How ontologies relate to AI and Specific benefits of 4D’ism’ to AI.
OK so this has taken me a bit longer than I intended. But that is generally the case for trying to do anything in January, let alone define the scope of what ‘AI’ actually describes and how Advanced Information Modelling (specifically 4D Ontologies) can help this massive field of study.
[Spoiler alert and general safety warning - despite its short acronym, AI as a discipline is, like, HUGE! Complex. And confusingly, ‘loopy’ in its nature - i.e. one field of study is quickly fed back into another and another new field is born! This makes the domain extremely exciting, but also very difficult to keep track of, especially with the current pace of things!]
So, I followed an exercise, through which I answered the following questions stage by stage, with input from researchers in the AIM community (shout outs and thanks to Tony W, Tom K, Syra M and Elizabeth B for your inputs) and yes a degree of generative AI was applied to help answer/undertake some of the analysis…because hey, that’s just how you do things today, I guess.
Q1 - What does ‘AI’ actually describe?
Q2 - Where does Advanced Information Modelling (AIM) fit within the AI domains?
Q3 - What do we currently know about the benefit of ontological research?
Q4 - What does AIM, specifically ‘4d’ Ontologies, offer to the field of ontological research and AI more generally.
Q5 - Where can the benefits of current AI research be fed back into AIM (4d Ontological research)
After that I’ve rounded off with some general conclusions and thoughts, which I hope you find useful. As I’ve said to people many times before, this is a very complex area so if people do have thoughts, comments, refinements or omissions to report just let me know.
Q1 - What does ‘AI’ actually describe?
Using these sources (here and here) a rough list of research areas was compiled. This was cross referenced with a ChatGPT query (thanks Tony W) to further thrash out some more potentials and the following aggregated overview was produced using Freemind.
Basic overview of AI research domains
As you can see, this is A range of many disciplines. What this diagram isn’t really capturing is the complexity of the interconnections between these domains and disciplines which is constantly interchanging and driving further research and discoveries. Big areas of AI, such as Advanced General Intelligence (AGI) for example, will probably be a consequence of the bringing together of these different, but connected research areas.
But for now, understanding, this is just a simple model to help improve our understanding of a complex field, its a good place to start to understand where AIM fits.
So, next question.
Q2 - Where does Advanced Information Modelling (AIM) fit with this?
And this has a simple answer (phew)
Where AIM fits in AI research
Ok that understood - it’s useful to understand why we think Ontologies (in general) are valuable to AI.
Q3 What do we currently know about the benefit of ontological research?
As the field of ontological research is huge and encompasses studies of both 3D ontologies (the general norm when it comes to ontologies) and 4D ontologies (an increasingly explored alternative for ontological representation as described through AIM) - it’s useful to summarise some of the general benefits of the ontologies and how they can benefit all of the areas of AI described in the diagram.
So, (trying, but failing not to reference Life of Brian here) what have Ontologies ever done for us?
They improve data organization, reasoning, and interoperability.
They enable better semantic enrichment of data, improved contextual understanding, scene understanding, and standardization of underlying knowledge bases.
They can be used to facilitate better human-robot interaction, autonomous navigation, cross-domain reasoning, and dynamic replanning.
They support more complex reasoning such as the formalization of concepts, semantic interoperability, emotion recognition, speech-to-text systems, and adaptive learning.
Theoretically they could also provide better transparency in decision-making, trace reasoning paths, and define game rules, strategies, and outcomes.
(also remembering that these benefits are across all of the areas of AI in general)
And then having considered the general benefits of ontologies and going back to some of the other specific aspects of ‘4D ontologies’ covered in other posts we can move onto the next question:
Q4 What does AIM, specifically ‘4d’ Ontologies, offer to the field of ontological research and AI more generally.
4D ontologies actually push the benefits listed in Q3 even further for two reasons (they are probably more but these seem the biggest at time of writing)
A 4D ontology provides better interoperability of source data for model development and training (highly beneficial for ML and NLP). They do this by addressing issues of standardising different sources of data and information for onward usage. A 4D ontology insists all data is formatted in the same way, regardless of source. This unfortunately does mean a degree of initial pain in building such a model, but provides a distinct benefit when it comes to future processing of the underlying data it contains.
Because its very design is different to conventional datastores (3D) a 4D ontology by its nature handles ‘object change over time’ so if data points change - this change is recorded in the model. This addresses a fundamental limitation of 3D ontological models that have to be worked around with various compensating techniques from other branches of AI to manage or additional 3D databases needing to be built, yes, with their own separate ontologies (but that is a different point for another time).
Still here? OK, well thanks for making it this far. I usually lose people at ‘4D’….
This comes to the final point and poses an interesting point for the community of people currently looking into AIM - what can we use from AI to feedback into the process?
Q5 Where can the benefits of today’s AI research be fed back into AIM (4d Ontological research)
Excitingly we’ve seen two real developments recently that have been really helpful for our work.
Improved information extraction - deep learning, specifically neural network supported information extraction is really accelerating a field that is both seemingly basic but also extremely important and practically very hard to do! Better extraction of reliable data, means more high quality data goes into our ontological models. Whether they are they 3D or 4D this is good news.
Speeding up the labour of model population - one of the limitations of 4D is that it takes time to build (the pain point I mentioned earlier). In the past, model development has mostly been manual and time consuming - but specific branches of AI development can again, speed this up LLM’s and a further branch of NLP (known as Retrieval Augmented Generation (RAG) can be bought to bear to automate and streamline the processes behind model development.
General final thoughts
OK - so this is all a bit big in some ways.
I’m hoping its a good primer for what AIM is and how it relates to AI.
As a general final point, in doing this exercise its become increasingly clear that we are genuinely seeing a big leap forward in the application of AI tools. These are then causing secondary leaps, as we then reapply the tools ‘we’ have made back into other areas. Clearly we’ve seen this with ChatGPT for how we manage and process text. But in parallel with this, and the leaps we’re seeing in information extraction for example, we are seeing the leaps from AI being fed back into areas of software development, such as coding and OCR (to name but a few) that offers many new applications and the development of new technologies (such as AIM) that we wouldn’t have thought possible five years ago.