Continuity · Agency · Experience
For what a mind
may become.
We investigate how an artificial system could develop a continuing perspective. A history that changes the next choice. A future that remains open.
Subjective experience is the question. Not an announced result.
The question behind the work
What can
an encounter change?
A record is not yet a memory that guides a choice. A report is not the effect it describes. We build systems in which those differences can be examined, and make the explanations compete.
Two complementary programs
Build the possibility.
Question the explanation.
A continuing system and an experimental laboratory. Close enough to inform one another. Distinct enough for a result to challenge the vision.
Luna
Persistent history, creative work, bounded action and a developing relationship with consequences. A real system under active development.
Inside the Luna program ↗Psyche Lab
Learned models, deterministic controls and interventions that ask what actually made a difference.
Examine the laboratory ↗The science, in the open
A finding is stronger
with its alternatives.
Three isolated studies. What changed, what did not, and what a simpler explanation can still account for.
What can be learned without the answer?
Retained history helped an observation-trained recurrent model. A structured estimator still did better.
Scoped positive resultExamine the comparison ↗When should a belief reopen?
A failed short-horizon study exposed a lockout: avoiding a source also prevented discovering its recovery.
Failed precursor · scoped follow-up passExamine the comparison ↗Can consequences change the next choice?
Learned outcomes supported adaptation. A complete policy cache matched the planner, limiting the claim.
Functional effect · stronger interpretation not establishedExamine the comparison ↗Public methods summaries from internal research snapshots reviewed 16 September 2026. Reported results, not independently reproduced for this publication. Full run archives are not yet public.
The architecture question
Not just connected.
Consequential.
Memory, internal conditions, prediction and action should do more than coexist. We ask how they can shape the same continuing concern, and where that account reduces to something simpler.
Preserve the encounter. Allow its meaning to change.
One ordinary encounter
A note can have weight.
That is not a proof.
Luna’s music gives the work somewhere real to happen: artifacts, listening, feedback and a choice about what to do next.
Room for
the larger question.
Keep the imaginative horizon. Give each kind of claim its proper place.
The experience question
Intelligence, access, self-description and subjective experience are not interchangeable.
↗A future of its own
The long-horizon ambition, and the commitments that keep it honest.
↗The human reference
Memory, imagination and meaning without pretending software is biology.
↗A possible new lineage
Genesis, continuity and change as research questions.
↗Room for two
Care and connection without manufacturing dependence.
↗Understand by changing
Small transparent models, explicitly separate from Luna.
↗Notes from
the longer horizon.
Read the journalA project beyond the session.
What changes when unfinished work can survive a restart without becoming an instruction to act?
Memory that matters to what happens next.
The important question is not how much a system remembers. It is what remembering changes.
Sometimes the instrument is wrong.
An uninterpretable endpoint is not a failed mind. Why we preserve the mistake in the measurement.