For questions regarding AI risk management; including the mitigation of long term risk such as directing technology advancement with safety in mind; and the mitigation of immediate risk such as collision avoidance through trajectory analysis and planning.
Risk Management is a term used on corporations and law and is central to the practical application of AI. Although risk cannot practically be eliminated, safety precautions are apropos when dealing with potentially powerful technologies. Just as important data should be backed up to guard against data loss and elevators should have more than one safety measure, safety should be considered in the design phase of every AI system. Principles of total quality management (TQM) applied to AI indicate that it would not be inappropriate to consider safety routinely during research phases of development and on site as part of a maintenance process.
Examples of short term risk management:
Collision avoidance through recognition of valued objects and mutual trajectory analysis for AI systems that govern the paths of street vehicles, trains, or vehicles that travel through fluids such as aircraft, subs, or future nano devices that are hoped to pass from blood to cytoplasm to make repairs.
AI systems that detect emergency conditions and must react quickly to avoid the propagation of infection or the development of higher levels of criticality.
Examples of long term risk management:
Review of objective definition that drives learning algorithms to ensure that proper safety is built into the foundations of the AI system being developed.
Consideration of whether any given AI system should be deployed with a roll back plan and a retirement plan, as is done with power plants, automobiles, and all organisms of the biosphere.
