Selected publications
The record. 2024 – 2026.
- Sep / 2026 Active Inference is Optimal Control Shows that continuous-time Active Inference is finite-horizon path-integral stochastic MPC, and its canonical discretization is classical KL control on belief space. Tree-level minimization of the derived expected free energy exactly recovers Bellman dynamic programming — weakly dominating both extant EFE formulations — letting Active Inference researchers directly leverage modern control engineering. ssrn.com ↗
- Jun / 2026 The CRISTAL Method: Neurosymbolic analysis from AI-synthesized world models. A neurosymbolic framework that turns natural-language prior knowledge into an interpretable probabilistic program with full Bayesian inference and uncertainty quantification. On a synthetic-equities benchmark, CRISTAL reaches Bayes-optimal accuracy with just 5 examples and a 5-second budget — where state-of-the-art LLMs plateau around 40% even with far more data and compute. arXiv.org ↗
- Apr / 2026 Hybrid World Models for Fusion Plasma Simulation FUSIONWM, a physics-informed world model for fusion plasma based on JEPA, trained with 11 differentiable physics constraints. It predicts plasma-state evolution across four tokamak configurations at 1.5× the speed of a conventional PDE solver — and identifies architectures suited to photonic hardware, targeting 30–50× energy-efficiency gains via Q.ANT's photonic processor. zenodo.org ↗
- Feb / 2026 Escape Velocity: Why Fusion Energy Is a Race Against Civilizational Decay A stochastic simulation over 2026–2100 showing that the timing of fusion commercialization is the single strongest determinant of long-run economic outcomes — more influential than climate sensitivity, institutional resilience, or pre-fusion clean energy. The window is narrow, and no adjustment to non-fusion parameters compensates for late fusion. zenodo.org ↗
- Dec / 2025 The GAIA Tech Tree: A Neurosymbolic AI Framework for Strategic Technological Decision-Making Describes the Tech Tree's system architecture: neurosymbolic AI combined with a technology tree for strategic decision-making in the fusion and fission energy sector — and how an open-source approach lets the energy community contribute knowledge and software. zenodo.org ↗
- Feb / 2025 Free Energy Risk Metrics for Systemically Safe AI: Gatekeeping Multi-Agent Study A risk metric for multi-agent AI systems grounded in the Free Energy Principle. “Gatekeeper” agents monitor collective risk in real time and intervene in individual agents’ policies — in a simulated autonomous-vehicle fleet, even a low share of gatekeepers produces significant gains in overall system safety. arXiv.org ↗
- Jun / 2024 Presentation: Gaia: Distributed planetary-scale AI safety Introduces Gaia, a decentralized, crowdsourced, model-based safety oracle drawing on collective intelligence and active inference — covering the key design challenges: scalable world-model and preference elicitation, credit assignment, and computing safety assessments at planetary scale. youtube.com ↗
- Early / 2024 Primordia: Neurosymbolic AI for Fundamental Finance primordia.ai ↗