# Artificial Deterministic Intelligence (ADI) — New Research Paper & Weekly Scientific Series > Source: > Published: 2026-10-27 14:07:38+00:00 Introducing Artificial Deterministic Intelligence (ADI) **A Continuing Research Series from Cortex Agentics Global** *Probabilistic Reasoning. Deterministic Authority.* I’ve published the first paper in a continuing research series investigating a different approach to engineering artificial intelligence. The central question is: **What if more capable, accountable, and computationally efficient AI systems don’t always require larger language models, but better architecture around them?** Artificial Deterministic Intelligence (ADI) is an architectural approach in which probabilistic models perform bounded inference while external deterministic systems govern memory admission, identity, permissions, execution authority, verification, and release. Instead of requiring a language model to independently maintain and enforce every operational responsibility, ADI investigates which responsibilities can be established and controlled outside the model. ### **Artificial Deterministic Intelligence: Probabilistic Reasoning, Deterministic Authority** Published October 9, 2026. Full research preprint: ### In a first-party synthetic historical-memory benchmark using 30 task designs repeated three times per configuration: - **8B model with governed scoped context:** 90/90 accepted runs. - **35B model with governed scoped context:** 90/90 accepted runs. - **4B model with governed scoped context:** 72/90 accepted runs. - **35B model with broader context:** 46/90 accepted runs, including 43 truncated responses under the fixed output budget. The scoped 8B configuration satisfied the same recorded acceptance contract as the scoped 35B configuration on this bounded workload. This finding suggests an important research direction: how much of an AI system’s operational burden can be handled by governed architecture rather than increased model size? These results do not establish general model equivalence, measured RAM savings, universal reliability, or a general safety guarantee. Independent replication and stronger optimized comparisons remain necessary. ### We are preparing a series of supporting scientific papers, with a target of one evidence-reviewed release each week. **Paper 2 — Externalized Intelligence and Model Substitution** Can smaller models achieve acceptable performance when historical context is governed and prepared before inference? **Paper 3 — Deterministic Worker Construction Before Inference** Investigating worker construction, repeated prompt overhead, token consumption, and inference efficiency. **Paper 4 — Governed Shared Memory for Multi-Agent Systems** Examining shared organizational memory, authorization boundaries, provenance, and record admission. **Paper 5 — Model-Independent Artificial Workers** Exploring whether operational worker continuity can survive changes in the underlying inference model. **Paper 6 — Temporal and Authoritative AI Memory** Investigating memory expiration, supersession, historical validity, and deterministic retrieval controls. **Paper 7 — Behavioral Governance of Artificial Workers** Exploring externally governed worker behavior, permissions, and accountability. **Paper 8 — Governed Persistent Student Memory for Educational AI** Examining personalized educational continuity under human authority and privacy constraints. **Paper 9 — Evidence-Calibrated Computational Persona** Investigating behavioral representation grounded in authorized evidence. **Paper 10 — Persistent Artificial Identity Across Model Generations** Researching continuity across changing models and computing platforms. **Paper 11 — Persistent Artificial Identity in Robotic Embodiment** Exploring governed continuity between digital AI relationships and physical robotic systems. Publication timing depends on evidence verification, technical review, and disclosure approval. Planned titles are not claims of completed experiments. ### We’ve also released the **Cortex Governed Memory Challenge v0.1**, with public synthetic test specifications and behavioral fixtures for examining governed-memory decisions. Challenge documentation: Research series and publication updates: I’m particularly interested in technical feedback on: - Deterministic enforcement boundaries around probabilistic inference. - Optimized retrieval and context-management baselines. - Model substitution and inference resource requirements. - Memory provenance, temporal validity, and access control. - Independent test designs that could challenge or falsify our findings. The underlying Cortex implementation remains proprietary, while approved research descriptions, synthetic challenge fixtures, and published results are available for examination. I welcome constructive criticism, alternative explanations, and independent evaluation. **Branden Laskowski** Cortex Agentics Global *Bridging Humanity and Technology — Human-Centered AI Governed by Construction*