Unsolved Problems in ML Safety
It aims to guide research priorities for ensuring safety in ML systems, particularly as they grow in scale and are deployed in high-stakes settings, but it is incremental as it builds on existing safety concerns without introducing new methods.
The paper addresses emerging safety challenges in machine learning by providing a new roadmap and refining technical problems for the field to tackle, focusing on four key areas: robustness, monitoring, alignment, and systemic safety.
Machine learning (ML) systems are rapidly increasing in size, are acquiring new capabilities, and are increasingly deployed in high-stakes settings. As with other powerful technologies, safety for ML should be a leading research priority. In response to emerging safety challenges in ML, such as those introduced by recent large-scale models, we provide a new roadmap for ML Safety and refine the technical problems that the field needs to address. We present four problems ready for research, namely withstanding hazards ("Robustness"), identifying hazards ("Monitoring"), reducing inherent model hazards ("Alignment"), and reducing systemic hazards ("Systemic Safety"). Throughout, we clarify each problem's motivation and provide concrete research directions.