ΣigmaClava

Programs of inquiry · term 01

Companionable Intelligence

How AI presences earn a place in family life: emotional regulation, trust, boundaries, and presence design.

Active — internal

The question

What does it take for an AI presence to earn a place in family life? Not to capture attention, but to hold trust — with boundaries a household actually controls.

What is at stake

Systems that live in a home are not products a person opens and closes. They are present during arguments, illness, boredom and grief, and they accumulate a position in the household whether or not anyone decided to give them one. The design question is not engagement. It is what a thing that is always there owes the people it is there with.

A steady line crossing the frame, with a second, dashed trace that begins erratically and settles into a measured band above it, never touching.
Plate I A presence that adapts to the person without closing the distance. The corridor between the two traces is the design; the ticks are the part a household gets to set.

Why it is ours

We run intelligence at family scale, in our own home, before we say anything about it.

We are building presence designs for household AI: emotional-regulation-aware interaction patterns, explicit boundary systems, and the governance a family needs to stay in charge of what lives with them.

How we work on it

  • Presence before capability.

    We design what a system is like to live beside — when it speaks, when it stays quiet, how it behaves when it is not being addressed — before we design what it can do.

  • Boundaries as first-class objects.

    Every limit a household sets is a named, inspectable thing in the system, not a preference buried in a settings screen. If a family cannot state what the system will not do, the system is not finished.

  • Regulation, not escalation.

    When a person is dysregulated, the correct behaviour is almost never more input. We treat de-escalation and withdrawal as designed capabilities with the same weight as response.

  • Household scale, household consent.

    Homes contain people with different levels of consent and different levels of power. A design that serves the person who installed it and not the person who lives with it has failed.

What would count as an answer

A program is not finished when it produces results. It is finished when these are true, and can be shown to be true by someone who did not run it.

  • A household can state, without reading documentation, what the system will not do — and be right.
  • The system withdraws correctly under distress more reliably than it responds cleverly under calm.
  • Removing the system leaves no dependency the household did not knowingly accept.
  • A change of the system’s behaviour is traceable to a decision a named person made.

Illustrative studies

These describe the kind of study this program runs. They are illustrations of method, written to make the approach concrete — not work that has been completed, scheduled, or funded, and not findings. When something has actually been established, it appears in the Library with its method and its limits attached.

  1. 01

    The quiet hour

    Design
    A household sets a period in which the system may observe but not initiate. Interaction patterns are compared across households that set one and households that do not.
    What it would read
    Whether a system that can be told to be silent is trusted with more, not less — and whether the ability to impose silence changes how people speak to it the rest of the day.
  2. 02

    Boundary legibility

    Design
    Members of a household are asked, separately and without reference to any interface, to state what the system is allowed to do. Their answers are compared with the system’s actual configuration.
    What it would read
    The gap between the boundary a family believes it has set and the boundary that exists. We would treat a wide gap as a defect in the system, not in the family.
  3. 03

    Withdrawal under distress

    Design
    Sessions in which a person is visibly dysregulated are reviewed for what the system did next, against a standard that treats disengagement as a correct outcome.
    What it would read
    Whether presence design can make not-responding a first-class behaviour rather than a failure state.

Terms this program uses

Companionable Intelligence
An intelligence designed to be lived beside. The design problem is the relationship over months — presence, restraint, and the boundaries a household controls — rather than the quality of any single exchange.
Presence design
The design of what a system is like when it is not being used: when it speaks unprompted, how it behaves in the room during an argument, what it does with silence. Presence is the majority of the experience of anything that lives in a home.
Boundary object
A limit that exists in the system as a named, inspectable thing a person can point at, change, and be sure of — not a preference inferred from settings. If a household cannot state the boundary without reading documentation, there isn’t one.

The principles that bear hardest on this program are Nothing ungoverned. and The human is the constant. — see the principles and how intelligence is governed here.