When the German government presented the first details of its AI strategy in mid-November, Peter Altmeier felt compelled to create a new term to describe the impact of the technology: Artificial intelligence is "not just any innovation", but "basic innovation that will change and improve our economy and our lives as a whole", said the Federal Minister of Economics. The artificial intelligence megatrend has also long since arrived in industry. And here too, the expectations are enormous. Bold forecasts are talking about additional growth of up to 10 billion euros per year and an increase in productivity of a third - the term basic innovation seems almost an understatement.

The term artificial intelligence is actually over 60 years old. There are various reasons why it is experiencing a renaissance right now: On the one hand, the massive increase in available computing power means that the necessary parallel data processing is becoming faster and cheaper, while at the same time the available storage space is also increasing massively thanks to largely freely scalable cloud computing. On the other hand, it has become much easier to collect machine data in the factory thanks to the inexpensive equipment of machinery with sensors. And finally, providers such as Siemens, IBM and GE now provide platforms with which AI applications can be integrated extremely easily thanks to the Open API interface. In short: AI has become suitable for the factory and is increasingly finding its way into the smart factory in one form or another.

Machine learning, AI and neural networks

At the same time, however, the approaches that often operate under the AI label are so broad and not very clear-cut that the aspirations and reality of industrial AI applications are often far apart. In order to clear up any misunderstandings, a distinction should therefore first be made between some central approaches:

Weak AI vs. strong AI: A key misconception about AI systems is that they have the same intellectual abilities as humans, such as the ability to think logically, make decisions even in the face of uncertainty or plan and learn. However, such a so-called "strong AI" is still only a vague vision whose practical implementation is not yet in sight. Today's AI applications can all be classified as "weak AI". They are geared towards the fulfillment of a clearly defined task and rely on learned methods without varying the approach to problems or gaining a deeper understanding of the problem solution. Typical fields of application include text, image or speech recognition, automated translations or expert systems that provide recommendations for action based on knowledge databases.

Machine learning: The terms AI, machine learning and deep learning are often mistakenly equated in everyday life. However, machine learning and deep learning are each only sub-areas of artificial intelligence or of each other. While AI as a generic term describes all methods and approaches for replicating human intelligence, machine learning refers to a collection of mathematical methods of pattern recognition that enable a system to independently generate knowledge from experience.

Deep learning: Deep learning, in turn, is a sub-area of machine learning using neural networks. They function along the lines of the human brain with several layers between the input and output of information. The advantage of artificial neural networks is the gradual abstraction of correlations in the intermediate layers. Instead of a deterministic algorithm, deep learning therefore relies on statistical data analysis and thus enables self-learning and self-optimizing systems. Almost all current AI applications use the methods of deep learning or neural networks.

In summary, today's AI applications can be described as self-learning methods of statistical data analysis that are able to recognize patterns within large amounts of data and derive recommendations for action.