The author discusses the lack of clear guidelines or standards in the field of artificial intelligence (AI) software engineering. They highlight the high failure rate of AI projects and the issue of irreproducible results in research articles. The author defines AI as the study and construction of agents that act rationally. They also differentiate between AI and machine learning, with machine learning being a subfield of AI. The author explores the model-centric and data-centric approaches to AI/ML and suggests that the data-centric approach is more suitable for real-world applications. They propose an AI Engineering Process (AIP) consisting of various steps, such as problem definition, dataset selection, data preparation, model design, tuning, and deployment. The author emphasizes the importance of literature review and problem decomposition in the problem definition phase. They also explain the PEAS description for specifying the task environment. The article concludes by highlighting the need to decide whether supervised, unsupervised, or reinforcement learning is required for the problem at hand.

source update: The AI Process – Towards AI