ПРИЛОЖЕНИЕ НА ГЕНЕРАТИВЕН ИЗКУСТВЕН ИНТЕЛЕКТ В ПРОИЗВОДСТВОТО НА ОФИЦИАЛНА СТАТИСТИКА
APPLICATION OF GENERATIVE ARTIFICIAL INTELLIGENCE IN THE PRODUCTION OF OFFICIAL STATISTICS

Summary/Abstract: The article examines the possibilities for the application of generative artificial intelligence (GenAI) in the production of official statistics, with a special focus on the potential for its introduction into the National Statistical Institute (NSI). The main classes of generative models are described - variational autoencoders (VAE), generative adversarial networks (GAN and DP-GAN), diffusion models and large language models (LLM), as well as their typical applications: imputation of missing data, generation of synthetic microdata, automated text coding, migration and optimization of code, and short-term forecasting of key indicators. Through a systematic review of scientific and institutional sources and thematic analysis (LDA), four main directions have been identified: synthetic data, data editing, automation of the programming code and short term forecasts. Specific examples from the practice of several national statistical institutes are presented, demonstrating significant improvements in terms of statistical utility, risk of disclosure, and quality of text outputs, measured by indicators such as pMSE, Kullback–Leibler divergence, record linkage, F1-measure and Cohen’s κ. On this basis, the requirements for quality management, confidentiality, and cybersecurity are discussed, including the integration of GenAI into the Generic Statistical Business Process Model (GSBPM) framework and the ISO 9001 standard, as well as the relationship with the upcoming EU regulation on artificial intelligence. The conclusion emphasizes that, with a strategic, phased, and ethically justified approach, GenAI can become a key tool for the modernization of official statistics in Bulgaria.

• Details
• Contents