So the last topic in our session before lunch was to compare artificial intelligence and

human intelligence. Now the point is why there are other ways to speak about it and to think

about it. As I told you, the sociological way in doing science is not the mathematical one,

but there every three years or so you have to define a new buzzword to get more money for the

research. And one of the other buzzwords in the same direction here is the point data analytics.

And data analytics here means you give the imagination that you speak about something new,

but I want to show to you that is absolutely the same what we have discussed before. So in data

analytics standard view on looking at this is to say we have to speak about descriptive analytics,

diagnostic analytics, predictive analytics, and prescriptive analytics. And so what is the idea?

Descriptive analytics means you have to say something about the question what happened.

Yeah, this is nothing else than organizing data. So this is the description of the past.

Then the diagnostic analytics means do model building because you have to analyze the data

and then you have to look for the interaction between the data. And then predictive analytics

is nothing else than forecasting, which means you have to think about how will the dynamics go on

or what will happen next. So that then finally at the end of the day, prescriptive analytics means

to find an optimal action plan. And you see this is completely the same story what we have done

before, but it's written down in other wordings. And therefore some people have the feelings,

oh, we do something new, we do something different. No, it's not. The underlying mathematics is

exactly continuation from what we have done before with neural network stuff. And so there

you can discuss a lot of sub questions which I will avoid in the moment. Nevertheless, my

focus on the whole story still is predictive analytics, which means the forecasting thing.

Another way to operate with all of this is we speak about data-driven

modeling, which means we say in the beginning we have perceptions and there we have data.

And another way to think about all of this is simulation. And in large companies,

maybe from our Siemens or wherever, there might be a competition between this because all these

people want to get money to do research projects. And then therefore sometimes it looks like that

simulation and data-driven modeling is a competition. But in my eyes, it's not. Why?

They have a complete different starting point. If you do simulation, you think that you have

a microscopic understanding of the system that you want to model, maybe physical equations or

whatever it is. And then from the knowledge of the microscopic understanding, you want to compute

the microscopic behavior. So even if you have the weather equations, navier-stoke equations,

partial differential equations, then these equations suggest that you have a microscopic

understanding of what's going on in a low locality. And from this, you want to

extend to the whole weather behavior in a large country.

In data-driven modeling, you start with a macroscopic observation, which means

the interaction of all the variables together of a system. And then you want to

reconstruct the underlying microscopic causality of it by finding a neural network doing so.

And so the point here is, if you look at simulation versus data-driven modeling, you see

it's absolutely a complete opposite view on the world, what you are doing there.

For example, if you have the traffic there on a highway, you can say, I can do it by data-driven

modeling simply by measuring how many cars are driving there per minute. And then I do a forecast,

or you can say, cars can drive in this and this way. And now I do a forecast about the stop and

go behavior, what's going on there. Now there are applications where you can do both sides,

but in principle, they have a complete opposite view on the world.

Now, instead of arguing, I want to have to research money instead of you should get it,

the other department. One should think about, are there chances to combine both concepts here,

and so that the overall concept is much better. Now, if you build up simulation models, which

means assuming that you know what's going on microscopically and then producing a global

computation of it, then there still will be errors in it. And one possibility in combining both sides

would be to say, let's do the simulation and then let's take the residual errors, which I could not