Why do we need Digital Twins?

The key capability that a digital twin delivers is that it collates and represents everything that is known about the "world" of an enterprise, where the "world" is those facts or elements that are important to each individual enterprise. As an example a lemonade stands 'world' could consist of the price, quality and availability of lemons, sugar & water, the cost and availability of labor, transport costs, customer volumes and requirements, the waether forecast etc.

In order for a business to be able to accurately and rapidly decide on what is the best next action it should take it first needs to know the current state of its 'world'. If it has a digital twin of this world that state is available for evaluation in real time, all the time.

What is a Digital Twin?

From the introduction above you might be thinking "it's a database", which is not entirely incorrect but is also not really capturing the essense of what a digital twin is either. Our traditional view of a database, the relational database, is like a bunch of linked excel sheets. That is, pretty rudimentary, static and 2 dimentional. A digital twin can better be thought of as a "graph" (i.e. a graph database) that has "nodes" which are facts or elements in the world, such as 'a customer', that are connected to each other and to different node elements to show how they are related or, well, connected.

Think of the Facebook graph that links its customers together by their relationships with not only other people like friends and relatives but with products they like, brands they buy, restaurants they visit, politicians they vote for. This is a much richer multi-dimentional picture of a 'world' where newly discovered or evaluated data can be added or updated to "enrich" the nodes and connections, creating a living data model, ever changing and evolving to better reflect the real world it is modelling.

Where do we integrate a Digital Twin?

The data landscape in an organization (or "data fabric" if we want to be hip and cool) generally consists of a disparate (random?) scattering of databases, data lakes, data stores, data in applications and paper... In the autonomous enterprise world, in its pure form (somewhat aspirational I know) there is only the digital twin as the central store of all enterprise knowledge. It is the brain, without the processing (that would be analytics).

So an enterprise would have a digital twin at its center and all digital data input streams, the 'senses' of the enterprise, would feed into it to keep it updated and current. All evaluation of that data and creation of new knowledge or 'insight' would happen on that data structure. Models or analytics would be built that effectively apply expertize to that knowledge to deliver decisions to the organization.  

How do they work?

Like our own memory the digital twin does not "do" anything per se but it does need to constantly update and enrich its representation of the world by either 'learning' new information about each element (or 'entity' in graph speak) or 'forgetting' information that is no longer relevant.

They also need to constantly evaluate and re-evaluate how these entities are connected together, the number and strength of those connections, as well as discovering indirect connections or multi step connections between entities. That is no small feat and many graph databases are heavily AI powered in order to be able to achieve that.