Artificial Intelligence and the Control Deficit: On the Growing Gap Between Capability and Governance

This paper, Artificial Intelligence and the Control Deficit: On the Growing Gap Between Capability and Governance, examines the structural divergence between the rapid advancement of artificial intelligence systems and the comparatively slower development of mechanisms for their understanding, control, and governance.

The work introduces the concept of a capability–control gap, defined as the widening mismatch between AI capability growth—driven by scaling laws, increased compute, larger datasets, and model architectures—and control capacity, which depends on interpretability, alignment research, and institutional regulation.

Through a technical and conceptual synthesis, the paper explores three interrelated dynamics: (1) the scaling mechanisms underlying modern AI systems, particularly transformer-based architectures and optimization processes; (2) the fragility of control in high-capacity systems, including challenges in AI alignment and interpretability; and (3) the governance lag, where institutional and regulatory frameworks struggle to match the pace of technological change.

The paper argues that this gap is not temporary but structural, emerging from fundamentally different scaling properties of capability and control. While capability scales efficiently through computation and data, control scales slowly through understanding, specification, and governance adaptation.

Overall, the paper frames the capability–control gap as a central challenge for the safe and sustainable development of advanced AI systems, with implications for both technical research and global policy design.