Building the Next Generation of AI Infrastructure

Chapter 3. There Are Multiple Factors

Up to now, you have gotten a mix of industry context, layered approach, tangible checklists, and recommendations to approach your AI infrastructure exercise. But nothing in life is absolute and you will need to face some trade-offs and deal with unclear decision paths.

In this chapter, we will try to bring a new dimension to your decision-making process—specifically, all the variables that will help you adjust your AI infrastructure to your company context. Let’s start with the notion of lifecycle, as that will provide a structured scope to your AI developments.

Think in Terms of AI Lifecycle

Imagine the AI infrastructure as that set of interconnected vertical layers, stacked one after the other, like you saw in Figure 1-2. The notion of the AI lifecycle can bring you a two-dimensional approach, because you can organize it as a horizontal sequence that defines your AI developments.

The notion and details of an AI lifecycle depends on the author. You have technical lifecycles like the classic Team Data Science Process (TDSP) or compliance-oriented versions like ISO 5338 or NIST AI. Ultimately, regardless of the version you choose, those lifecycles typically follow a more or less granular sequence of ideation, data preparation, AI modeling, testing, and deployment activities. Some of them include continuous post-production monitoring and the final decommission of the AI application.

For the sake of simplicity, we will define five simple AI ...