Proving the Value of Knowledge Management

With KM as the substrate, the myriad expressions of AI (particularly nonstatistical ones) become much more reliable or, as some may put it, valuable. “You need that foundation in terms of knowledge that exists today that’s structured, that’s easily understood when you need to find something,” Phelps pointed out. The terminology definitions, business concepts, and best practices of KM can supply that structure for machine learning models. In fact, this information is critical for optimizing the performance of these models for common KM outcomes such as process automation, workflow management, and information retrieval. “When business applications or process automation tools execute actions, incomplete or low-quality metadata and data will prevent actions from being reliably achieved,” noted Vaidyanathan. “Hence, KM is the foundation to succeed at any AI or automation-related initiatives.”

KM-Improved AI

The salutary effects of KM on AI increase the value of both KM and AI. There is an abundance of ways in which the former makes the latter better, thereby proving KM’s value. One of the most immediate involves applications of retrieval-augmented generation (RAG) for semantic search, question-answering, content summarization, content comparisons, and much more. For these deployments, KM “has doubled its value,” Smith observed. “As well as for traditional outcomes, it’s now needed for generative grounding.” The generative grounding Smith described entails KM broadening the context of generative models for applications of AI such as RAG. Specific KM tools, including wikis, can provide this context for closed source large language models to answer questions about domain-specific use cases.

According to Phelps, wikis include valuable information such as “specific processes and procedures for certain activities,” which might be useful for models answering questions about contracts, for example. The metadata tags (based on vocabularies and taxonomies) for enterprise content are extremely helpful for RAG use cases. Most organizations don’t realize that for the majority of RAG applications, the models embedding their content and answering questions about their domain-specific knowledge have not been trained. Metadata tagging can make up for this omission, so organizations are not only “getting the right answers to the question, but also ensuring, for content owners, the RAG system is going to the preferred sources for the answers,” Schuerman said.

Governance and Compliance

It’s an oversimplification to solely ascribe proof of the value of KM to aspects of AI. Granted, many of these use cases readily quantify the value of KM. But KM is just as valuable for other considerations, including data governance, data security, and regulatory compliance. Oftentimes, taxonomies and ontologies are useful as metadata models for tagging content according to data governance concerns. Their specificity for doing so is valuable, as organizations can create definitions and tags according to particular regulations or certain industries. “In addition to helping automate business decisions and actions, modeling an enterprise’s data in an ontological form also allows for governing it better,” Vaidyanathan said. “Due to the domain data model, metadata and classification can now be surfaced in a much better manner.”