However, the researchers noted in the FAQ that the Repository has several limitations, including being limited to risks from 43 taxonomies, so it “may miss emerging, domain-specific, and unpublished risks, and has the potential for errors and subject bias ; we used one expert reviewer to extract and write the codes.”
Despite those flaws, the MIT Technology Review article said the findings “could have implications for how we evaluate AI,” and contained the following from Neil Thompson, director of MIT FutureTech and one of the website’s creators: “What (it) says , the range of risks is large, not all of them can be detected in advance.”
Live work
In this illustration, Thompson and other project participants wrote that the Repository is, “to our knowledge, the first attempt to rigorously process, analyze, and extract AI risk frameworks into a publicly accessible, comprehensive, extensible, and hierarchical risk database. This creates the basis for an integrated approach , consistent, and comprehensive approach to defining, assessing and managing the risks posed by AI systems.”
Source link