NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·3h ago
AI Research
arXiv CS.AI·Jun 9·5 min read
RiskNet: A large-scale dataset of AI risk incidents from news with alignment and multi-dimensional annotations
TL;DR
As AI systems are increasingly used in critical areas, reports of AI-related risks have surged, highlighting a lack of comprehensive data for analysis. RiskNet was developed as a large-scale dataset that organizes AI risk incidents from multilingual news sources, featuring structured classifications and annotations.
✦ Why It Matters
Engineers and researchers can leverage RiskNet to enhance their understanding of AI risks and improve safety measures.
Key Takeaways
- Utilize RiskNet for empirical analysis of AI-related harms.
- Explore the dataset for longitudinal studies on AI safety.
- Implement findings to inform AI governance frameworks.
- Access structured incident classifications for better risk assessment.
Full Summary
The rise of artificial intelligence (AI) in socially significant domains has led to a growing number of reports detailing AI-related incidents, yet existing datasets for analyzing these events are often small and manually curated. RiskNet addresses this gap by providing a large-scale dataset of AI risk incidents sourced from extensive multilingual news articles.
The dataset employs a structured pipeline for identifying AI risk news, screening reports at the event level, aligning incidents, and classifying them across multiple dimensions. Currently, RiskNet encompasses hundreds of millions of source records, resulting in a comprehensive collection of AI risk reports, including incident clusters and annotated benchmark subsets.
It is accessible through an online platform for exploration and analysis. This resource is designed to facilitate research on AI safety, governance, and risk analysis, bridging the divide between theoretical governance principles and real-world AI risks.