Automating GUI-based Software Testing with GPT-3
Daniel Zimmermann, Anne Koziolek
Abstract
This paper introduces a new method for GUI-based software testing that utilizes GPT-3, a state-of-the-art language model. The approach uses GPT-3’s transformer architecture to interpret natural language test cases and programmatically navigate through the application under test. To overcome the memory limitations of the transformer architecture, we propose incorporating the current state of all GUI elements into the input prompt at each time step. Additionally, we suggest using a test automation framework to interact with the GUI elements and provide GPT-3 with information about the application’s current state. To simplify the process of acquiring training data, we also present a tool for this purpose. The proposed approach has the potential to improve the efficiency of software testing by eliminating the need for manual input and allowing non-technical users to easily input test cases for both desktop and mobile applications.