• Written by Azhar Aliyar, Director

What is grounding, and how can it be applied in a large organisation to ensure that answers to questions are tied to retrieved sources, asked a data analyst during a meeting to evaluate AsscherAi. Grounding refers to the process of linking every claim in an answer to a specific source, such as a database table or a document, to provide evidence for the claim. This is a critical aspect of AsscherAi, as it enables users to verify the accuracy of the answers and understand the context in which they were generated.

Definition and explanation

Grounding is a mechanism that ensures that answers are not generated based on assumptions or inferences, but rather on actual data that can be retrieved from a source. This is particularly important in a large organisation, where data is often scattered across multiple systems and databases. By grounding answers in retrieved sources, AsscherAi provides a clear audit trail and enables users to drill down into the data to understand the underlying evidence.

The process of grounding involves identifying the relevant sources of data and linking them to the claims made in an answer. This requires a deep understanding of the data structures and relationships within the organisation, as well as the ability to map these relationships to the natural language queries posed by users.

Distinguishing grounding from related concepts

Grounding is often confused with data integration, which refers to the process of combining data from multiple sources into a single, unified view. While data integration is an important aspect of AsscherAi, it is distinct from grounding, which focuses on linking answers to specific sources rather than combining data from multiple sources. Most people get wrong that grounding is just about combining data, but this is incorrect, as it is about providing a clear link between the answer and the source.

Another related concept is data virtualisation, which involves creating a virtual layer on top of multiple data sources to provide a unified view of the data. While data virtualisation can be used to support grounding, it is not the same thing, and organisations should be careful not to conflate these two concepts.

Applying grounding in a large organisation

In a large organisation, grounding can be applied in a variety of contexts, such as when a category manager compares two suppliers, or when a shift supervisor checks the maintenance log. For example, a user may ask AsscherAi "What is the current stock level of product X?", and the system will respond with an answer that is grounded in the inventory management system. The user can then drill down into the data to see the underlying evidence for the answer, such as the inventory levels and transaction history.

By applying grounding in this way, organisations can increase transparency and accountability, and provide users with a clear understanding of the evidence behind the answers. This can be particularly useful in regulated industries, where audit trails and evidence-based decision making are critical.

Limitations of grounding

While grounding is a powerful mechanism for ensuring the accuracy and transparency of answers, it is not a panacea, and there are limitations to its application. For example, grounding may not be possible in situations where the data is incomplete or inconsistent, or where the relationships between data sources are complex or unclear.

In such cases, AsscherAi may not be able to provide a grounded answer, and users should be aware of these limitations when evaluating the system. Organisations should also be careful not to over-rely on grounding, and should consider other factors, such as data quality and user needs, when designing and implementing AsscherAi. To learn more about how AsscherAi can be applied in your organisation, visit our website at www.asscher.ai.

When grounding is the wrong choice

There are situations where grounding may not be the best approach, such as when the question is highly subjective or open-ended, or when the data is highly uncertain or probabilistic. In such cases, other mechanisms, such as natural language processing or machine learning, may be more appropriate. Users should be aware of these limitations and consider alternative approaches when evaluating AsscherAi. If you have questions about how AsscherAi can be applied in your organisation, or would like to discuss your specific use case, please contact us at contact-us@asscher.ai.