This paper argues that the world is being taken over by AI as a system that models pixels, words, and phonemes, and that the world is not made up of pixels, words, and phonemes, but of entities (including objects, things, and events) with properties and relationships. Therefore, it argues that we should model these entities rather than perception or description. The reason we currently focus on modeling words and pixels is that the world's valuable data exists in the form of text and images, but we emphasize that most companies' most important data is stored in relational formats such as spreadsheets and databases, which are different from the forms dealt with in existing machine learning. We explain why this field, which is called by various names such as relational learning and statistical relational AI, has not taken over the world except in a few cases with limited relationships, and what steps are needed to increase its importance.
