Industrial AI systems relying solely on large language models (LLMs) face significant risks due to their probabilistic nature, which can result in hallucinated or incorrect outputs. For example, an AI incorrectly recommended WD-40 as a lubricant, which could have caused catastrophic equipment failure. This highlights the fundamental challenge: LLMs are pattern-matchers, not reasoning engines, making them unsuitable for safety-critical industrial tasks without augmented verification.
To overcome this, a robust knowledge management strategy is essential. This involves transforming fragmented, unstructured data such as manuals and diagrams into structured, AI-readable formats using shared ontologies or knowledge graphs. Incorporating formal logic enables the system to recognize missing data and avoid guessing, instead escalating queries to human experts. Such an approach provides explainable, trustworthy results essential for precision industries.
GlobalLogic, a Hitachi company, is developing this integrated system combining generative AI as a user interface and a deterministic reasoning engine for core logic. This knowledge-first model supports safer autonomous operations, increases efficiency by limiting GenAI’s computational use, and facilitates edge AI deployment. The structured, verifiable foundation enhances trust and ensures industrial AI systems can be reliably adopted in environments where errors have costly implications.
👉 Pročitaj original: CIO Magazine
