Model-based test execution from high-level natural language instructions using GPT-4

Official URL: https://dx.doi.org/10.1007/s11219-025-09712-9

This work presents a novel, model-based approach to generating and executing UI tests directly from natural language instructions, addressing the high cost and complexity of traditional user interface (UI) testing, particularly in domains like smart TVs, which lack comprehensive automation support for testing. By combining generative artificial intelligence (AI) models with automatically discovered UI models, the method translates natural language test scenarios into executable, black-box test cases through semantic interpretation of screenshots. Evaluated on a smart TV platform and an open-source media manager, the approach outperforms alternatives, demonstrating its effectiveness and adaptability. Key contributions include tackling smart TV-specific testing challenges, introducing a model-based generative AI framework, and validating its utility across diverse real-world applications.