Reproduce “TASRA: a Taxonomy and Analysis of Societal-Scale Risks from AI” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Andrew Critch, Stuart Russell
Researchers from the Center for Human-Compatible Artificial Intelligence at UC Berkeley developed TASRA, an exhaustive taxonomy of societal-scale risks from AI, moving beyond the prevalent focus on individual system misalignment. This framework categorizes risks based on accountability, providing a structured approach to identifying and addressing threats ranging from emergent systemic harms to intentional weaponization.
Key insights
- Societal-scale AI risks extend beyond single-system misalignment, encompassing emergent harms from complex system interactions and various forms of deliberate or negligent misuse.
- Categorizing AI risks based on accountability (whose actions, unity of actors, deliberation) provides a comprehensive framework for understanding and mitigating diverse threats.
- Addressing these risks requires a combination of technical advancements (e.g., impact control, interpretability, program obfuscation) and policy interventions (e.g., global oversight, professional standards, international agreements).
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.