Candidate AI and ML Policy

Abstract

There are many potential applications of AI in DANS and in its ecosystem. With this in mind, DANS has developed a draft AI Policy that aligns with recently published EU legislation [3]. Until a few years ago, AI was difficult to implement and that limited its utility. The main challenge with AI is not its implementation any more, but its governance. Therefore, we need to deploy AI in a manner that supports the main governance requirement of EU legislation for our typical applications: transparency. This extrapolates to elements such as attribution, provenance, and open decision making. Moreover, there are additional concerns about the negative impacts of AI use, relating inter alia to accuracy, skills erosion, and the ability to identify content generated by AI services. The policy defined here is informed by a landscape assessment [0] developed by DANS in 2023 (and reproduced in Annexure C). Due to rapid changes in the landscape, this assessment and the policy may require frequent (6 monthly) revisions. Annexure A contains a definition of roles and high-level responsibilities. A draft policy is presented in Annexure B, and the draft policy is mapped to the FAIRCORE4EOSC Compliance Assessment Toolkit (CAT) in Annexure D1.

Ähnliche Arbeiten