Konzeptionelle Grundlagen der Künstlichen Intelligenz: KI und Naturwissenschaften

(English): In the first textbook of modern physics, "Mathematical Principles of Natural Philosophy" (Principia Mathematica Philosophiae Naturalis), Newton describes the methods of natural research (regulae philosophandi) that still determine research practice today: Natural research should first "inductively" derive causal relationships from observation and measurement data and formulate them in mathematical laws, from which explanations and predictions are then "deductively" derived (1). The deductive method was formalised in mathematical logic at the beginning of the 20th century and became the basis in the first AI phase of automatic reasoning and expert systems ("symbolic AI") (2). Today, the inductive method is used in AI based on statistical learning theory to discover data correlations and is the basis of modern machine learning. However, statistical correlations do not replace causal relationships (3). Thus, practical security and verification issues of computer programmes are closely related. Statistical methods play a central role in the modern natural sciences, from physics and chemistry to biology and the life sciences. However, AI is not only applied in the natural sciences. Conversely, methods of mathematical physics are now used to understand and computationally master the vast neural networks (Deeper Learning) in science and technical practice (4). AI and the natural sciences are thus methodically growing together in every respect.