A research agenda that organises machine-learning safety into four areas — robustness, monitoring, alignment and systemic safety — with concrete open problems in each.
This is an external publication hosted at arxiv.org. AI Governance LK did not write it; the summary and notes on this page are our own.
Why it matters
This paper is one of the clearest maps of technical safety research. It groups the work into withstanding hazards (robustness), identifying hazards (monitoring), reducing inherent model hazards (alignment), and reducing systemic hazards. For newcomers looking for a research direction, the per-area lists of open problems are especially useful.
How to read it
Treat it as a reference. Read the overview, then the section closest to your interests, and follow its citations.
What you will learn
- Use the four-area framing to place new papers in context
- Identify open problems suitable for a first research project
Before you start
Suggested background. This is editorial guidance, not a requirement.
- Introductory machine learning
- Concrete Problems in AI Safety (recommended)
How to cite
Hendrycks, D., Carlini, N., Schulman, J., & Steinhardt, J. (2021). Unsolved Problems in ML Safety. arXiv:2109.13916.