Many applications have implemented AI in a spectrum of agentive to assistive. The UI patterns that were developed to make interfaces user-friendly sometimes fall short when we apply them to AI-powered applications. How do we balance the needs of the AI that controls the application and the needs of a user who wants to be in charge? How do we keep our interfaces user-friendly when AI is doing much of the work?
To identify AI's design challenges in our industry, I researched how big companies like Microsoft, Google, IBM, and the like are dealing with AI implementation today. Many have publicly issued their AI manifesto and policies around fairness, privacy, security, explainability, etc. I collected and organized these policies into three main themes that would help us organize our design patterns in a meaningful way.
I collected and audited AI-powered applications in various industries, including healthcare, education and e-commerce. I found examples of interface elements that allowed a form of interaction between the user and the machine’s predictions and actions. I then organized these examples into patterns addressing each of the three main themes.
An interaction pattern library consists of the pattern’s name, a clear problem statement and a description for the solution describing how that specific pattern behaves, and examples of the pattern being used in context. I wrote definitions for the initial patterns and our team continue to write new interaction patterns as they become available in the industry.