According to a recent study, software companies may be managed quickly and economically by artificial intelligence chatbots like OpenAI’s ChatGPT. The findings followed the publication of another study in which big language model-powered AI agents were capable of managing a virtual town on their own.
In a recent publication, a group of scientists from Brown University and other Chinese universities tested whether ChatGPT 3.5 model-powered AI bots could complete the software development process without any prior training. Researchers developed ChatDev, a fictitious software development firm, to test this. The company was divided into four phases in chronological order based on the waterfall model, a sequential technique for developing software: planning, coding, testing, and documentation.
Researchers then gave each AI bot “vital details” that explained the “designated task and roles, communication protocols, termination criteria, and constraints” before assigning the AI bots to those positions.
Each AI bot was assigned to its appropriate stage once the researchers gave it its tasks. For instance, ChatDev’s “CEO” and “CTO” worked during the “designing” phase, while the “programmer” and “art designer” contributed during the “coding” phase. The AI workers conversed with one another to fulfill specific tasks of the software-development process at each stage, from selecting a programming language to finding faults in the code, until the project was finished.
To determine how long it would take ChatDev to develop each type of software and how much it would cost, the researchers experimented using several software scenarios. They then used several analyses for each scenario. For instance, researchers asked ChatDev to “design a basic Gomoku game,” a strategy board game with an abstract theme commonly known as “Five in a Row.”
The CEO requested the CTO at the design stage to “propose a concrete programming language” that would “satisfy the new user’s demand,” and the CTO offered Python in response. The CEO responded, “Great,” and stated that the programming language is “popular among both novice and seasoned developers alike” because of its simplicity and readability.
ChatDev then moved on to the coding stage, where the CTO asked the programmer to write a file before the programmer asked the designer to provide the software with a “beautiful graphical user interface.” After the CTO responded, “Let’s get started,” ChatDev continued. Until the program was developed, the chat chain was repeated at each stage.
The study found that after giving ChatDev 70 tasks, the AI-powered company was able to finish the entire software development process “in under seven minutes at a cost of less than one dollar,” on average. This was achieved while identifying and resolving “potential vulnerabilities” thanks to its “memory” and “self-reflection” capabilities.
According to the research, 86.66% of the software systems created were “executed flawlessly.”
The researchers stated in their report that their experimental findings “demonstrate the efficiency and cost-effectiveness of the automated software development process driven by CHATDEV.”
Before the article was published, Insider contacted the researchers for comment, but they did not react right away.
The study’s results show one of many ways that generative AI tools like ChatGPT can carry out particular job functions. Workers from many industries have been using the AI chatbot at work to save time and increase productivity since its release in November.
In particular, programmers may benefit from generative AI technologies in both their personal and professional lives. It was discovered that Amazon employees utilize ChatGPT for software development, and a Berlin-based programmer named Daniel Dippold used it to create a tool that assisted him in finding housing.
However, there were certain drawbacks to the study that could affect the development of software, including biases and inaccuracies in the language models. The findings, according to the researchers, “may potentially help junior programmers or engineers in the real world” in the future.
