The rise of Generative AI and Large Language Models in the last two years has led to a wave of interest in semantic relationships and the expression of information and data through machine learning tools. In the rush to jump on the GenAI hype-wagon and find ever more obscure use case applications, we have seen a concomitant rise in magic bean sellers and Wizard of Oz curtain-twitchers telling this is all novel technology and new ideas. However, the fundamental underpinning of these technologies and concepts is the ontological and relational understanding of data and information, and that is thousands of years old. It’s time to demystify this topic and delve into the deep roots of where AI and LLMs actually come from and why ontology matters.
I’ve been getting asked the question, ‘What is the value of ontology for AI?’ The answer I give – all the usual benefits you get from using ontology for information and data management, plus the ability to apply GenAI across cultures, the ability to tailor generically pre-trained transformers (GPTs) and LLMs to your particular organisational and cultural context.
A fully relational ontological approach to information/data modelling means the resultant models are pre-designed and conceptually pre-defined, to natively work in Datalake/Graph Database environments (across and within organisations), are interoperable across broader Digital Twin approaches, and can be munged into multi-dimensional standardised industrial models like 7D/BIM.
So, why does AI provide a new range of Use Cases for ontology and metamodeling? Ontology allows us to move beyond the barriers of legacy data models, which have traditionally used hierarchical models, and it allows us to transition sematic context between languages and cultures. Therefore, both of those business barriers of AI uptake, i.e. data trapped in hierarchical structures and semantically fixed within those inflexible hierarchies can be resolved.
As for the old-school folder-structure style hierarchical Linnaean 18th century ‘one-to-many’ taxonomical model? Well, that just becomes one of the many available models that you can visualise and represent in an ontology. So, you can still accommodate and map to those legacy models, you’re just not constrained to only being able to visualise and compare/crosswalk across data in rigid hierarchical representations. Therefore, you can also map to modern many-to-many, relational database, and multi-entity AI/ML models too, it provides options; it’s not just hierarchical.
Some practitioners may say, ‘Why not use both?’ Well yes, they both have their practical uses; however, the key point is that with an ontological data/information model, you can do both, with a hierarchical model you can only do hierarchical structures. So, if you design and build an ontological system, you can do whichever one you need.
We can also use the ontology to be a part of an overarching conceptual framework for a tiered conceptual approach to data and information management and as a basis for developing and integrating data into Digital Twins.
A key consideration behind proposing this ontological approach is that this isn’t a new development. The complex ontological approach has been around for well over 2000 years (formalised by Socrates in the Greek philosophical tradition drawing on earlier Indian and Chinese traditions). Then a huge leap forward was made in the 9th Century Islamic golden age by Al Khwarizmi (his name is translated into Latin as Algorithm) with the introduction of algorithms as a conceptual and data modelling approach. For context, the constricting and narrow Taxonomy hierarchical model wasn’t introduced broadly into science until the 1700s by Linneas.
From the late 1700s the ‘Respect des Fonds’ ontological modelling approach of understanding information in its original context (rather than imposing a model over it, like Linneas) for metadata management grew out of the newly created Revolutionary French State, in its need to manage information at a national level to found a new Republican model. This approach to data management in capturing complex relationships may originally have been used by the ancient Sumerians and Egyptians c. 5000 BC for metadata and data management. Anyway, the key point stands, it’s a tried and tested approach, and it still offers as many advantages for data management now as it has over the preceding millennia.
NB: The images used here (unless a source is provided) are from my own slide decks and are originally derived from work I undertook on the Asset Management Data Standard and Technical Specifications for Land Transport, NZ Transport Agency, 2019.
Stephen is an Information Management (IM) and AI Consultant. A former engineer, he has over twenty years’ experience as an IM specialist, as a Chief Data Officer, Chief Archivist and thought leader. His drive to establish a strong intellectual foundation for IM has been expressed through writing articles, and contributing to International Standards such as ISO15489, ISO13028, and ISO16175. As a social anthropologist he understands human systems, and as a technical expert he understands information systems, using technology to connect the two is his life mission. He has a disturbing obsession with ontology and is a recovering historian.
See Stephen’s profile here.