OpenAI is working on technical safety mechanisms to ensure that systems built align with human values, by looking at data and building proof of concept models that can learn human preferences.