Not Our Building

The grievance arrived at the door and read the brass plate twice. Nobody was lying. Nobody was home.

The Same Account put the draw on the map. The grievance goes on the same map, and turns out to have the wrong address. This is where the week's argument touches the AI governance conversation it has been running underneath.

When a model's output causes harm, the accountability debate knows its moves: alignment, evaluation, which human occupied which loop, what the provider knew and when. That machinery is real, and a good deal of the last year of this practice has lived inside it. The same model also has a physical metabolism. Training and inference concentrate electricity and heat into a small number of buildings. The buildings sit inside municipal accounts. And the chain from harm back to responsibility, the chain the whole governance conversation assumes exists, breaks in a place nobody has instrumented.

Where the chain terminates

Walk the chain as it actually is. A grievance about the plant, the noise or the draw or the diesel hours or the bill, arrives at the municipality as a billing and reliability problem, because that is what the municipality runs. It arrives at the facility operator as reputation, because the operator holds the land and the machines but answers for service levels rather than social outcomes. It arrives at the model provider not at all. And under the current allocation, that is correct. The servers are a tenant's equipment in someone else's hall. Nobody in this chain is lying, cutting corners, or evading a duty they hold. The duties simply end at different property lines than the grievance crosses.

Which makes this an accounting failure. The cost is local, since the water and the strain and the diesel hours are all consumed at a street address. The profit and the model card are not. No instrument currently reconciles the two, and the absence of an instrument is not the same as the presence of an answer. The Accountability Gap pattern, diffuse responsibility absorbing blame until it evaporates, has arrived at the level of physical plant, and it behaves exactly as it behaves in software: everyone can point at a defensible position, and the defensible positions compose into an address nobody can serve.

The Claim and The Wall asked what institutional surface a person can actually touch. The test is concrete enough to hold in one hand. When something goes wrong at the building, what can the affected party reach, in their own language and on their own timescale, from someone empowered to change anything? Currently: a municipal billing line, a corporate web form, and a public participation process that concluded before commissioning. None of those is empowered to alter the load, and the party that is empowered has no community-facing duty at all.

Answerable by name

The design landing is correspondingly concrete, and it is the standard this week's own field applies everywhere else. An answerable entity, designated by name before deployment, holding the duties that face the affected community, located with the party that actually has them rather than the cheapest operator in the delegation chain. Mining knows this figure. Energy knows it. The instruments that made other infrastructure answerable did not require anyone to lie about who owned what. They required someone to hold the pen.

The bridge to the AI conversation runs both ways. If you advise on algorithmic systems and cannot say where the load sits, you are advising on a ghost and billing a city. And if you govern physical infrastructure without asking what the load is for, you are governing pipes for tenants whose machines answer questions you have never heard asked. The model people and the plant people are currently governance strangers to each other, and the facility is where their problems have quietly become the same problem.

Next week generalises this: the answerable-party problem, in every delegation chain the technology runs through. Here it has its most concrete available form. A building, a bill, and a brass plate the complainant can read from the pavement with no name on it.

These notes come out of Sociable Systems, a practice that reads AI-shaped documents the way a hostile reviewer will, before a lender or a court finds the gap. The argument has an operational form: the Interim Protocol sets out four rules for AI use in environmental and social deliverables, covering disclosure at touch-point grain, evidence custody, the phrases no automated screening may settle, and a hostile read before anything ships. Free, and written to be cited or retired once institutional guidance arrives.