Methods of artificial intelligence (AI) are increasingly present in our daily lives. Until recently, only computer science experts dealt with them, but now almost everyone encounters them regularly. This article describes the development of the field of AI and outlines how modern AI methods actually work. We present an overview of the historical development of AI, which reveals cyclical phases of high expectations and disappointments in the field. We explain the basic AI principle, which captures human knowledge through input data and parameters. The model, known as the universal approximator f(x, p), seeks optimal parameter values via an optimisation process based on a large number of training examples and manual labels. Such trained models form the foundations for practical AI applications. We also present an example of »manual« learning with a discriminative model that employs a linear boundary, i.e., a straight line, where our understanding of the problem is stored in just two parameters.
|
