The practice of governing agent artificial intelligence systems

Agent AI techniques—synthetic intelligence techniques that may pursue complicated targets with restricted direct supervision—may very well be broadly helpful if we might responsibly combine them into our society. Whereas such techniques have nice potential to assist individuals obtain their targets extra effectively and successfully, in addition they carry dangers of hurt. On this white paper, we suggest a definition of agent AI techniques and events within the agent AI system life cycle, and spotlight the significance of agreeing on a baseline set of duties and safety finest practices for all events. As our major contribution, we offer an preliminary set of practices to make sure brokers function safely and responsibly, which we hope will develop into the constructing blocks for the event of agreed baseline finest practices. We enumerate the questions and uncertainties that should be addressed earlier than implementing these practices. We then concentrate on the classes of oblique impacts arising from the large-scale adoption of agent AI techniques, which can require further governance frameworks.