E. Santi

Eduardo Di Santi

Artificial intelligence is increasingly embedded in human decision-making, where it can either enhance human reasoning or induce excessive cognitive dependence. This paper introduces a conceptual and mathematical framework for distinguishing cognitive amplification, in which AI improves hybrid human-AI performance while preserving human expertise, from cognitive delegation, in which reasoning is progressively outsourced to AI systems. To characterize these regimes, we define a set of operational metrics: the Cognitive Amplification Index (CAI*), the Dependency Ratio (D), the Human Reliance Index (HRI), and the Human Cognitive Drift Rate (HCDR). Together, these quantities provide a low-dimensional metric space for evaluating not only whether human-AI systems achieve genuine synergistic performance, but also whether such performance is cognitively sustainable for the human component over time. The framework highlights a central design tension in human-AI systems: maximizing short-term hybrid capability does not necessarily preserve long-term human cognitive competence. We therefore argue that human-AI systems should be designed under a cognitive sustainability constraint, such that gains in hybrid performance do not come at the cost of degradation in human expertise.

Eduardo Di Santi

Real-world phenomena do not generate arbitrary variability: their signals concentrate on compact, low-variability subsets of functional space, enabling rapid generalization from few examples. A small child can recognize a dog after extremely limited exposure because the perceptual manifold of "dog" is compact, structured, and low-dimensional. We formalize this principle through a deterministic functional-topological framework in which the set of valid realizations produced by a physical process forms a compact subset of a Banach space, endowed with stable invariants, a finite Hausdorff radius, and an induced continuous perceptual functional. This geometry provides explicit limits on knowledge, conditions for identifiability, and guarantees for generalization from sparse evidence -- properties fundamental to both natural and artificial intelligence. Across electromechanical, electrochemical, and physiological domains, we show that real-world processes consistently generate compact perceptual manifolds with the same geometric characteristics. Their boundaries can be discovered in a fully self-supervised manner as the empirical radius saturates with increasing sampling, even when the governing equations are unknown. These results demonstrate that deterministic functional topology offers a unified mathematical foundation for perception, representation, and world-model construction. It provides a geometric explanation for why biological learners and self-supervised AI systems can generalize from few observations, and establishes compact perceptual manifolds as a fundamental building block for future AI architectures. Finally, this work unifies biological perception and modern self-supervised models under a single geometric principle: both derive their generalization ability from the compactness and invariants of real-world perceptual manifolds.

Debdeep Mukherjee, Eduardo Di Santi, Clément Lefebvre et al.

Track circuits are critical for railway operations, acting as the main signalling sub-system to locate trains. Continuous Variable Current Modulation (CVCM) is one such technology. Like any field-deployed, safety-critical asset, it can fail, triggering cascading disruptions. Many failures originate as subtle anomalies that evolve over time, often not visually apparent in monitored signals. Conventional approaches, which rely on clear signal changes, struggle to detect them early. Early identification of failure types is essential to improve maintenance planning, minimising downtime and revenue loss. Leveraging deep neural networks, we propose a predictive maintenance framework that classifies anomalies well before they escalate into failures. Validated on 10 CVCM failure cases across different installations, the method is ISO-17359 compliant and outperforms conventional techniques, achieving 99.31% overall accuracy with detection within 1% of anomaly onset. Through conformal prediction, we provide uncertainty estimates, reaching 99% confidence with consistent coverage across classes. Given CVCMs global deployment, the approach is scalable and adaptable to other track circuits and railway systems, enhancing operational reliability.