What is Artificial Intelligence?

Artificial intelligence (AI for short) is an exciting and fast-growing technology. It is already influencing our lives in many ways. Artificial intelligence is in the voice assistantthat sits in the living room. The streaming service streaming service uses AI to suggest suitable movies and series. And the navigation system in the car calculates the best way to avoid the next traffic jam while the car is still moving. To understand how all this works, we explain the basics on this page: What is AI? What is an algorithm? What is machine learning? And what types of AI are there?

On the subpages of the topic area, we then devote ourselves to more specific questions about AI. Everyone is talking about generative AI. These are systems that can create texts, images, videos and music or speech. We explain what generative AI can already do and how it works. We also look at how AI can be used in schools schools and what challenges lie ahead for the protection of youth media protection media protection. And we discuss which risks AI systems can entail.

To get a better idea of what an algorithm is, a comparison is often made with a baking recipe. When baking a cake, you need certain ingredients in a certain quantity. These are then processed together in a specified order.

The basic idea of an algorithm is that certain data ("ingredients"), are processed according to predetermined mathematical rules ("recipe") to achieve a specific result ("cake").

In order for a computer to read such algorithms, they must be described in a specific form - this is called a program or software.

The terms "algorithm" and "artificial intelligence" are often used interchangeably, but there are differences. In general, it can be said that algorithms are the basic building blocks of AI systems. With their help, they process data, recognize patterns and correlations, and derive decisions from them. But not all algorithms are suitable for use in AI systems. They must have certain properties.

There is no generally valid definition of the term. However, one can look at the meaning of the two words that make up the term. "Artificial" means that it is a system artificially produced by humans, i.e., a machine. This machine is used to try to simulate "intelligence" - that is, certain human cognitive abilities.

The term, " artificial intelligence" (AI for short), was coined in the run-up to a working meeting of scientists in the USA in 1955. They held the view that properties of human intelligence could be described in such a way that machines could read and execute them. Since then, AI technologies have developed as a subfield of computer science from this consideration.

The goal of AI technologies is to enable machines to mimic intelligent behavior. However, what exactly is to be understood as "real" intelligence is not clearly defined. It is generally understood to include abilities such as communicating, solving problems, thinking logically or being able to adapt to new situations. But people can also perform tasks better than humans, without necessarily requiring a high level of intelligence. A pocket calculator, for example, is far superior to humans in its ability to calculate, but is therefore not called "intelligent". And an AI can answer questions correctly, but understands neither the meaning of the question nor that of the answer.

Basically, there is no such thing as "one" AI. The term "artificial intelligence" covers a wide range of different methods and technologies. Depending on the task an AI is supposed to implement, the systems can be quite different in their function. However, AI systems can be divided into two types:

1. strong AI.

So-called "strong" AI (General AI)attempts to build computer systems that can mimic or even surpass the intellectual capabilities of humans. Such AI would be capable of understanding and accomplishing various complex tasks in multiple application domains. This type of AI does not exist yet. It is a vision of the future and not currently a focus of research. There are also discussions about whether such a system is even technically feasible.

2. weak AI

The focus in AI development is primarily on so-called "weak" AI (Narrow AI). In other words, systems that are designed for a clearly defined task area in order to solve a specifictask there. Weak AI systems do not possess intelligence or a comprehensive understanding of the world. Nevertheless, these systems can do amazing things.

That's because, thanks to high computing power, the processing of enormous amounts of data and sophisticated functions, AI systems can arrive at solutions much faster than a human can. In 1997, for example, an AI called Deep Blue (IBM), which was programmed to play chess, defeated Garry Kasparov, the world chess champion at the time. In another application area, however, this AI would have been useless.

A very important subarea of (weak) artificial intelligence is formed by so-called "learning" AI systems. They are also known under the term "machine learning". The machine learning approach is currently one of the most widely used methods in AI. Machine learning is an approach in which computer systems learn by example by identifying patterns and relationships in data and recording this in a statistical model. But how does it work?

Phase 1: Training and evaluation of the model

In our example, we want to train a system that reliably recognizes cats in pictures. The computer learns which input belongs to the appropriate output based on examples and human feedback. In our case, the input would be a picture of a cat and the output would be the statement "That's a cat!". As training data, the system needs a large set of cat images. Using this training data, the system infers relevant patterns and rules that are present in this data. For example, that cats have a tail, or that they come in different colors. In the end, a statistical model is created on this basis, i.e. a mathematical equation that contains all the relevant correlations and decision rules. Developers correct inaccuracies and errors in the model and evaluate its accuracy.

Phase 2: Decisions based on the model  

The model can then be used to process new data and make similar decisions and predictions. Some models are dynamic and can continue to learn beyond the testing phase. They draw conclusions from the new data and improve their performance and functions. But there are also static models that do not adapt. They are used when data does not change or changes very slowly.