The hardest part of an ontology engineer’s job can be getting people to agree on the definitions of key concepts, so they know what to build. Without consensus on what words mean, and how they connect to one another, ontology engineers risk developing knowledge frameworks for AI services that users don’t trust and therefore don’t use.
Most AI needs precise, standardised definitions to automate tasks reliably and reason over knowledge at scale. The consensus of stakeholders on these definitions is both the foundation of an ontology build, and key to its eventual adoption, partly because the process gives users a sense of shared ownership.
What is an ontology?
An ontology is a map and rulebook for a domain’s knowledge, including controlled vocabulary, which enables people and machines to share data consistently. Ontologies have long been employed in domains such as medical research and advanced manufacturing, where accuracy is critically important, and collaboration across different disciplines and languages is a part of everyday life. Some become ontology standards.
The job of an ontology engineer is to design, implement and maintain ontologies. A 2022 paper argued that they should also be trained in specific social-negotiating skills which are currently lacking in most technical courses. The paper positions consensus creation as a fundamental activity in ontology engineering.
Why are ontology engineers in demand?
The advent of domain intelligent systems powered by vertical AI agents, where machines actually reason and share information between one another at scale, makes the role of an ontology engineer important for the future of AI and the semantic web. This is because the consequences of small errors or hallucinations that arise from poorly structured knowledge, tend to compound at scale and can happen very fast.
Natural language processing models (LLMs) are particularly susceptible to hallucination because they work on statistical probability rather than structured knowledge governed by constraints. Ontology engineering, which starts with consensus creation, is one potential answer to this problem.
How is ontology consensus achieved?
Traditionally domain experts such as senior researchers and academics would come together in working groups, standards bodies (ISO, W3C) and international conferences to debate and eventually compromise on an ontology that the sector could work with. This allowed engineers to develop an approved plan and proceed with building the ontology.
But obtaining consensus in this way can be a long, and sometimes interminable process, especially when there are political factors at play, or contentious issues that experts cannot resolve to everybody’s satisfaction. In the worst cases, ontologies end up gathering dust because they are not adopted.
For this reason, approaches such as the Delphi Method, which emphasises anonymity, can be applied to get a working ontology over the line. Digital tools are also being developed that provide an alternative to physical gatherings of experts, and encourage convergence on definitions that people can live with.
What are ontology consensus tools?
Emerging ontology-as-a-service platforms (OaaS) are in some cases offering consensus tools that aim to guide non-technical domain stakeholders towards agreement on ontology definitions, without the need for a dedicated knowledge engineer. They also aim to be more accessible than enterprise-level knowledge platforms such as Palantir, giving small and medium enterprises a path to accurate and trustworthy AI services, able to extract more value from their company data.