We are hearing the term ‘ontology’ more and more these days from folks less and less in the semantic world. Its exciting but a a bit confusing. No, it doesn’t solve all problems, but it can help a lot of them.
This is a valuable tool for CDOs and SVPs of Architecture to make their organizations data-focused and #AIReady. But, there’s a lot of hype which can distract from practical basics.

First, to get it out of the way – LLMs will commoditize. Not if but when. The nuances of language are numerous but finite. The race for AI dominance will be won by those with the most money and training data (my money is on google).

Why don’t our Fortune 100 clients just let a language model loose in their organizations?

Because these models don’t understand context. This is where ontology comes in. Simply, ontologies are models of the real world. They form the core of many data management processes, especially Generative AI. For example, consider the interactions between drugs, compounds, molecules, and trials in Pharma. Or transactions, people, accounts, and fraud in Payments. Ontologies are meaningless if built in isolation. They need to be tested against real world data.

When the data resolves (fits) the model is validated. When it doesn’t is where the fun begins. There could be three reasons:

1. Errors – data quality or ingestion errors that can help improve your data pipelines
2. Outliers – surprising insights where data deviates from your prediction
3. Emergent trends – the most interesting, helping refine and improve new ways to increase model predictability

This is crucial because refinement of your ontology is accuracy, and accuracy is value. When an ontology accurately represents your world – like the expected interactions between pilots and planes in the US Airforce or claims, adjusters, and payouts in insurance – it’s the best foundation for adapting generic AI to Enterprise-grade AI.

And its how we are seeing the largest organizations in the world turn pilots into value and create AI applications that actually work.