Systematic mapping of non-functional requirements and their impacts in architectures for artificial intelligence

Artificial intelligence (AI) is a branch of computer science that seeks algorithms capable of learning, reasoning, adapting, and performing tasks in a human-like manner. Software is determined by a set of functional requirements (FRs) and non-functional requirements (NFRs). NFRs, such as scalability, security, performance, and usability, define how a system should function beyond its basic functionalities. The architecture must be designed to meet these NFRs, ensuring system efficiency and reliability while maintaining a flexible and modular structure. Project-specific requirements, technological advancements, and the domain greatly influence the variety of NFRs and architectures that will be essential for the development of AI-based systems. Selecting an appropriate architecture is a process intrinsically guided by the prioritized NFRs. This systematic mapping seeks to provide the NFRs, architectures, and how these impact the selection of architectures for AI-based systems. The evidence gathered shows that NFRs tend to prioritize security, scalability, and accuracy. At the architectural level, there is a similar trend toward patterns such as microservices and lambda. Finally, in terms of impact, microservices is highly influenced by reliability and scalability, workflow orchestration by transparency and explainability, and federated learning by sensitive data security.