Discernimento entre Códigos-fonte Gerados por Pessoas e por Inteligência Artificial: Estudo com Programação em Java

Abstract

The technological revolution driven by the popularization of Artificial Intelligence (AI) has raised the pressing need for organizations to maintain teams of highly qualified professionals in Information Technology (IT), in areas such as software development, to discern between human creations and productions generated by machines. This monograph addresses the development of source code in Java generated from ChatGPT and Bard artificial intelligence and asks 10 CI-UFPB students to identify whether the code was created by a machine or a human programmer. The main objective is to study and understand the metrics and heuristic criteria used to identify the origin of codes using the Turing test to extract relevant data. The Turing test is a method that aims to determine whether a machine can exhibit intelligent behavior indistinguishable from human behavior. In this context, the Turing test is adapted to assess whether the code generated by artificial intelligence is convincing enough to be confused with human code by students. To achieve this goal, the research focuses on five heuristic criteria for code analysis, including structure, semantics, and syntactic characteristics. These heuristic criteria are essential for understanding the generation of source codes according to the results, as 100% of participants stated that the codes were useful for identifying the origin of the code.