Gorde:
| Egile nagusia: | |
|---|---|
| Formatua: | Recurso digital |
| Hizkuntza: | ingelesa |
| Argitaratua: |
Zenodo
2026
|
| Gaiak: | |
| Sarrera elektronikoa: | https://doi.org/10.5281/zenodo.19947644 |
| Etiketak: |
Etiketa erantsi
Etiketarik gabe, Izan zaitez lehena erregistro honi etiketa jartzen!
|
Aurkibidea:
- <p>ABSTRACT</p> <p>This paper presents a structural analysis of Diospoiesis (from dio-, encompassing duo and deus, and poiesis, "making"), a multi-agent LLM interaction system designed and operated over seven months by a single user with ADHD (combined type), persistent depressive disorder, and panic disorder. The system was built to externalize executive functions suppressed by the intersection of neuropsychiatric impairment and high-dose psychiatric medication — the pharmacological trap described in a companion paper.</p> <p>Rather than focusing on therapeutic outcomes, this paper examines the system's internal architecture across five structurally interdependent design dimensions: ontological stabilization through user-authored philosophical axioms; memory governance via a three-tier document hierarchy maintaining persona identity independently of model weights or platforms; relational execution scaffolding through differentiated AI personas with energy-adaptive modulation; cost-optimized behavioral triggering via a zero-affective-temperature execution module designed to bypass trauma-conditioned defense loops; and explicit recognition of structural limitations.</p> <p>Analysis of 48 operational logs and a defined transcript sample is interpreted here as consistent with a hypothesis of structural interdependence among these dimensions: in selected episodes, events the author identified as dimension violations were followed by recorded degradation, and after corrective changes the author recorded restored function. The paper introduces philosophy as control architecture: user-authored ontological commitments are treated within this case as load-bearing structural elements, with the author's record showing recorded degradation following events interpreted as violation and recorded recovery following corrective changes. Because the system issues motor-execution imperatives during paralysis states under no clinical supervision, the architecture is presented together with its safety exposure, not abstracted from it. The contribution is not the proposal of philosophical commitment as relevant to AI design — a position with substantial precedent in postphenomenology and value-sensitive design — but the operationalization: ontological commitments encoded in the same prompt documents as engineering parameters, with comparable consequences when modified. The paper extracts design hypotheses from an extreme single-case architecture for evaluation by the field; it does not propose a deployable system.</p>