MIT researchers have launched the AI Risk Repository, a comprehensive database that documents over 700 risks associated with AI systems.

This repository employs a two-dimensional classification system that categorizes risks by their causes and into seven distinct domains, including discrimination, privacy, and misinformation.

Project lead Peter Slattery emphasized that existing classifications of AI risks were fragmented and incomplete, highlighting the need for a comprehensive overview.

The efforts to document AI risks have been largely uncoordinated, leading to a pressing need for a unified classification system to effectively address these challenges.

Researchers found that existing AI risk frameworks overlook approximately 30% of identified risks, indicating significant gaps in current understanding.

Adopting AI technology carries numerous risks, including bias, misinformation, addiction, and the potential for misuse in creating biological or chemical weapons.

Most identified risks emerge only after AI models are publicly accessible, with only 10% detected prior to deployment.

Despite the creation of the repository, there are doubts about its adoption and whether it will effectively influence AI regulation, which is currently inconsistent globally.

The initiative aims to assist decision-makers in government, research, and industry in understanding and assessing the evolving risks of AI technology.

The database is intended to evolve, inviting feedback and further research on under-researched risks and potential gaps in understanding.

The AI Risk Repository is publicly accessible and intended for use by organizations for risk assessment and mitigation in AI development and deployment.

The repository is intended to save time for researchers and policymakers by providing a thorough overview of AI risks instead of relying on fragmented literature.