Right now, many questions are being asked about artificial intelligence (AI). The topic is hyped and makes daily headlines in the media. We’ve been working with AI since the late 1980s.
First, we asked fundamental questions about semantics and a possible architecture. In the early 1990s, we then considered possible applications on the LISP machines that were designed for AI development at the time. Even in the early stages, there were various definitions of AI and thus also different approaches. On the one hand, for example, attempts were made to describe the basic principles of AI using dialectical laws (Hegel and Engels); other approaches were based on mathematical and cybernetic models. The crucial question, however, was the possible self-awareness of a machine-based AI system. Simply put, we asked: Can a machine learn?
This question is still bitterly debated today. It begins with the question: What does learning mean? Currently, AI systems are based on data collections. This can be texts and/or links between mathematical and textual relationships. Image information can also represent a data collection.
Today’s AI systems are rule-based. These systems are usually based on “if-then” relationships or pattern matching. An AI system can perform tasks and classify the result – again according to corresponding rules – as successful or unsuccessful. Good AI systems can refine and adapt their own rules based on the results. However, this is not yet learning, as the “knowledge” component is missing. According to Duden, knowledge is defined as follows: “Knowledge is an insight gained through mental processing of impressions and experiences.” However, AI systems cannot yet generate “insight.” Because this component is (still) missing, they are also referred to as “simple AI.” Cross-connections to simulations based on mathematical models are not entirely wrong.