Putting the Simple in Simulation via Scalable Hybrid-Learning AI

Since the advent of modern computing, scientists and engineers have leveraged its power for resolving large calculations of the physical world.  As early as 1958, NASA used IBM computers for telemetry monitoring, trajectory calculations, and commanding of the Vanguard satellite. Today, scientists, engineers, and manufacturers enjoy a host of high-performance-computing (HPC) products that enable virtual design and analysis of structural features, thermal loads, fluid dynamics and prototypes. With the advent of cloud computing, the speed and capability of HPC has grown exponentially. Yet, in spite of these available resources, only 8% of US SMEs employ HPC in their designs and prototypes. This is due to a skills gap; most engineers will learn one or two software suites — SolidWorks for CAD and ANSYS for structural analysis, for example. A given platform tends to focus on a limited number of things and suites are siloed. These engineers would also need years of additional training to effectively deploy these tools onto cloud computing. Thus, compute scenarios are limited to the few known platforms, rarely deployed on HPC, and opportunities for tying in other physics (e.g., surface oxidation rates) are missed.  MSBAI is filling that gap with GURU – a  cognitive AI assistant that serves as an overarching guide and interface coordinating simulation input and output across multiple platforms. GURU ‘learns’ how to run complicated engineering software so you don’t have to. It enables experts to run end-to-end design and multiphysics analysis for meaningful simulations and virtual prototyping – with little to no training. It allows engineers, scientists, and designers to focus their talents on the applied physics, rather than the programming. To accomplish this feat, it is necessary to enforce best practices and rules (like F=ma), so we have adopted a hybrid symbolic-geometric AI approach employing both innate structure encoded in logic trees, and we achieve sufficient generalization of the system by building and training machine learning models, invoked when problem-specific decisions need to be made (the hybrid cures the common drawbacks of pure symbolic AI or pure machine learning). We call GURU’s agent tooling platform ARTIST, an acronym based on the core approaches:  Automated Reinforcement-learning-guided Tree-based hybrid-Intelligence Synthesis Trainer. ARTIST is not just hybrid intelligence, it is AI that makes AI. ARTIST is our answer to the question: what would it take to capture the expertise from a human user with a particular specialty, and embody it in intelligently driven software that can be served to a broad audience so that they can then perform that specialist task?