Can AI prevent the next Fukushima?

Nuclear power is complex and heavily regulated. Every scrap of paper and entry is captured. There are no exceptions.   In the 1980s, three engineers could support a nuclear station’s maintenance function. Today, the same job requires sixty engineers.  Why?  Mountains of unstructured data spread across multiple databases. Overwhelmed engineers grind away, armed only with their experience, intuition, and keyword searching of poorly tagged and often large documents. It is a prescription for disaster.   In 2016, SavantX’s self-learning AI began “thinking” about one station’s data. Each day the station generates more data while SavantX’s AI expands its vocabulary. For this one nuclear station, SavantX’s vocabulary is now more than 125,000 terms with untold relationships across tens of millions records.   SavantX allows the user to issue queries and the AI then classifies, bins, and returns passages. The AI goes further, suggesting documents and revealing relationships it “thinks” are interesting and relevant to the query. The AI never sleeps; it is constantly refining its understanding of relationships within the data.   It’s a challenging use case: The data are heterogeneous, most requires OCRing of scanned images. Compounding this are observations and corrective action programs powered by humans often using varied language over time to describe similar events.   Identifying and sharing experience within and between stations is the first line of defense for safe operations so mistakes are not repeated and best practices can prevail. The nuclear industry avows operating experience as a beacon of safety for the future. Today, this “operating experience” is arguably powered by tribal knowledge. AI is emerging as an important next step to further safeguard against another Fukushima.   The SavantX AI has taught us that critical knowledge for safe operations is indeed retrievable, and reproducible across a wide range of mission-critical industries.