PROSPECTS OF USING AI FOR AUTOMATED SOFTWARE TESTING
DOI:
https://doi.org/10.30857/2786-5371.2025.5.1Keywords:
software testing, test automation, artificial intelligence, machine learning, test generation, self-healing, defect prediction, quality assurance, neural networks, CI/CDAbstract
Purpose. The research method is to analyze the prospects for using artificial intelligence to overcome key limitations of traditional automated software testing, in particular, test fragility, low adaptability, large volumes of test scenarios and the inconsistency of testing speed with the pace of modern development, as well as to justify the transition to intelligent, self-learning quality assurance systems.
Methodology. The work uses the method of systematic review and synthesis of modern scientific literature and publications of leading technology companies, a comparative analysis of traditional and AI-oriented approaches to testing, general practical use of examples using large language models, reinforcement learning, graph neural networks and self-healing technologies, as well as illustrative modeling of the benefits of AI through schematic diagrams and examples.
Findings. The study identified six main problems of classical testing and demonstrated that the integration of AI allows you to automatically generate test scripts (including from natural language), predict defects, implement self-healing mechanisms, optimize test priority in CI/CD, detect flaky tests, and significantly increase code coverage and process stability while simultaneously reducing support costs autotests.
Originality. The novelty lies in the systematic synthesis and updating of trends in the use of large language models for generating tests based on natural language, graph neural networks for dependency analysis, as well as reinforcement learning methods for adaptive navigation in interfaces, which simultaneously contribute to the formation of the concept of proactive, intelligent, and context-oriented software testing provision.
Practical value. The results of the work have direct practical value for QA teams and DevOps processes: the proposed AI approaches allow you to reduce time and resources for supporting automated tests, accelerate CI/CD cycles, increase the reliability of releases, optimize the focus of testing on risky areas of the code, and ensure the availability of processes even for a team with limited technical resources.
