Concept drift and human-centric vulnerabilities in cognitive cities : A scoping review

Sammanfattning

The increased use of AI-driven technologies in cognitive cities has introduced new challenges related to cybersecurity. While previous research has primarily focused on traditional cybersecurity threats, less attention has been given to emerging challenges related to adaptive AI systems and human involvement, particularly concept drift and human-centric vulnerabilities. The aim of this study was therefore to explore how these threats are defined, conceptualized and addressed within existing smart and cognitive city research. To address this, a scoping review was conducted to examine existing literature related to these emerging challenges. The findings showed that concept drift is primarily described as changes in data distribution over time that reduce model performance and require continuous adaptation strategies. The findings also described cybersecurity concerns with concept drift, particularly in relation to intrusion detection, anomaly detection, and malware classification systems. Human-centric vulnerabilities are mainly discussed through multiple themes such as participation, bias, trust, transparency, governance and human behavior. The review identified potential connections between these threats, suggesting that human-generated data may influence the data that AI systems rely on. The study concludes that existing research remains fragmented between technical and socio-technical perspectives, highlighting the need for more integrated approaches to support the development of adaptive, trustworthy and resilient cognitive city systems.