Answers are grounded in this paper and cite their sources.
MIRA/SENTIA is an experimental agent architecture in which autobiographical identity, self-model observations, falsification criteria, epistemic revision, authority boundaries, and action authorization are maintained outside the replaceable generative language model. The work evaluates whether persistent computational identity can be represented as durable, source-linked, falsifiable state rather than primarily as persona prompting or conversational history. Controlled experiments examine evidence-triggered self-revision, fork divergence, cold-start persistence, reasoning-model replacement, owner epistemic pressure, behavior versus self-report, evidence-gated agency, and component ablation. In the reported experiments, a durable identity observation was revised from 0.85 confidence / ACTIVE to approximately 0.62 / UNCERTAIN after source-linked evidence triggered a preregistered falsifier. Identical starting forks developed distinct durable self-model states after different evidence histories. Component ablation produced measurable losses when the persistent self-model, falsification layer, epistemic-authority boundary, or evidence gate was removed. The results support bounded claims about persistent computational identity, falsifiable self-revision, provenance-backed state change, authority separation, and evidence-gated agent behavior. They do not establish consciousness, sentience, phenomenal subjective experience, human-equivalent autonomy, or free will.
No takes yet. Share an insight, caveat, or question.
Nikki Pommells (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: