Concerns about AI safety are growing. Two researchers explain the unintended consequences of training AI models and what questions remain.