"68% of machine learning papers didn't mention societal need and only 4% provided a rigorous justification for how the work addressed a social problem" https://arxiv.org/abs/2106.15590 (via @jackstilgoe on Twitter)