Artificial Intelligence (Al) has the potential to offer huge improvements both to the performance and the efficiency of systems. When engineering any system, it is important to ensure that the user of the system can be confident that the behaviour and performance of that system is as needed. Al presents challenges to achieveing this confidence when introduced into a system. This paper broadly considers these challenges, including the new failure modes exhibited by Al, the unpredictable and evolving behaviour of Al systems and the inability to explicitly validate due the level of trust in the systems and its robustness to future environments. Traditional system assurance techniques have proved ineffective to address these challenges. The paper then looks at potential solutions that are currently being developed that have the potential to address this need and assist in the assurance of the performance of Al enabled systems throughout their lifecycle.
展开▼
