You could also introduce some checks and balances like having a small representative committee from various parts of communities exercise judgement on the judgment of the AI agent to ensure it doesn’t do anything radical.