MIT Researchers Share AI Risk Database

In response to the growing concerns surrounding artificial intelligence (AI) and its potential risks, researchers at the Massachusetts Institute of Technology (MIT) have launched an AI "risk repository." This initiative aims to create a comprehensive and publicly accessible database that categorizes and analyzes various AI risks, particularly those associated with critical infrastructure. Peter Slattery, a lead researcher on the project, emphasized the need for such a resource, noting that many stakeholders in the AI industry and academia require a structured approach to understanding these risks. The repository highlights significant gaps in existing frameworks, revealing that many only address a small fraction of the identified risks. According to Slattery, the average frameworks cover just 34% of the 23 risk subdomains recognized by the researchers, with some frameworks addressing less than 20%. This fragmentation in the literature raises concerns about the collective understanding of AI risks, particularly as over half of the frameworks focus on discrimination and misrepresentation, while only a small percentage address issues like the "pollution of the information ecosystem" caused by AI-generated spam. Slattery believes that the new database can serve as a foundational tool for researchers and policymakers, enabling them to build upon existing knowledge and conduct more targeted investigations into AI risks. By providing a centralized resource, the repository aims to streamline the review process and enhance oversight in AI development and usage. However, questions remain about the repository's adoption and its potential impact on the AI landscape had it been available earlier.