This chapter is from the book
This chapter is from the book
This chapter is from the book
1.7 Summary
This book discusses AI engineering—the application of software engineering to AI systems. AI systems are machine-based systems that use AI in one or more of their components, but always include other, non-AI components. Given that the overall quality of a system depends on all of its parts and their interplay, it is important to practice AI engineering well, with the aim of creating high-quality AI systems and operating them effectively.
The quality of an AI system depends on three factors: the life-cycle processes used, the software architecture, and the quality of the AI model. The processes used help the designer detect errors—both logical and in the model output—early in the development process.
The software architecture contributes to the achievement of one set of quality attributes requirements; it does this through the use of architectural tactics. The AI model used also contributes to the achievement of quality attribute requirements; it does so through the choice of model type, training data, and training algorithms. Data preparation is important in ensuring model quality. Data cleaning and feature engineering are two aspects of data preparation.
Three types of AI models are symbolic models, narrow ML models, and foundation models. FMs and narrow ML models are specialized types of ML models. The probabilistic nature of AI models means an organization should perform a risk assessment to determine the effects of incorrect outputs of the AI models and AI systems.
