Data Science using openAI: testing their new capabilities focused on data science

Abstract

Even though statistics is taught in several courses, including life sciences, their applications tend to be challenging; basics errors are inevitably done by researchers. It is not uncommon discussions on errors done during statistical analysis done on data by non-experts in statistics, in scientific papers. However, statistics is too important and crucial for science for being left to statisticians only. On this paper, we discuss the possibilities opened by openAI latest API called coder interpreter. This API is able to read CSV files, and perform statistical analysis on the dataset, having as starting point human's commands. Different from classical statistical tools, it is able to choose what method and library to use, get the results back, and give an evidence-based answer. The user need only to know what to ask, which may require minimal knowledge on statistical inference, most likely, covered on any course on statistical analysis. Several areas can benefit from this tool, we think, especially evidence-based medicine, an area where the researchers are also medical doctors. This tools is actually using open source python libraries: the python community was able to build a rich and powerful set of libraries for data science, and this new API from openAI is exploring those libraries. We believe that those tools can be used on real research, and we present a couple of examples, one of them, studying a dataset in diabetic patients. We hope to influence the usage of this tool by non-experts, as well experts, on statistics.