This featured resource is from NIST: “Language of Trustworthy AI: An In-Depth Glossary of Terms”.
Work to produce it started in 2022, then it was published in 2023 and updated in August 2024. It’s not exactly new, but is fairly recent.
NIST states that “The goal of this common vocabulary is not to declare one specific meaning for identified terms, but to provide interested parties with a broader awareness of the multiple meanings of commonly used terms within the interdisciplinary field of Trustworthy and Responsible AI.”
Each of the 447 terms has one or more definitions. These are from the OECD, IEEE, ISO standards, and a range of other sources. A handful of terms, including accuracy, have five definitions each.
There’s even a couple of definitions of fraud detection, including this one: “Monitoring the behavior of populations of users in order to estimate, detect, or avoid undesirable behavior”. The glossary provides the full citation for this and the other definitions.
There’s also a doc outlining the full background and motivation for creating the glossary.
Of course, a web search or chat thread might be the simplest way to find definitions. But we know that those don’t always produce the same results each time, so it’s useful to have a reliable place to go to. In practice, I might go here first. Then perhaps supplement with a search if a definition is unclear, or just to see if there’s anything newer.
Disclaimer: The information in this article does not constitute legal advice. It may not be relevant to your circumstances. It was written for specific algorithmic contexts within banks and insurance companies, may not apply to other contexts, and may not be relevant to other types of organisations.