CIPAC: A framework of automated software construction based on collective intelligence

Abstract

In software development, constructing programs efficiently and accurately is essential, which leads to the rise of program synthesis. However, the complexity of program spaces and the diversity of user intents restrict the scale and quality of the code generated, resulting in many works focused on generating function-level code. To address the automated generation of complex code, we propose an automated program construction process model CIPAC based on collective intelligence, which can generate software-level code automatically. CIPAC incorporates methods for the automated aggregation of collective intelligence, the construction of software structures and task specifications, efficient program generation and search, and the optimization of code quality and composition. CIPAC employs explainable program synthesis methods as its core to ensure reliability and leverages collective intelligence throughout the entire software development lifecycle. To validate the effectiveness of CIPAC, we conduct a case study for developing a matrix operation application and explore complex tasks in the aerospace domain. The results show that our platform enables the construction of the software project compared to existing methods. To demonstrate reliability, we conduct integration testing, system testing, and security validation, with results indicating that the generated project passes all tests without security issues.