The AI Subject Matter Expert – A New Role

By Malcolm Chisholm

For years in Data Governance, we have been working diligently to implement and maintain mature Data Catalogs.  There have been many reasons for doing so including, compliance, data democratization and keeping an accurate data asset inventory.

In the age of Generative AI, we can now see bigger picture solutions that were not obvious before.  With the newer Generative AI solutions, data is an even more essential resource than previously thought.  Generative AI solutions are very data hungry, and require the right data and metadata – and lots of it.  However, it is time to start thinking about Generative AI solutions that Data Governance can provide that will benefit the whole enterprise and offer tangible business value.

One such solution would be a LLM chatbot trained on your proprietary data and metadata for a very specific domain.  This solution would also be known as an AI Subject Matter Expert – let’s refer to it as an AI SME.  The goal would be to have a 24/7 resource that anyone could ask questions to for specific domains of knowledge.  The AI SME’s can be developed today with existing concepts including, low parameter LLM’s, fine tuning, RAG, prompt engineering, and Reinforcement Learning from Human Feedback (RLHF).

Let’s explore some reasons why we need AI SME’s.

Reason 1:  Reduce SME Burnout

In Data Governanec we have seen that SME burnout is real (in fact, it may have even happened to you, dear reader). Often SME’s just want to get their own regular work done without being peppered by questions throughout the day.  This leads to SME’s not wanting to be documented as experts for fear of increased workload.  A properly solutioned AI SME could answer most of these questions, leaving the SME more time for their day job.  The AI SME will not replace the SME, but will take much of the pressure off of them.

Reason 2:  Your Tribal Knowledge Has Left the Property

For various reasons human SME’s with a lot of tribal knowledge will leave the enterprise.  This can happen for a variety of reasons, such as reassignment, people getting new jobs, illness, or retirement.  In these cases, especially with retirement, it can be difficult to train new people in time.  Solutioning a low parameter AI SME would take load off regular staff and provide a resource to preserve what would otherwise be tribal knowledge.

Reason 3:  Multi Agent Solutioning

Once an enterprise has successfully developed multiple AI SME’s, this opens the door for a whole new, grander, Generative AI solution.  Multi Agent Solutioning is when AI SME’s work synergistically to solve problems and answer complex questions spanning multiple domains.  Quick note: the AI SME’s can be generically referred to as Agents in AI parlance.  Think of being able to ask a team of experts a difficult question that they need to research, and this is the crux of Multi Agent Solutioning.  The framework solutions already exist, for example Microsoft’s AutoGen.  AutoGen is open source and ready for you to try right now, and is already being used in many places (the AI race is on!).

This article is intentionally light technical details, however we wanted to present concepts and trends in Generative AI that Data Governance must be aware of.  With awareness of these trends, and knowledge of data assets throughout the enterprise, Data Governance can position itself to lead its own Generative AI initiatives that will have solid enterprise-wide benefits.

Have you ever thought about what an AI SME could do for you, or do you already have one in place? Please let us know below.

#ai #aigovernance #llm #datacatalog #datagovernance #ailiteracy