Selective memory

Added · Seed

Seeded from a cross-catalogue triage of six AI pattern libraries (references/ai-pattern-catalogues.md); not yet written from close product study.

What the system remembers about the actor is theirs to inspect, correct, scope, and erase — per fact, per conversation, or by stepping outside memory altogether.

Forms

Recurring forms: a reviewable memory list with per-item deletion, memory scope set per message or per thread, incognito sessions that leave no trace, and graph views of what has accumulated.

Resources & references

To-do

The defaults-govern force wants grounding in consent-design literature. Whether “memory” here and the delegation foundation’s context handling are one territory or two is unresolved.

Consequences

  • what the system holds about the actor is inspectable and editable: continuity becomes a visible, revocable grant rather than an ambient accumulation
  • forgetting is now a feature — per item, per session, or per mode — and it has an interface cost the invisible version never paid
  • curating memory is work most actors will not do until something goes wrong, so the defaults of what gets kept do most of the governing

Related patterns

Enacts

  • Privacy — turns held-forward data into something the actor can see, scope, and revoke
  • Adaptability — memory is the raw material the system adapts from

Complements

  • AI tuning — remembered facts and shaped behaviour are two channels of the same persistence

Related

  • Settings — where memory controls conventionally live
  • Activity log — what the system did versus what it keeps