Knowledge engineering is the process of acquiring, structuring, and then implementing expert knowledge in a machine-readable format so that AI systems can reason with it. Collaboration with subject matter experts enables knowledge engineers to maintain and update intelligent systems, and constrain them with the domain’s rules, so they are trusted by users.
What is the job of a knowledge engineer?
There is a human side and a technical side to the knowledge engineer’s job. On the human side, they gather domain knowledge in collaboration with subject matter experts, data scientists and software developers, which may involve building consensus using versions of the Delphi Method. This knowledge is fitted into logical structures.
The technical side involves building knowledge graphs, defining rules and constraints for reasoning, and ensuring the system can draw correct conclusions. Typically, knowledge engineers encode the system using description logics, OWL (Web Ontology Language) and RDF schemas that are features of ontological engineering.
Knowledge engineers also need to ensure systems align with emerging trends in the semantic web, such as vertical AI agents.
Is a knowledge engineer the same as an ontology engineer?
Knowledge engineering is a broader category than ontological engineering, although in practice the two disciplines frequently coincide. An ontology engineer focuses on the conceptual scaffolding of knowledge. These frameworks formally define concepts in a domain, using controlled vocabulary for example.
Ontologies set out how key concepts relate to one another, and enable information to be shared consistently between systems (interoperability). This is important in sectors such as medical research. Ontologies are also adopted by global organisations to automate communications in different languages and ensure that meaning is consistent in each cultural context.
While ontology engineering provides the high-level framework for a domain’s expertise, knowledge engineering extends to building and maintaining the system for its intended purpose. The system needs to reason, learn, and interact with users in a way that is both useful and reliable.
This involves designing inference mechanisms, validation workflows, and integration with other AI components – all in collaboration with domain experts.
Is there a method to knowledge engineering?
A 2024 paper describes a knowledge engineering method that was developed using the COVID-19 pandemic. It shows how human expertise in disease spread is captured and formalised to support intelligent reasoning systems in a high-stakes domain.
The method sets out how to elicit, encode, validate, and document causal assumptions from experts when data is sparse or uncertain.
When is a knowledge engineer not needed?
A knowledge engineer isn’t always needed if a domain’s knowledge has already been captured as clean, structured, machine-readable data. Ontology-as-a-service (OaaS) platforms are also emerging that make it easier for subject matter experts to build their own intelligent systems without significant technical support.
Novel approaches such as neuro-symbolic AI, using pattern discovery and data-driven inference, can be implemented without an engineer encoding structured knowledge, although these tend to work best with image recognition and speech processing. Some simpler forms of workflow automation don’t require serious engineering either.
For complex builds where human expertise needs to be explicitly captured, structured, reasoned over, and made explainable (XAI) – a knowledge engineer is very likely to be needed.