Add On: AI
​

Artificial intelligence has already found its way into many areas and plays an important role in our everyday life. An understanding and knowledge of machine learning and algorithms enables pupils to develop a responsible approach and an independent attitude to the topic. 
The TXT 4.0 makes it very easy for pupils to take their first steps with artificial intelligence (AI). They find out what is behind it and how they can use it for their own models.

The signals arrive on the left-hand side (input layer) – this is comparable with our sensory cells. The sensory cells are connected to neurons, and the strength of the connection indicates how strongly a neuron reacts to the stimuli from the left.   

On the right-hand side (output layer) we can see the output neurons – put simply, these stand for certain terms which were perceived by the senses on the left-hand side. If we “see” an image on the left-hand side, for example, the term “cat” could be associated on the right.

Between the sensory cells on the left and the output neurons on the right, there are structures made up of many neurons. These are usually arranged in individual “layers” and every layer has a certain task or function. 

The division of the nodes of an artificial neuronal network into layers is for reasons of clarity. We talk about the input layer, hidden layer and output layer.

• The input layer picks up the data from the environment.
• The output layer delivers the data to the outside again.
• Hidden layers usually connect the input layer with the output layer. There can be any number of these layers, or none at all.

There are different types of links for different tasks. Certain links are suitable for recognising parts in images, and others are good at recognising sequences in time, such as words in a sentence.

Terms such as deep learning, deep mind or the program DeepL, which translates text to another language, refer to the structure of the neuronal network – a so-called “deep” network. The deep. Deep simply means that there are many layers in succession, with each layer reacting to the outputs of the previous layer.

In simple terms, learning in a neuronal network works by stimuli e.g. images being set up on the left-hand side and then seeing whether the right term arises on the right-hand side. If this is not the case, the strength of the connections is adapted from right to left and a new attempt is made; as long as necessary until the network reacts correctly. 

The constant processing of a large network is very time-consuming. This is why tasks for AI are trained at the PC and not on the TXT 4.0. 

Work on artificial intelligence has been carried out for many decades. For a long time, however, computers and networks were not powerful enough to train large neuronal networks. For a network to be trained from scratch so that it can recognise objects, millions of images are required, which the network processes constantly from left to right and then corrects again from right to left. 

For AI to be able to do anything, it has to learn first and then apply what it has learned later. The basic principle always works according to the following three steps: 

1. Collection of training data – examples are collected which the AI uses for learning later. This can be e.g. images of different animals or objects. We then have to name these images so that the AI knows what it has to learn. 

2. Training the AI – in this step we show the AI our examples and what we mean by them. Then the AI adapts in such a way that it is able to find the respective term to a given image as well as possible. 

3. Use of AI – once training has been completed, we can use the network and try to recognise new or other images. 

We will quickly realise that the training is not as easy as you might think. You need the right images and plenty of them, and the light conditions have to be right. But this should be found out by exploring and experimenting with the models themselves. 

To stop training of the AI being too difficult, we use a pre-trained AI as a basis with our examples. This is able to recognise objects and only has to be taught the “names” of the things it is expected to recognise. This type of learning is called transfer learning.