Practical AI on the Google Cloud Platform

Chapter 8. Building AI Assistants in the Cloud

As the world plunged into a pandemic in early 2020, the need for smarter AI became clear. Essential workers all over the globe were at high risk of being carriers and sources of contagions. We needed AI that could not only make better decisions but also help buffer those essential workers with smarter AI assistants. Now, we’re not talking about incarnating Data, the android from Star Trek. It is only 2020 after all. What we are talking about is creating smarter AI assistants that can assist us in physical tasks—AI that can assist us in times of crisis, like we’ve seen in 2020.

We already have many assistant platforms such as Siri, Alexa, and Google itself. Couldn’t they do what we need from smarter assistants? In part, yes, but most of these systems themselves are not real AI. Many are in fact rule-based systems developed by programmers and defined through extensive testing and training. Most of these assistant services are transitioning to real AI, but this takes time. The other main downside of these services is the input mechanism itself. In almost all cases, these services rely on voice- or text-based natural language input. These are great, but what about the old adage “a picture is worth a thousand words”?

In this chapter we explore a new domain of smarter AI that uses reinforcement learning. Reinforcement learning (RL) is a special form of learning that is intended to be dynamic and self-learning. We will explore the basics ...