New framework clarifies how AI systems should handle uncertainty in the real world
Researchers have created the first unified model for how autonomous systems—from self-driving cars to industrial robots—should represent and manage different types of uncertainty. The work addresses a critical gap in AI reliability: existing approaches use inconsistent terminology and fail to distinguish between software, hardware, and autonomous decision failures, making it harder for companies to build trustworthy systems.
Originaltitel: Surveying uncertainty representation: a unified model for cyber-physical systems
<p>Cyber-Physical Systems (CPS) operate in dynamic environments, leading to different types of uncertainty that affect their design, operation, and reliability. This work provides a comprehensive review of uncertainty representations and categorizes them based on the dimensions used to represent uncertainty. Through this categorization, key gaps and limitations in existing approaches are identified, such as inconsistent terminology, the lack of systematic differentiation between CPS components, and the absence of explicit consideration of autonomy in CPS. To address these issues, a Conceptual Model of Uncertainty Representations in CPS is introduced, which unifies the terminology used in existing frameworks while introducing missing categories specifically tailored to CPS. Our model incorporates distinctions between cyber, physical, and platform components, as well as between autonomous and non-autonomous subsystems, offering a more precise characterization of uncertainty. Its applicability is demonstrated through examples from the automotive domain, showing its effectiveness in capturing and structuring uncertainty in real-world scenarios. This contribution not only harmonizes existing approaches but also establishes a foundation for future research on expressive and domain-aware representations of uncertainty in CPS.</p>