D is for Domain Models

From A to I to Z: Jaid’s Guide to Artificial Intelligence

A domain model encompasses all the knowledge AI needs in order to understand a particular topic or perform a specific activity. 

For example, a domain model might contain the rules of chess and the various strategies and approaches AI would need to know to become a good chess player. 

Or it might include the basics of financial services, information on products, rules, and regulations, and best practices for advising or making recommendations to clients, so AI can understand the industry and discuss its finer points in a competent way. 

It’s useful to think of domain models like maps or recipes. 

Maps make it easy for anyone to find their way around a specific area even if they’ve never been there before. And recipes make it possible to replicate a particular dish.

Similarly, domain models provide AI with the knowledge it needs to excel in a given area, task, or activity. 

Domain models are also useful because they give the people involved in an AI development project a shared understanding of what they’re trying to achieve. This helps them make sure they give the AI enough training so it can perform the tasks it’s meant to perform as accurately and effectively as possible.

Some facts:

Building a domain model is a collaborative effort. Once the development team identifies the domain it wants to model, it gathers as much information as possible about it and organises it in a suitable format. 

So, if the domain is financial services, for instance, the development team would build a data bank of the rules, regulations, and best practices pertaining to the industry, as well as input from subject-matter experts. They may also use additional techniques — natural language processing, for instance — to extract further knowledge, such as identifying patterns and connections. The idea is to create a representation of financial services that is comprehensive and coherent enough for AI to operate competently and confidently.  

Once the development team has the basic model down, it’s time to refine it. Typically, there’s a testing and evaluation process in which the model is used to train AI. If the AI keeps making the same mistakes, or struggles with a particular task, the development team will check whether the domain model has any inaccurate or outdated data or knowledge gaps. The development team also uses feedback from the testing process to find ways to expand and improve the domain model.

A domain model isn’t necessarily built specifically for a particular AI system or tool. Many domain models — OpenAI’s GPT model, a natural language processing model used by ChatGPT, is a case in point — are shared across projects. 

Sharing domain models makes it possible for AI systems that perform similar activities to draw on the same knowledge and for development teams on different projects to share expertise. It also means more people can contribute to strengthening the domain model and making it more accurate and comprehensive.

Domain models aren’t just used in AI development, they also have applications in other fields. For example, linguists use domain models to understand the rules and structure of different languages. Similarly, in behavioral science, a domain model can help explain how humans are influenced to make decisions.

Want to know more?

This article is an excellent primer on the fundamentals of data modelling, including key concepts, techniques, and best practices.

And if you fancy putting your data modelling skills to the test, LinkedIn Learning has a series of training courses you can sink your teeth into.

Jaid’s perspective

While domain models are just one aspect of AI, they’re a key enabler, because they make it possible for AI systems to become intimately acquainted with a vast range of subjects and activities — from natural language to the finer points of fund administration.

If you’re interested in learning more about Jaid and how our AI platform utilizes domain models, contact us today for a free demo.