So today we will start with a new chapter and this means we have a new starting point

in thinking about problems.

And this time the starting point is the following one.

See if you do learning from data, the underlying assumption is that you have enough data to

characterize your task and then you can do the evaluation with neural networks, whatever

we have done up to now.

Now let's assume you have a situation where you do not have enough data but you might

have knowledge from the engineers or the economists who are working in this field for many years.

So then you could build up a rule system which basically is an expert system.

And so the story of expert systems is even older than the story of the neural networks

here.

Let's look at it from the following viewpoint.

If you have something to solve as a thinking scheme, let's take you have to do a forecast

of dollar, euro, whatever is your forecasting problem, so it's only an example.

So you have a problem and then you need an observer who observes the markets to generate

data to describe the underlying mechanisms of this market.

Which means so this is the interaction between the real world and then the representation

of your problem in form of data.

If this isn't done well, everything else will fail.

So you really have to do a copy of the original problem in form of data then.

And afterwards, starting from this data here, you can say now let's apply the learning algorithms

to produce a neural network there which characterizes the interaction of the variables so that you

have done what we have done in the last days.

That's the man way of thinking.

The alternative way which was even more hype wave in the 70s is to think as follows.

The problem is the same, but this time you do not have simply observables, observers,

but you have experts which can describe the inner relationship of the variables in the

problem in form of rules.

Please have in mind also an observer has to be an expert because otherwise you would not

know what's going on there.

But here the underlying is that the expert is able to give you an understanding of the

interaction of the variables in form of rules.

But if you ask 10 different experts, you might get a lot of different rules which are partly

confusing which means they do not fit to each other.

And so the interpretation of this summarized structure there by a computer expert at least

partially has to work through this inconsistencies between all these rules there.

Especially if you speak about economics, you have the situation that 10 different people

can tell you complete different stories.

Nevertheless, let's assume we have done it and the computer expert then have designed

a consistent expert system.

Then you would have another answer to the same problem what we have started with.

But now you have two sides here.

You can say I want to rely only on the data or you can say I want to rely on the interpretation

of the structure by the experts.

In principle, you can fight for both sides because if you have enough data and if you

have good experts then on both sides you might get a good answer.

But on the other side if you have bad data and inconsistency expert descriptions then

on both sides you will fail.

So the point for me is not so much to say the one side is the only good one, the other

side is bad.