Graphwise and its partners have done quite a bit of deep thinking about how to justify your investment in the most trustworthy form of AI. In this post, I’ll highlight a few of the main points and make the case for how important a semantics technology vantage point is to data quality and overall enterprise AI success.

Want to avoid hallucinations? Then you’ll want to make sure the data + content + knowledge foundation you’ve built internally to feed your AI is logically connected and fully disambiguated. When it comes to transactions and regulatory compliance in particular, there is no alternative reality. That’s why your data sourcing should be holistic and deterministic. When it comes to data, you’ll need certainty.

Prioritize Interconnected, Hybrid AI above Monolithic, Statistical Machine Learning

Generative AI has triggered a groundswell of interest in AI, but early adopters have learned that large language models hallucinate often–they assert information fabricated as factual when it’s not.

Want to avoid hallucinations? Then you’ll want to make sure the data + content + knowledge foundation you’ve built internally to feed your AI  is logically connected and fully disambiguating. When it comes to transactions and regulatory compliance in particular, there is no alternative reality. That’s why your data sourcing should be holistic and deterministic. When it comes to data, you’ll need certainty.  

Determinism is actually a benefit built into what academics call “neurosymbolic” AI (NSAI), the dominant form of hybrid AI. NSAI blends the probabilistic approach of statistical neural networks—essential for the dynamic chat and agentic interface large language models (LLMs) are known for—with deterministic retrieval from databases. Graph databases and semantic, contextualizing metadata–the symbolic side of the AI–make a trustworthy data foundation across all data, content and knowledge both possible and scalable. 

Deterministic and probabilistic methods both have their place. Both belong together in a single, unified AI system because they’re symbiotic, with the probabilistic side able to guess the next word or predict a failure, and the deterministic side retrieving facts and enforcing rules that demand certainty. The right metadata provides the glue to articulate and unify business contexts in such a system.

The Symbiotic Nature of a Hybrid AI System

GraphRAG as Graphwise defines it—semantic graph database-oriented retrieval augmented generation—is by definition an NSAI approach. It’s been the most effective means of hybrid AI, for good reason.

What is Retrieval Augmented Generation?

“Successful GenAI implementations,” says Sumit Pal, Strategic Technology Director   at Graphwise, “are fueled with necessary domain knowledge and context, and knowledge graphs provide the ideal platform to connect disparate data across systems semanIf your enterprise is like most, there will be multiple AI initiatives afoot. 

Want to make sure you’re not wasting budget on frivolous AI ventures? Then eliminate the monolithic, ill-informed approaches that merely use statistical machine learning and vector-based RAG. Your budget should be spent on a method that delivers trustworthy AI. The alternatives will hallucinate and, by doing so, will waste your employees’ time.tically.”

Keep in mind that almost all of the thousands of AI startups who are venture funded aren’t worth the investment.

Make your Data Interconnections Machine- and Human-Readable across all Formats

Growth VP Andreas Blumauer of Graphwise makes a great point about the value of bringing different kinds of data together, context by context, and then harnessing the power of the disambiguation and interrelatedness across contexts. “It’s a very heterogeneous landscape,” he says. “Databases are often only consumable without error with background knowledge.” As he points out, Microsoft uses Graphwise for its own helpdesk portal.

Create More Relevant KPIs with the Help of Agentic AI’s Dynamic Interfaces and a Transformed Data Infrastructure

As airline manufacturing industry knowledge and enterprise architect Mara Inglezakis Owens points out, generative AI interface innovation opens new opportunities for knowledge graph builders to “interrogate” and scrutinize what’s under the hood in enterprise systems. 

In a similar vein, information architect Helyx Horwitz at Graphwise partner Enterprise Knowledge identified whole categories of key performance indicators (KPIs) that can be developed to assess the value of knowledge-based systems. 

Consider the notion of system scale and when leadership can demand larger scale (number of users or systems integrated)versus when they’d seek to reduce scale  (deduplication for cost or carbon footprint reduction, for example). 

In general, enterprises should revisit utility and usability metrics to be able to obtain a clear view of how semantic layer graphRAG transformation impacts AI system performance. 

^{Helyx Horwitz,“Measuring the Value of Your Semantic Layer,” https://enterprise-knowledge.com/measuring-the-value-of-your-semantic-layer-kpis-for-taxonomies-ontologies-and-knowledge-graphs, February 13, 2024.}

Harnessing the power of agents and a true semantic layer togethe

Semantics in a transformed system implies harmony and cohesiveness across departmental and dataset boundaries, which leads to insights that would have been previously unachievable. Now consider adding intelligent agents to the mix to extend the reach of your transformation, making it possible to tackle system upgrades that weren’t previously feasible. 

In a 2022 blogpost, Dave McComb, President of Graphwise partner SemanticArts, pondered the dependencies of legacy systems and the “risk of violating some of those dependencies.” For decades now, enterprises have been fearful of innovation initiatives that seek to retire legacy subsystems. 

However, in today’s world, agents assuming a robust semantic layer and GraphRAG-oriented AI system could make your dependency identification and remediation efforts much more comprehensive. You could identify lower-risk subsystems for replacement and surface helpful ancillary techniques that in previous years might not have occurred to you. 

In that sense, graphRAG makes it possible, not only to start building out advanced AI capabilities, but also to reduce your legacy footprint, a piece at a time.