Explainable AI for the Sciences: Towards Novel Insights