Probabilistic Relational Learning and Inductive Logic Programming at a Global Scale

摘要

Building on advances in statistical-relational AI and the Semantic Web, this talk outlined how to create knowledge, how to evaluate knowledge that has been published, and how to go beyond the sum of human knowledge. If there is some claim of truth, it is reasonable to ask what evidence there is for that claim, and to not believe claims that do not provide evidence. Thus we need to publish data that can provide evidence. Given such data, we can also learn from it. This talk outlines how publishing ontologies, data, and probabilistic hypotheses/theories can let us base beliefs on evidence, and how the resulting world-wide mind can go beyond the aggregation of human knowledge. Much of the world's data is relational, and we want to make probabilistic predictions in order to make rational decisions. Thus probabilistic relational learning and inductive logic programming need to be a foundation of the semantic web. This talk overviewed the technology behind this vision and the considerable technical and social problem that remain.