Two Defeasible Challenges for algorithmic Scientific Discovery