Data Analysis Using Generative AI: Opportunities and Challenges
Volodymyr I Skitsko,Kyiv National Economic University named after Vadym Hetman,Volodymyr I. Skitsko,
DOI: https://doi.org/10.32983/2222-0712-2023-4-217-225
2023-01-01
THE PROBLEMS OF ECONOMY
Abstract:The article examines topical issues of using generative artificial intelligence, in particular large language models ChatGPT and Claude, for data analysis. The essence of the term «artificial intelligence» changes over time. And if earlier, when using this term, was talked about expert systems, machine learning, etc., now this term means primarily large language models, among which the most famous are ChatGPT, Claude, Bing, Bard. These models allow you to generate texts, images, audio, and video based on user requests. The purpose of the article is to explore the opportunities and challenges of using ChatGPT and Claude in data analysis based on existing publications and our own experience. After all, the ability of large language models to communicate in natural language makes them a powerful analytics tool. The paper considers aspects of the main stages of data analysis in the context of the use of large language models: obtaining, collecting and loading input data, their pre-processing, application of mathematical models, visualization and interpretation of results. Practical recommendations for formulating requests to ChatGPT and Claude at each stage of data analysis are provided. It is noted that ChatGPT, thanks to the built-in Advanced Data Analysis service, allows an effectively analysis of data, powered by Python language. This provides higher accuracy of results compared to other large language models. Using a conditional example, a comparison of the capabilities of ChatGPT and Claude in data analysis is carried out. It is shown that ChatGPT allows you to build dependency models, generate graphs and give meaningful explanations of the results obtained. At the same time, Claude’s capabilities in data analysis are quite limited. It is concluded that ChatGPT has significantly greater potential for data analysis compared to Claude and other chatbots. However, so far, large language models cannot completely replace data analysts and powerful decision support systems. In further research, it is proposed to focus on the study of the practical application of the capabilities of ChatGPT and, in particular, its Advanced Data Analysis service to solve various data analysis problems.