Generative AI

Translating user queries into machine-readable queries

New use cases and solutions result from a combined process and the symbiosis of GenAI and data linking and knowledge representation technologies. For example, typical users in large companies need to be proficient in domain-specific languages if they want to query intra- and inter-domain-specific knowledge. This requirement is usually not met and this is where GenAI comes into the game: because the natural language queries are understood by AI and transformed into machine-readable queries in a comprehensible way.

‍

These queries must be displayed to users and be understandable for them; only then are the queries sent to knowledge bases with "verified" knowledge. This ensures transparency of the process and traceability of the results - and thus follows the objective of "Responsible AI" & "Explainable AI", i.e. concepts that are designed to make the use of artificial intelligence more ethical, transparent and comprehensible.

‍

‍Verified facts instead of "AI hallucination"

The well-known phenomenon of "hallucinating" in GenAI applications is avoided. This is because knowledge graphs are used for reliable knowledge representation: the machine-readable query in the second step is executed against the knowledge graph, which searches reliable data for the patterns and relationships that correspond to the query.

‍

However, training samples for formulating the correct machine-readable queries from the heterogeneous, unpredictable user queries are often not available in large companies and cannot be provided manually due to the high effort involved. The methodology to be researched assumes that it makes limited sense to learn from past queries due to the inhomogeneity of the queries and the dynamics of the schemas. Instead, GenAI should be able to understand and apply schemas and their documentation in a similar way to a human.

‍

The challenge is to develop industrially scalable parallel queries of dynamically changing, distributed knowledge graphs with large amounts of data while taking security aspects into account. CONWEAVER relies on an architecture that is based on the automated collaboration of more than 100 different models and data sets instead of a single large model with data.

‍

GenAI and corporate data integration

• GenAI must work with reliable data in order to create added value for businesses.

• Knowledge Graphs provide the business context that connects business objects within the Industrial Metaverse.

• GenAI and Knowledge Graphs make use of this reliable knowledge to avoid hallucinations and provide useful insights.

‍

‍The core technology of CONWEAVER is currently one of the most business-relevant and decisive trends according to Gartner. Knowledge Graph and Generative AI are classified as "Critical Enablers" in the "2023 Gartner Emerging Technologies and Trends Impact Radar".

‍

‍

‍

‍

‍

‍

‍