Abstract
With the rapid adoption of Machine Learning (ML) in Computing, there has been a flurry of recent research considering using ML to build the internal component of database systems. While initial work in this area has shown interesting results, the jury is still out on whether these methods will replace existing methods. A group of researchers with opinions on both sides of this issue met to assess the state of this area and to formulate a plan for the next steps that would be needed to determine the potential role of these new ML-based methods in building future database systems. This article summarizes the collective perspectives that resulted from these discussions. First, this article describes broad forces that are changing the landscape in which database systems are deployed, connecting several trends that likely require rethinking how future database engines are built. Next, this article describes the different perspectives on this topic of using ML methods to replace existing internal database components. Finally, the key takeaways from this discussion are presented, and these takeaways also point to directions for future research.