Artificial Intelligence

Production plants or machines often go through regular maintenance cycles. With the help of sensor-based AI systems, these cycles can be optimized. On the basis of automatic classifications, wear and tear or incipient defects are predicted or detected at an early stage. Machines are thus only maintained when necessary – not proactively. Downtimes are minimized.

The digital age makes it possible to almost completely monitor the supplier market for individual products. For example, all current prices for electronic items can be identified via price search engines. If the big players adjust their prices, it can be observed that smaller retailers react to these price adjustments within hours. The goal: to be taken into account by the consumer or, if market prices are too low, to deliberately avoid demand.

This is possible for a large number of items, as identical items are easily identifiable via product numbers. However, for items such as clothing, this is difficult to realize. While similarity can be easily assessed by humans, it is difficult for simple algorithms. Artificial intelligence can help here. By automatically comparing item properties and images, similar items can be identified.

Another typical use case for artificial intelligence is the subdivision of customers into different customer segments. AI is significantly more powerful than the classic ABC analysis, in which customers are simply divided into segments according to a central metric.

A considerable number of key figures or customer characteristics can be included in the analysis with the help of artificial intelligence, and additional information can be provided for a suitable or targeted approach. Ideally, the result of the customer segmentation is automatically fed into the campaign management tool and updated when new contact is made with the customer. This means that the campaign still has to be designed by the marketing manager and assigned to a customer segment. The customer approach is thus automated.

Artificial intelligence can help to predict the ability of customers to pay. Data on customers who did not comply with their payment requests in the past can be used as a basis. Based on this data, AI-supported algorithms predict the willingness of new customers to pay. This information can then be used in the online store, for example, to automatically restrict the payment terms of the new customer.

Today, information is often available in keywords or individual data points. For example, stock market prices are simple lists of prices at specific points in time, news agencies, provide news as collections of keywords or articles are initially provided by the manufacturer as descriptions for individual properties, such as color or material.

Artificial intelligence enables the generation of complete continuous texts from individual terms that are easy for the user to understand. Web pages on the development of stock market values, as well as summaries of soccer matches or detailed product descriptions, are already generated today using appropriate methods. They are only reviewed by humans before they are made available to the consumer.

Customer care, or helping the customer make a purchase, is the most common AI application. Today, it is possible to analyze how a prospect or customer interacts with a company across all touchpoints.

This information is used to specifically serve or address customers as well as to uncover and improve the weak points in the customer interaction. Customer satisfaction thus increases significantly.

In addition, transparency about customers and prospects is created, which strategically creates far-reaching added value throughout the entire customer journey​​​​​​​.

Although the term " Artificial Intelligence" is quite old, it is not clearly defined. This is due to the fact that the word intelligence itself is not clearly delineated.

Usually, Artificial Intelligence is understood to mean a system that replicates human behavior. In the most simple case, Artificial Intelligence can be based on a rule system that implements predefined decision rules.

Artificial Intelligence can also include systems that independently monitor the environment, develop behavior based on the monitored data, observe the reaction of the environment to its own behavior, and make decisions again for future situations based on the observations.

Today, Artificial Intelligence usually subsumes systems that use machine learning to recognize (learn) patterns on historical data and apply what they have learned to current data. A current trend is not to learn patterns once and apply them until what is learned becomes outdated, but rather to learn continuously and adapt the model to new data on an ongoing basis. Machine learning identifies and abstracts patterns and regularities in the learning data.

Deep Learning is a subfield of Machine Learning. Deep Learning refers to modeling approaches with artificial neural networks that have at least one intermediate layer called a "hidden layer" in addition to an input and output layer. More complex models, especially in the area of image recognition or text processing, have architectures with dozens of such hidden layers. In addition, there are architectures deviating from feed-forward networks with more complex interconnections between individual layers and knots.

Artificial Intelligence as a term was shaped in the 1950s. Due to the rapidly increasing processing speed of computer systems at that time, people were sure that they would be able to develop algorithms within a short period of time that would eventually prove to be strong AI systems. After initial enthusiasm for this topic, the lack of success quickly led to disillusionment.

The following epoch with little interest on the part of the research community is generally referred to as the "AI winter".

It was not until the increasing processing speed of computers, improved software and the ubiquity of data that the topic of artificial intelligence experienced a renaissance around the turn of the millennium.

The learning of Artificial Intelligence begins in principle with the Data Scientist's understanding of the Data Science problem to be solved and the ability to recognize data that contribute to the solution of the problem.

Required data is obtained by the Data Scientist, cleaned and aggregated if necessary. Subsequently, a model is created. Usually, training and validation is performed on historical data used for learning the patterns and relationships.

In the training step, the model parameters are adjusted in such a way that the results of a training data set can be optimally predicted. Subsequently, the model is validated using a test data set whose data was not used to train the model. This ensures that the model will produce valid results when new data is used in the future.

If the quality of the model is satisfactory, the AI solution is ready to be deployed and can be provided as a service or integrated into dashboards. If the quality is not sufficient, the process starts again. If necessary, additional data or other models are used or the question is adapted.

In short: The learning of an Artificial Intelligence is the interaction of human intelligence, a lot of data and the selection of a suitable algorithm.

Artificial intelligence is essentially about imitating human behavior. The focus is on replicating human perception, learning from historical data and imitating human decisions

Recognition

One of the strengths of humans is their ability to quickly separate the important from the unimportant. For example, people immediately recognize other people in pictures and can quickly assign them to people they know. A subfield of Artificial Intelligence is concerned with replicating this behavior in various areas. The focus is on image, speech and text processing.

Understanding and learning

The strengths in human perception are the result of "lifelong learning". How this works is not yet fully understood. Artificial Intelligence follows different approaches to replicate this, such as machine learning, reinforcement learning or even crowd sourcing.

The goal is to use past experiences to recognize patterns with a computer-based system and make them usable for the future. AI systems learn through feedback loops during their use.

Acting

If appropriate patterns have been learned, they can be used to act in future decision-making situations. A distinction is made here as to how strongly behavior is intervened in.

With predictive analytics, the human decision-maker is presented with data and figures based on which he or she can make decisions. Expert systems go one step further. Here, Artificial Intelligence acts as an advisor, making a proposal for the decision that the human must accept or reject. In robotics, artificial intelligence makes simple decisions without human intervention.