Signals from the noise: Decoding global AI discourse in central bank communications

Abstract

This study examines how artificial intelligence (AI) has entered and evolved within the official communication of central banks. Using an archive of central bankers’ speeches covering 1996–2024, we develop a multi-stage NLP framework that combines an AI–finance lexicon validated by a fine-tuned transformer model, sentiment analysis, and structural topic modeling with temporal and country covariates. We identify 1032 AI-related speeches and show that AI discourse was marginal before the mid-2010s, rose sharply during the fintech wave of 2017–2018, and accelerated again with the emergence of generative AI in 2023–2024. Central banks frame AI primarily in positive and neutral terms, while caution is concentrated around generative AI, model governance, cyber threats, and supervisory accountability. Topic modeling reveals ten stable AI-centered themes spanning sustainability, cyber risk, payments, supervision, productivity, policy analysis, monetary transmission, and fintech experimentation. The findings suggest that AI has become a policy-relevant element of central bank communication, through which monetary authorities signal institutional readiness, frame emerging risks, and guide expectations in a rapidly changing financial system.