At the beginning of the 21st century, the development of the Internet made massive amounts of data immediately available. Big Data technologies have made it possible to use these datasets in a meaningful way. Moreover, massively parallel computing power became affordable outside dedicated data centers. The path to practical application of neural networks was clear. AI had come a long way from its initial early vision.
Modern AI is data-driven and problem-solving oriented. It processes large amounts of data, describing how specific problems were solved in the past, and computes a so-called model. This model can then solve new, previously unknown problems of the same type. For example, a pharmaceutical AI can learn a model of the relationships between active ingredients and their effects from a large set of known formulations. It can then predict the effects of a new formulation, or even propose formulations with certain effects. Within the context of the problem and the data, artificial intelligence shows itself to be highly intelligent. Advancements in AI are particularly impressive in human language processing.
Modern language models can thematically categorize and summarize arbitrary texts, and they can even compose new texts based on a given topic. The historical dream of the artificial brain seems within reach if an AI can summarize thousands of descriptions of side effects by patients in a meaningful and grammatically correct way. This is a false impression, as the AI always stays within the space defined by the underlying data. It does, however, arrange them in ever-new patterns under enormous speed, which is often helpful.