The future of telco data management could hinge on the Semantic Web

AI models, particularly generative AI (GenAI) and agentic AI, require real-time access to high-quality, trusted data. This is forcing communications service providers (CSPs) to evolve their data management strategies, and some experts are pointing to a hybrid approach that combines Semantic Web concepts like formal ontologies with large language models (LLMs) as the future of telco data architecture.

An upcoming report from TM Forum about data democratization and modernization of telco data architectures focuses on the more immediate challenges CSPs are facing when it comes to data, such as capturing, processing and securing it and making it trustworthy and widely available. But the experts we interviewed stressed that operators must also be thinking ahead about semantics to ensure that data is understood in context – not just as raw values, but as information that conveys relationships and intent.

As one principal network architect who participated in our survey and report highlighted, a distinction must be made between data, information and knowledge: “Data without information is of little value. You need to have knowledge about the data to make decisions upon [it].”

The difference between data, knowledge and information

TM Forum Chief Architect Dave Milham emphasizes the challenges around data interpretation, sharing an example of how misunderstood semantics led to a real-world close call for one CSP. “In this case, a team wanted to send a promotional mail to high-usage customers, and they found a list of addresses that looked good,” Milham explains. “Fortunately, whilst checking they realized that these were site addresses including a large number of traffic-light controllers and not customer billing addresses.”

This happened because engineers misinterpreted the meaning of “address”. In this case, the word referred to a site location, including traffic-light controllers, not a billing or customer address. Such semantic challenges arise because companies focus too much on format and moving data without deeper understanding of semantic context, according to Milham.

He explains the difference between data, information and knowledge like this (with some help from an LLM):

• Data is raw, unprocessed facts and figures without context. It’s like the individual pieces of a puzzle before they are assembled.
• When data is processed, organized or structured in a meaningful way, it becomes information. This is where context comes into play, turning the raw data into something understandable and useful.
• Knowledge is information that has been further processed, contextualized and understood. It is information combined with experience, insights and intuition, which enables decision making and problem solving.
• In the context of data, an ontology is a structured framework that organizes information into categories and defines the relationships between those categories. It’s like a blueprint for knowledge, helping to ensure that data is interpreted consistently using formal methods of reasoning.

Marrying Semantic Web and LLMs

These distinctions will become increasingly important as agentic AI and LLMs are used in mission-critical telco operations. Dr. Lester Thomas, Head of New Technologies & Innovation at Vodafone Group, believes Semantic Web principles can play a complementary role alongside LLMs to make communication between AI systems more robust.

“We’re looking at using these GenAI models – generic, open models that are not telco-specific – to run our network operations, which is mission-critical national infrastructure,” Thomas explains. “It’s not quite enough just to use these GenAI models – I think there’s something in it to make how they work more robust.”

The Semantic Web, also called Web 3.0, is a set of World Wide Web Consortium (W3C) standards that aim to make data machine-readable. The vision includes an RDF, or Resource Description Framework, which represents relationships to ensure that data is connected in a meaningful way, and like OWL (Web Ontology Language), which provides structured frameworks for categorizing and reasoning about Information and data.

Thomas points to the unique advantages of formal OWL-based ontologies, which include interoperability, logical reasoning, data provenance and standards compliance as areas where LLMs alone are insufficient.

“LLMs operate as ‘black boxes’ with limited insight into their decision-making processes,” he explains, adding that “LLMs cannot inherently track where their knowledge comes from, making it challenging to audit outputs in high-stakes applications.”

‘Hybrid intelligence’

Thomas makes the case for “hybrid intelligence”, where LLMs enhance the usability of intent-driven APIs, such as TM Forum’s Intent Management APIs. Ontologies can be used to define intents and provide a shared understanding of intent across systems. They can also be used to resolve conflicts between overlapping intents and integrate diverse systems.

“I think all this work on intent and ontology could help ensure communications. When you have agents talking to each other and talking to each other in English – how can you make sure that their conversations are robust?” Thomas explains. “Using those small, structured ontologies, and almost forcing a standard way of speaking, might be necessary in order to use Gen AI on things which are mission critical.”