Adel, T; McCrory, M; Tynan, C; Thompson, A; Duncan, P (2025) Characterising and Testing the Trustworthiness of Artificial Intelligence Systems. NPL Report. MS 62

Preview

Text (NPL Report MS 62) MS62.pdf - Published Version Download (321kB) | Preview

Abstract

In many applications, decision making has recently become dependent on artificial intelligence (AI) systems. In order to ensure a safe integration of such systems within these applications, not only should their accuracy and performance be tested, but also their trustworthiness. We discuss here the basic phases involved when testing the trustworthiness of an AI system, as well as some of the steps that can be taken to ensure an AI system is trustworthy. We begin by discussing characteristics which should be considered for most AI systems, prior to moving on to other characteristics of trustworthiness which can be essential for some AI systems, particularly those which are sensitive and have a direct impacton people’s lives. We also shed light on the fact that trustworthiness, along with its evaluation, should be fit for purpose and should be aligned with the original context in which the respective AI system will be deployed. We also examine the role of third-party testing in the development and deployment of AI and ML systems, outlining some of the related benefits, risks, and best practices for mitigating these risks.

Item Type:	Report/Guide (NPL Report)
NPL Report No.:	MS 62
Subjects:	Mathematics and Scientific Computing > Modelling
Divisions:	Data Science
Identification number/DOI:	10.47120/npl.MS62
Last Modified:	22 Aug 2025 09:52
URI:	https://eprintspublications.npl.co.uk/id/eprint/10216