Peter Eckersley, “Impossibility and Uncertainty Theorems in AI Value Alignment (or why your AGI should not have a utility function)”, (Poster at NIPS 2017, Paper at SafeAI 2019): https://arxiv.org/abs/1901.00064