For reward hacking, see, e.g., Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman & Dan Mané, Concrete Problems in AI safety, arXiv: 1606.06565 [cs.AI] (2016).

Principle of safety

● To make efforts to conduct verification and validation in advance in order to assess and mitigate the risks related to the safety of the AI systems.

● To make efforts to implement measures, throughout the development stage of AI systems to the extent possible in light of the characteristics of the technologies to be adopted, to contribute to the intrinsic safety (reduction of essential risk factors such as kinetic energy of actuators) and the functional safety (mitigation of risks by operation of additional control devices such as automatic braking) when AI systems work with actuators or other devices.

● To make efforts to implement measures, throughout the development stage of AI systems to the extent possible in light of the characteristics of the technologies to be adopted, to contribute to the intrinsic safety (reduction of essential risk factors such as kinetic energy of actuators) and the functional safety (mitigation of risks by operation of additional control devices such as automatic braking) when AI systems work with actuators or other devices.

● To make efforts to explain the designers’ intent of AI systems and the reasons for it to stakeholders such as users, when developing AI systems to be used for making judgments regarding the safety of life, body, or property of users and third parties (for example, such judgments that prioritizes life, body, property to be protected at the time of an accident of a robot equipped with AI).