Research Challenges
Dependability is a broad system property that encompasses attributes such as reliability and safety. Reliability refers to a system’s ability to function continuously without failure, while safety concerns its ability to avoid hazardous failures that could cause harm to people or the environment.
Complexity
Modern systems are expanding in scale and include vast numbers of interacting components, extensive software, and high levels of automation. This complexity introduces new failure modes and hazards that are difficult to predict and control.
Intelligence
Intelligent systems, particularly those based on Machine Learning (ML), represent a paradigm shift. These systems are not explicitly “designed” in the classical sense as they learn behaviour from large datasets. For instance, an ML algorithm may learn to detect pedestrians by analysing labelled images. While this capability is revolutionary, the learning process is inherently imperfect, and failures can occur in unpredictable ways. Crucially, traditional dependability methods which rely on explicit models of system design and they cannot be applied because no such models exist for ML-based components.
Autonomy
Autonomous systems operate without human intervention. If they encounter unforeseen situations or hazards, there may be no human operator present to intervene or correct course. This raises the stakes for reliable and safe decision-making.
Openness of Cooperative SoS
Many modern systems operate as part of cooperative Systems of Systems, communicating with other devices and infrastructures (e.g., vehicle-to-vehicle or vehicle-to-infrastructure communication). These systems form dynamic, unpredictable configurations in real-world conditions and configurations that cannot be fully anticipated or tested in advance. Yet, traditional dependability assurance is largely performed before deployment.
CIAO encapsulates the key dependability challenges for the next generation of intelligent systems. Without addressing these challenges, safe deployment of such technologies cannot be guaranteed.
