Toward Implementable AI Standards
Keywords: Ai standards, algotrithms, training data, pre-release testing, post-release evaluation
TL;DR: This paper makes the case for standard-based governance and identifies four major elements that must be part of any AI standard
Abstract: The U.S. government has proposed a standards-based approach to AI governance, with the precise contours of that standard to be developed over time. This article lays out the case for a standards-based approach and identifies four major elements that must be part of any AI standard. It also emphasizes that any AI standard incorporate a notion of optimality built around cost-benefit analysis and some measure of acceptable risk.
Submission Number: 6
Loading