We will introduce the concept of 'Semantic AI', which is based on a fusion of NLP, machine learning (ML) and semantic knowledge graphs. We will take a look at two concrete use cases: How to improve ML-based classifiers for legal documents with semantic enrichment. How to combine various entity extraction methods, partly based on ML, to identify complex rules and restrictions from legal documents. We will elaborate how traditional NLP/text mining methods can be outperformed when combining various AI algorithms and methodologies, and we will introduce additional use cases for the legal domain that could benefit from this approach.