Chapter 6. The Simplest Way to Scale Artificial Intelligence
Scale is about making what you build as useful as possible to as many people as possible without drowning in complexity or expense. Scaling any kind of software system (including AI) isn’t simple. However, there are things that you can do to make scaling AI as simple as possible. The simplest way to scale AI is to build and deploy utility services.
The AI Infrastructure
Utilities set standards that make technologies work together and ensure that everything is available through a common interface. Treat AI platforms as commodities. Design the AI applications so that platform components can be plugged into and swapped out of the existing legacy infrastructure. Choose technologies capable of running AI as widely available, reliable services.
Figure 6-1 shows the blueprints of an AI algorithm. These are the basic technical elements needed to deliver a functional, operational, and industrial AI.
Figure 6-1. A blueprint for the AI utility
Utilities set standards for simplicity. Consistently choose the minimum viable algorithm you need to test out your hypothesis. Don’t, for example, build convolution neural networks when linear regression will do. Pull in as little data as possible and map all of the data consumed to a standard domain model. Don’t build a world-class data warehouse when a simple file store will do. However, ...
Get Artificial Intelligence now with the O’Reilly learning platform.
O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.
