Skip to main content

What is a knowledge base?

The AI Glossary

In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.

A knowledge base is an organized collection of information that people or software can consult. It might contain product instructions, organizational policies, clinical guidance or answers to common questions. The name describes its purpose, not a guarantee that everything inside it is correct.

Some knowledge bases are collections of documents. Others store structured facts and relationships. What makes them useful is not simply the quantity of material, but how information is identified, maintained and retrieved. A current policy and an obsolete draft should not appear interchangeable.

An AI application can search a knowledge base and supply relevant material to a language model before asking it to answer. This is part of retrieval-augmented generation, or RAG. Retrieval finds information; generation produces the response. The knowledge base is the source collection, not the model itself.

This arrangement can make information easier to update without retraining a model. It can also provide sources that a reader can inspect. Neither benefit happens automatically. Search can miss the right passage, retrieve an outdated version or return material that resembles the question without answering it. The model can then misinterpret even a relevant source.

A useful collection records who owns each item, when it was reviewed and what it applies to. Access controls matter as well: a person’s permission to ask a question should not silently grant access to every document the system can search. Sensitive records and general guidance often need different handling.

For a clinical operations team, a knowledge base might bring together referral requirements, service eligibility rules and approved patient communications. When two documents disagree, the assistant should identify the conflict rather than invent a compromise. Staff need a straightforward way to report an error and have the underlying source corrected. That maintenance loop is as important as the search technology. A well-written answer built from stale policy remains a stale-policy problem, however capable the language model may be.

Sources: Lewis and colleagues — Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020); Sentence Transformers — Semantic Search

To learn more, take a look at the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.


AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.

For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.

Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy

Lead with AI in Health Care

The AI Pragmatist is a monthly newsletter from ALA faculty, with new explainers & essays from the Advanced Learning Academy