Verification and Validation Methodology for AI
conference paper
Defence must continuously search for opportunities offered by new technology to improve its operational effect. Artificial Intelligence (AI) is a disruptive technology that offers a diverse set of such opportunities. The nature of many AI algorithms, however, is that they are essentially a black box. Safety concerns is one of the limiting factors for the use of AI in defence operations. Verification and validation (V&V) are processes that can be used to gain insights into the qualities and risks. But there are several important differences between V&V of traditional systems compared to V&V for AI. We are building V&V processes that cover the whole life-cycle of systems that include a data-driven AI component. The training phase of the human operators of AI applications, which indeed may include additional training of the AI as well, should result in suitable human-machine-team configurations. This training phase, is one of the key steps in the AI life-cycle. In this paper we describe the background and the V&V methodology we are developing.
TNO Identifier
992037
Source title
IT2EC, 2023, Rotterdam
Files
To receive the publication files, please send an e-mail request to TNO Repository.