions, dashboards, and plans. The shadow life consists of the exceptions, judgments, private agreements, and improvised repairs through which work actually gets done. One makes the institution legible; the other allows it to adapt. They depend on each other, but never fully coincide.\n\nAI is introduced to the official form and expected to operate within the shadow life. The distance between them is where the machine finds the cracks.\n\nThe official form is the institution's account of itself: how work should happen, who may decide, and what counts as legitimate. Documentation allows that account to outlast its authors and orient people who were not present when it was made. A school's procedures guide new teachers; a foundation's criteria shape later grant rounds; a public agency's mandate coordinates action across offices. Yet the account is never complete. A policy may preserve an intention rather than a settled practice. Minutes record what was said, not what changed anyone's mind. A role description names responsibility while actual authority rests elsewhere.\n\nThe lived institution fills those gaps. It is the colleague who knows which process is ceremonial, the programme officer who remembers why an exception was granted, and the person consulted before an old decision becomes precedent. This knowledge is often practical and generous, but fragile.\n\nIn *The Tacit Dimension*, Michael Polanyi gives the shadow life its most compact formulation:\n\n> We can know more than we can tell.\n>\n> — Michael Polanyi\n\nThis is not a failure of documentation. Practical knowledge always exceeds its formal account. The problem begins when an institution depends on what people know without knowing who carries it, how it can be challenged, or what happens when they leave.\n\nAI enters precisely here. It is asked to retrieve a policy, explain a decision, compare precedents, recommend a next step, or help someone act without the person who normally supplies the missing context. The system can work quickly across documents, but it cannot recover context an institution never made durable. Its failures can reveal where the institution cannot remember, explain, or authorize its own decisions.\n\nThat is the central test. A missing answer is not always a retrieval problem. The relevant reason may never have been recorded. The document may not say whether it is current. The decision may have been made by someone whose authority was understood socially but never defined. The system is not merely failing to find the answer; it is showing that the institution has not made the answer durable enough to travel.\n\nThe machine finds the cracks.\n\nAI can reveal an existing crack, amplify one through unwarranted fluency, or create one through unclear permissions, categories, or authority. It reveals an existing crack when it surfaces a stale document still treated as policy, a decision without a recorded reason, or a role that exists on paper but not in practice. It amplifies a crack when a polished answer gives weak, contradictory, or outdated material more authority than it deserves. It creates a crack when its own boundaries are unclear: when nobody knows whether it is retrieving precedent, applying a rule, classifying a case, or making a new judgment.\n\nThe distinction matters because not every failure proves an old institutional defect. The tool and the institution interact. A system may misread a clear record, or a clear-looking record may conceal an unresolved disagreement. The useful response is to examine both: what the system did, and what the institution had—or had not—made available for it to do responsibly.\n\nIn *The Question Concerning Technology*, Martin Heidegger pushes us beyond a merely technical account:\n\n> The essence of technology is by no means anything technological.\n>\n> — Martin Heidegger\n\nHere that distinction becomes concrete. A failed system does not only report on its own limits. It also discloses the arrangement in which it was asked to act: whose judgment has been recorded, which decisions can be explained, and where responsibility is allowed to rest.\n\nRepair therefore starts with the institution, not with the fantasy of a perfect assistant. A document needs an owner, a status, a date, and a relationship to the practice it describes. A decision needs its reason attached closely enough that someone later can distinguish rationale from outcome. An exception needs to be marked as an exception. A role needs clear limits on what it may interpret, approve, or change.\n\nThis is what distributed institutional memory means: knowing where a claim came from, whether a document remains current, why a decision was made, who may interpret it, and what the system is allowed to do. It means making those connections available without requiring every person to rely on private access or on the memory of one experienced colleague. It also means recording disagreement, uncertainty, revision, and exceptions instead of smoothing them away for the sake of a cleaner archive.\n\nThe constructive response is not an all-knowing institutional brain. An assistant may retrieve a precedent, compare options, identify a contradiction, or show that a policy's status is unknown. It may help people see where a vocabulary fragments across teams or where a recurring repair has never been recognized as part of the work. But it should not silently decide which policy governs, approve an exception, or turn an inference into institutional fact. The system's authority must be narrower than the institution's responsibility.\n\nBuilding this memory is slower than adding a connector. It asks institutions to record why a decision was made, who owns its interpretation, when its status changed, and which exceptions should remain provisional. It asks them to make informal repair visible enough to show what knowledge the system carries and what knowledge still depends on particular people. That work may feel administrative, but it is also a way of making responsibility durable.\n\nAI can help with the repair because its failures make absences easier to see. A failed answer can point to a decision with no rationale, a policy whose status is unknown, or a category that different groups use differently. A fluent answer can be tested against its sources and authority rather than accepted for sounding complete. The system becomes useful not when it always produces an answer, but when it makes the limits of an answer legible enough for someone to act responsibly.\n\nThe machine finds the cracks. Sometimes it finds an old one. Sometimes it widens one. Sometimes, by entering the institution, it makes a new one. Our responsibility is to tell the difference, then decide what should be repaired, what should remain provisional, and what should hold.\n\nAI does not decide what should hold. That remains institutional work.",
  "$type": "ai.rizom.brain.post",
  "title": "The Machine Finds The Cracks",
  "format": "text/markdown",
  "series": "New Institutions",
  "summary": "AI systems that work with institutional memory reveal more than the quality of an archive. Their failures show whether an institution can preserve reasons, mark status, locate authority, and carry context into action.",
  "createdAt": "2026-07-11T17:54:52.100Z",
  "coverImage": {
    "blob": {
      "ref": {
        "$link": "bafkreieupowdu4upohnajgd5vmfhae7td6dqo5dpsxnyc3fwadfalpqine"
      },
      "size": 2691687,
      "$type": "blob",
      "mimeType": "image/png"
    },
    "width": 1536,
    "height": 1024
  },
  "seriesIndex": 5,
  "sourceEntityId": "The Machine Finds The Cracks",
  "sourceEntityType": "post"
}

did:plc:dtxrise7xa4kat6mh4zd4lqe | at://did:plc:dtxrise7xa4kat6mh4zd4lqe/ai.rizom.brain.post/The_Machine_Finds_The_Cracks