Ideological Inference Engines: Making Deontology Differentiable* — LessWrong