Designing AI Interfaces

Preface

In his book Thinkertoys (Ten Speed Press, 2006), Michael Michalko offers innovators a number of different frameworks for working through tricky problems. Using exercises and analogies, he invites readers to apply a variety of methodologies that promote abstract, creative thinking. In one part, he describes the “Phoenix Checklist,” a rich set of questions developed by, of all people, the CIA, designed to help agents move forward when a task seems uncertain—or impossible.

Many of the questions focus on exploring and defining the problem and the surrounding context. For example:

• What information do you have?

• What are the unknown factors?

• Why is it necessary to solve the problem?

But the heart of the process involves exploring other solutions and evaluating their relevance to the problem at hand:

• Have you seen this problem in a slightly different form?

• Suppose you find a problem similar to yours that has already been resolved. Can you use the same method?

• What have others done?

It’s this focus on foundational problems that piqued my interest. When I first came across this methodology, I had just begun working with AI at Google on its TensorFlow project and regularly felt baffled and hopeless about how to proceed. In the intervening years, I’ve returned to these questions as an establishing process for designing for AI, with a particular focus on how the early work on graphical user interfaces (GUIs) and human-computer interaction (HCI) laid a useful foundation ...