Towards Crafting Beer with Artificial Intelligence

摘要

The art of brewing beer has a long tradition that dates back to the very dawn of civilization. While the brewing process has been automated to a great extent, the creation of new beer recipes remains the result of creativity and human expertise with only minor support from software to validate chemical constraints. We collected a dataset of 157,000 publicly available recipes from all over the world and created a transformer-based model to support the creative process in brewing by suggesting new beer recipe templates. As a proof of concept, we crafted the IPA "Deeper" along a recipe generated by our model. Over 100 international newspapers and radio stations have reported on the first AI-crafted beer from Switzerland over the past few months. For the first time, this paper reveals the underlying pipeline architecture of eight transformer networks trained end-to-end that made this remarkable success possible.