Hybrid intelligence systems for reliable automation: advancing knowledge work and autonomous operations with scalable...

Area of Science:

• Artificial Intelligence
• Machine Learning
• Cognitive Science

Background:

• Mission-critical automation requires explainable, adaptive, and scalable decision-making, which traditional symbolic or data-driven methods struggle to provide.
• Existing approaches lack the integrated capabilities needed for complex, end-to-end autonomous workflows.

Purpose of the Study:

• To introduce a novel hybrid intelligence (H-I) system that fuses symbolic reasoning with advanced machine learning.
• To develop a hierarchical architecture inspired by cognitive frameworks for enhanced autonomous operations.

Main Methods:

• A three-level hierarchical architecture integrating Vision Transformers, graph-based neural networks, automated machine learning (AutoML), Joint Embedding Predictive Architectures (JEPA), and reinforcement learning.
• Navigation through file systems, databases, and interfaces; discrete decision-making via trained agents; and adaptive workflow orchestration through planning algorithms.

Main Results:

• Demonstrated near real-time anomaly detection in Space Domain Awareness with a 0.98 precision-recall score using a graph-based JEPA and multi-agent reinforcement learning.
• Achieved significant improvements in autonomously driven simulation setup for Computational Fluid Dynamics (CFD), boosting boundary layer capture propagation and reducing geometry failure rates.
• Validated linear scalability across 2,000 compute nodes for AI model training with minimal latency in cross-domain settings.

Conclusions:

• The hybrid intelligence system offers a robust and transferable framework for automating complex knowledge work in mission-critical domains.
• This approach paves the way for next-generation industrial AI systems in space operations, engineering simulations, and autonomous facility management.