Social Media as an Early Warning System: An AI-Driven Approach to Infoveillance

Over the past decade, as digital engagement has become deeply embedded in daily life, social media platforms have evolved into powerful, dynamic environments that capture public sentiment, behavioral shifts, and signs of emerging crises in near real-time. This talk presents an AI-driven infoveillance system, an intelligent early warning framework designed to mine large-scale public discourse on social media for sentiment patterns, misinformation trends, and early signals of emerging crises. The system builds on a series of peer-reviewed publications led by the speaker, including analyses of over 4 million Reddit posts, over half a million Instagram posts, and thousands of posts on X and YouTube related to COVID-19, Long COVID, and Mpox. This work demonstrates how modular AI pipelines can be designed to support global situational awareness. The architecture integrates multilingual sentiment analysis, toxicity detection, and temporal trend analysis into a unified, scalable system. The system leverages techniques such as cross-lingual model adaptation and transformer distillation to enhance performance across different social media platforms, enabling robust analysis even when processing noisy and unstructured social media data. To ensure trustworthiness, the system incorporates explainable AI modules and aligns with responsible AI principles such as transparency and robustness. In alignment with the IEEE Systems Council’s focus on AI in Systems, this talk emphasizes system-level design thinking, architectural trade-offs, and deployment challenges in large-scale, real-world social media analytics. This framework illustrates how AI systems can serve as scalable, explainable, and ethically grounded tools for supporting public safety and digital trust during moments of crisis.