So after our general introduction to what knowledge representation may be, I think I

would like to make a very concrete example.

Actually, it's the first example of knowledge representation for its own sake in AI, and

it goes by the name of semantic networks, and it's relatively intuitive, and you see

that it's coming again and again under new names.

The current new name without the AI flavor is mind maps.

That's also something you can understand from it.

And the idea is very simple, is that a semantic network is just a directed graph representing

knowledge and you have nodes which basically stand for objects or individuals like Jack

and John and Mary, or classes of objects.

We call them concepts.

A robin, that's a certain kind of a bird with a red breast, or birds, those are classes

of objects.

And we also use the edges, which we call links in this semantic networks area, and they represent

relationships between this.

For instance, Jack, a robin is a bird, or a bird has a part, a body part, that are wings.

And so here we have an example, which has some information, and I'm sure you can basically

directly get a picture of what the world is like that is represented by this semantic

network.

So the problem is that we have certain knowledge that's directly encoded in this network, but

very often we want to have more knowledge that we can derive from a network, and that's

kind of where the value of the network is.

And so the idea here that makes semantic networks tick is that you want to encode taxonomic information

about objects and concepts.

And special links, those are the ones that have the is a or inst labels, and those specify

things like property inheritance, and we treat them specially in the process model.

Remember we have the representation and we have the process model, and in this case the

process model is about inferring new knowledge from the network, from old network, knowledge

in the network.

So this is something that could be kind of the whole slogan of knowledge representation.

There's more knowledge in the whatever your approach is called, than is actually written

down.

In the network below, we know that Robins have wings, and so by just looking at the

network, even though there's no link between Robin and wings, we know that Jack is a bird,

or that Jack has wings from this network.

And so the idea here is that we have these special links, is a and inst are their labels,

and we have other links that kind of give us the relationships between the objects.

So to make this formal, we call a link labeled by an is a, an inclusion or is a link, and

if it's labeled by inst, an instants or an inst links.

And an is a link is for inclusion of concepts.

Robin and bird are both concepts, and in particular, Robin is always a bird.

So Robin is a sub-concept of bird, and we have that Jack, for instance, which is an

object, is an instance of a Robin.

In this case, Jack is not a person, but it's actually a special bird.

And then we define the, we define the process model.

And so we first of all specify some notation.

So any link labels that are not inst or is a, we're going to call relations.

And here's the process model.

If we have a semantic network and a relation are, whenever we have a situation where we

have an is a link from A to B and an are link to C after that, or an inst link from A to