DISPATCHES · Summit Cognitive

← All dispatches

LegitimacyJuly 27, 20265 min read

The memory of an institution

An organization remembers only what it wrote down. Everything else walks out the door.

There is a comfortable fiction that institutions remember things. We speak of corporate memory, of how a place knows its own history, of the accumulated wisdom that lets a long-standing organization avoid the mistakes a younger one would make. But an institution has no head in which to hold a memory. It cannot recall anything, because there is no one continuous thing there to do the recalling. What an institution has instead is a record — files, minutes, archives, the written residue of decisions past — and what it remembers is exactly that residue, no more and no less. The rest, the vast unwritten remainder, lives only in the people who happened to be present, and people are not a durable medium. They leave.

This is the quiet truth behind every senior person's departure that a place describes, ruefully, as a loss of knowledge. The knowledge was never the institution's; it was the person's, on loan. The reasoning behind a policy that everyone follows but no one can quite justify, the context that explains why a rule has the exact shape it has, the half-remembered argument that settled a question years ago and made all the later questions easier — these things lived in heads, and when the heads left, they went too. What remained was the policy, the rule, the settled question, stripped of the reasoning that produced them. The institution kept the conclusion and lost the argument. It remembered the *what* and forgot the *why*.

The why is the first thing to go

Notice which half survives, because it is not an accident. The *what* of a decision tends to get written down, because the *what* is operational — it is the thing people have to act on, the figure to use, the rule to apply, the outcome to implement. The *why* is reasoning, and reasoning is what people carry in their heads precisely because, at the moment of deciding, it feels too obvious or too intricate or too contingent to be worth recording. We write down the answer and trust that the answer carries its justification with it. It does not. An answer is a residue; the reasoning that produced it is a separate object, and unless that object is captured at the moment it exists, it is not captured at all. By the time anyone wants it — usually years later, when the answer is being questioned — the only copy was in a person who has moved on.

An institution in this state has a peculiar kind of amnesia. It can tell you precisely what it decided and is utterly unable to tell you why. It enforces rules whose rationale no living person can reconstruct. It carries forward commitments whose original purpose has evaporated. And because the conclusions remain perfectly legible, the loss is invisible from the outside; the place looks like it remembers, because all the *whats* are intact and in order. The forgetting shows up only when someone asks the one question the record cannot answer — and discovers that the organization has been operating, sometimes for years, on the authority of reasoning it can no longer produce.

An institution remembers only what it records. The reasoning that lives in a person's head is borrowed memory, due back the day they leave.

What automation does to the borrowing

Now consider what changes when the deciding moves from people to systems. The old arrangement was at least a borrowing — the reasoning lived in heads, and heads, while they were present, could be asked. You could walk down the corridor and ask the person who made the call what they had been thinking, and as long as they were still there, the institution had a kind of memory by proxy. It was fragile and it leaked, but it existed. A system that decides and retains nothing removes even that. There is no head to walk down the corridor to. The reasoning did not move into a person who might be asked later; it occurred and then ceased to exist the instant the output was produced. The institution does not have borrowed memory of these decisions. It has none.

This is the part that should give an organization pause as it hands more of its judgment to automated systems. The risk is not only that a system might decide badly. It is that a system might decide perfectly well and leave behind nothing of *why* — and that this nothing scales. A single departed expert takes one head's worth of reasoning with them. A system that retains nothing takes the reasoning behind every decision it ever makes, at whatever volume it makes them, and takes it continuously, in real time, as it goes. The institution can find itself accumulating a vast and growing inventory of conclusions for which no justification exists anywhere, not in a head, not in a file, not in the system itself. Every *what* present and accounted for; every *why* gone the moment it was formed.

The remedy is not nostalgia for the days of human judgment, which forgot plenty of its own reasoning too. The remedy is to stop treating the *why* as something that takes care of itself. If the reasoning behind a decision matters — and for any consequential decision it does, because the day will come when someone needs it — then it has to be captured at the moment it exists, deliberately, into a record the institution actually holds. The evidence consulted, the rules in force, the path from inputs to outcome: written down not because anyone is asking for them yet, but because the asking always comes later than the reasoning, and by then the only copy is gone. An institution remembers what it records. Automation does not change that rule. It only makes the cost of forgetting to record far steeper, and far more silent, than it has ever been.

— Dispatches · Summit Cognitive

Continue from here

Turn the argument into a practice.

Get new dispatches, assess how your organization handles consequential decisions, or explore Summit Cognitive.