When Do Data-Driven Systems Exhibit the Capability to Infer?

The article presents a framework for assessing the capability of data-driven systems to infer, particularly in the context of the European AI Act, which regulates high-risk AI systems. It highlights the ambiguity surrounding the definition of inference and applies the framework to credit scoring systems, demonstrating how inference can be evaluated across entire data processing workflows rather than just individual models. This work is significant for practitioners as it clarifies regulatory compliance and informs the development of AI systems that meet the legal requirements for inference capabilities.