Nothing ungoverned

The third of the house's principles reads: Nothing ungoverned. Intelligence in our work is always named, bounded, and answerable to a human. This note unpacks what those three words commit us to, because they are commitments, not slogans.

The three are conjunctive. A system that satisfies two of them is not partly governed. It is ungoverned, and each pair fails in a specific and predictable way.

Named

Every intelligence in a system we build is identifiable: what it is, what model of the world it runs on, what it is for. No ghost features, no unlabeled inference deciding things quietly in the plumbing. If a family cannot answer what is running in our house, and what does it do?, the design has already failed the first test, regardless of how well it performs.

Naming is the cheapest of the three and the precondition for the other two. A system that has not been named has no edges, so every conversation about it is a conversation about something slightly different, and nobody can say whether a given behaviour was in scope.

A product may contain several named systems. A brand is not a name in this sense.

Bounded

A boundary is a statement of what an intelligence may see, remember, say, and touch. It is written down before deployment, legible to the people living with it, and enforced by the system rather than by policy documents.

Boundaries at family scale are concrete: which rooms, which voices, which topics, which hours. Boundaries at commercial scale are the same discipline with different nouns: which data, which decisions, which workflows.

The distinction we hold to is between a boundary and a setting. A setting records an answer. A boundary is the question, the answer, who set it, and what the system does at the edge, as an object a person can point at rather than a value buried three screens deep.

Answerable

Answerable means a human can always ask why, get an honest account, and overrule. Not a feedback form, but an off switch and an explanation, held by someone with standing.

Answerability is a property of an organisation rather than of software. It requires a specific person, rather than a team or a committee or a policy, who is accountable for the behaviour and has the authority to change or stop it. A system answerable to everyone is answerable to no one.

Intelligence that cannot be interrogated or countermanded is not a tool; it is a tenant with no lease.

What the principle has cost us

A standard that never rules anything out is decoration. This one has ruled out work:

• Systems that optimise for attention. Engagement as an objective produces designs adversarial to the person using them, and this is not a close call.
• Anything that charges a person a disclosure, whether a diagnosis or a declaration, to receive behaviour that should have been the default.
• Deployments into a household that the household cannot switch off, completely and without penalty.
• Presentation of a persona as a relationship: a system may have a manner, but it may not imply reciprocity it does not have or a memory it is not keeping.

The same words in two registers

Where our research becomes commissioned work, it is delivered through OccultOpes, and this principle travels with it. OccultOpes' engagement structure is governance made operational: an assessment first, findings in plain language, incremental build, then ongoing stewardship. Nothing is deployed before it is understood, and nothing runs without a steward.

The research house and the commercial division say the same thing in different registers, which is the only way a principle survives contact with commerce.

We hold this principle because ambient intelligence is coming to households and businesses either way. The question is whether it arrives governed. In our work, it does.