How well does machine learning extract simple logic ???