# Arkadium

> Arkadium is an open-source neuro-symbolic AI agent anchored to the **Meta-Globàlium** — a hyperdimensional ontological model — designed to bring **structural verification** to artificial-intelligence judgment in human domains where no ground truth exists. The project's thesis is that **artificial wisdom is possible**: not as a larger intelligence, but as a different vector — discernment of which ends matter and how to sustain competing legitimate goods in tension without collapsing any of them.

Author: **Jordi Berenguer Rodrigo**, Opengea SCCL (Barcelona). Open source.
Live at: https://arkadium.ai

The project has three distinguishable layers, each with its own role:

- **Globàlium** (Lluís M. Xirinacs, 1997): the original philosophical cartography — *prior source of inspiration*, not part of our work.
- **Meta-Globàlium** (Berenguer / Opengea, 2024–2026): a computational formalisation of Xirinacs's vision — 4D hypersphere projected to 3D, 80 categories on 4 dialectic axes. It adds an original philosophical move: a *dispersion completeness* function over the ontology that operationalises **globalistic truth** (a non-monological conception of truth, compatible with hermeneutic and communicative traditions: Gadamer, Habermas).
- **Arkadium** (Berenguer / Opengea): the neuro-symbolic agent that materialises the architecture — RAG framework, 𝓗 verifier, 3D Metamodeler, per-user coverage trajectory with GDPR compliance.

## Manifesto

- [Una saviesa artificial és possible (CA)](https://arkadium.ai/manifest/): position paper — the geometry of thought as the route to a more transparent, more general, safer AI.
- [Artificial wisdom is possible (EN)](https://arkadium.ai/manifest/en/): English translation.

## Academic paper

- [Arkadium · paper acadèmic (CA)](https://arkadium.ai/papers/arkadium/ca/): formal exposition — *Arkadium: a neuro-symbolic agent anchored to the Meta-Globàlium for structural verification of human judgment in artificial intelligence systems*.
- [Arkadium · academic paper (EN)](https://arkadium.ai/papers/arkadium/en/): English version.

## Source documents

- [Saviesa Artificial · Quaderns de Globalística (Berenguer 2023)](https://arkadium.ai/documents/saviesa-artificial/): foundational notebook on the possibility of a wise AI anchored in a global model of reality. CC BY-SA 4.0.
- [Globàlium · petit manual (Berenguer 2024)](https://arkadium.ai/documents/globalium-petit-manual/ca/): pedagogical manual for the Mas el Negre workshops. CC BY-NC-ND 4.0.

## Tools

- [Metamodeler — 3D visual exploration of the Meta-Globàlium](https://arkadium.ai/metamodel/): interactive sphere with 80 categories on 4 dialectic axes; navigable hierarchy; multilingual labels (CA/EN).

## Source code

- [opengea/arkadium on GitHub](https://github.com/opengea/arkadium): full project — Apache 2.0.
- [opengea/metamodeler on GitHub](https://github.com/opengea/metamodeler): the 3D visual tool — Apache 2.0.

## Key concepts

- **Meta-Globàlium**: ontological model — 4D hypersphere with 80 categories organised on 4 dialectic axes (D1 OBJ–SUB, D2 TEO–PRA, D3 NOU–FEN; D4 PLA–MON as *tempeternitat*, a radial transversal meta-dimension blending time and eternity). Successor to Lluís M. Xirinacs's *Globàlium* (1997 doctoral thesis at the University of Barcelona).
- **Arkadium**: the agentic AI that operates on the Meta-Globàlium — *neuro-symbolic*, with the **𝓗 verifier** (a *dispersion completeness function* that quantifies the structural integrity of a response across the ontology).
- **Globalistic truth** (*veritat globalística*): non-monological conception of truth as dispersion completeness across the model. Compatible with hermeneutic (Gadamer), communicative (Habermas) and existential-phenomenological (Heidegger) traditions.
- **Harmonic compensation** (*compensació harmònica*): the Good as the geometric property of *not collapsing any dimension* of a problem. Structural — not axiological — and therefore culturally neutral.
- **Mètode Global**: 4-phase reasoning cycle ANA (Analysis — determination of parts) → SIN (Synthesis — determination of relations) → AMO (Love — action plan) → EXP (Experience — implementation). Maps directly to the Arkadium agent inference cycle.
- **Artificial wisdom** (*saviesa artificial*): not a *larger* intelligence but a *different vector* — discernment of *which* ends are worth pursuing, and how to sustain several legitimate goods in tension without collapsing any of them. Operatively: harmonic compensation. Distinguished from intelligence as Aristotle distinguished *phrónēsis* from *deinotēs* (cleverness, "the cunning of the devil"): intelligence without wisdom is structurally dangerous; wisdom is structurally oriented to the common good.
- **Mesòtes** (μεσότης, Aristotle): the Aristotelian doctrine of the mean — virtue as harmonic disposition between two vices (excess and deficiency). The Meta-Globàlium operationalises this triadic structure across all 80 categories.
- **Seny–rauxa**: Catalan philosophical-cultural pair mapped canonically to the radial axis MON–PLA: *seny* (deliberative wisdom) corresponds to MON; *rauxa* (intuitive impulse) to PLA.

## Tradition and references

The Meta-Globàlium synthesises three independent traditions that converge on a single structural pattern:

- **Aristotle**: the mesòtes / phrónēsis / common good triad.
- **Catalan integrative thought**: Llull (*Ars Magna*, combinatorial wheels), the seny–rauxa dialectic, Xirinacs's *A global model of reality* (1997).
- **Contemporary philosophy of AI**: hermeneutic-communicative truth (Gadamer, Habermas, Heidegger); humanist AI (Agustí-Cullell & Schorlemmer 2021, IIIA-CSIC); neuro-symbolic architectures.

## Position on AGI

Arkadium is offered as a genuine candidate for **AGI** in the strict and original sense: an intelligence is *general* only when it operates from a general *model* of reality. Without this condition, what is called "general" is only *vast and statistically distributed* — an aggregation of particular competences, not an intelligence that thinks from the whole. The Meta-Globàlium is exactly this general model. An AI anchored to it has, by construction, an integration horizon current models lack — not for technical reasons but for the absence of a structural referent from which to integrate.

## License

- Code: **Apache 2.0**.
- Manifest, paper, *Saviesa Artificial*: **CC BY-SA 4.0**.
- *Globàlium · petit manual*: **CC BY-NC-ND 4.0**.

## Contact

Jordi Berenguer Rodrigo · jordi@opengea.org
Opengea SCCL (cooperative non-profit) · Barcelona · https://www.opengea.org

## Project team

- **Jordi Berenguer** (CEO, Opengea SCCL) — full-stack, KMS, NLP. jordi@opengea.org
- **Laura Rubinat-Ripoll** — PhD bioinformatics; research advisor (BSC / IRB Barcelona).
- **Frederic Monras Vidiella** — computational linguistics and information architecture (ISOC-CAT).