“Artificial Intelligence shapes how we think, feel and behave. It drives the decisions that define our future. We have the responsibility to use this potential for humane technology. Building an AI based on our diverse values and needs requires thoughtful design.”
– Lennart Ziburski, UX of AI
Things have been developing rapidly since I posted my first Artificial Intelligence Experience Design Principles in 2018. We’ve had more experience designing with artificial intelligence (AI) and AI has quickly integrated in almost every part of our daily lives (well, for some of us). Here are some more AI experience design principles to consider:
Start with the understanding your audience. Think about how the people do the task today. Figure out what their goals are and how your technology is going to make it easier for them to accomplish their objectives.
Set the right expectations. People will expect your AI to be both “smarter” and “dumber” than it is. Try to explain – in plain language – what your solution can do, and where its limitations are.
Explain your results. Trace results back to the supporting data. If that’s not possible, explain the basic operation of the algorithm, the data sources you use, and which qualities the AI focuses on.
Communicate your confidence. Show a percentage, star ratings, or other cues.
Degrade gracefully. When the input is clear and the answer certain, awesome! Less confident results need to be presented differently – like “toning down” the boldness or how you “frame” the results.
Know what not to automate. Some tasks are best done by humans (understanding emotions or motivations), have an intrinsic value to the manual process (providing dignity or enjoyment), or requires subjective evaluation (ethical or moral decisions).
Keep the human in control. Provide feedback, reverse bad actions, and reward good ones.
Build trust over time Don’t require personal data (or minimize as much as you can). Make suggestions – NOT decisions. As your AI gets to know who it’s helping, it can automate more and ask for permission less.
Adapt to evolving changes. Values and needs change over time. If your AI is stuck with what it has learned in the past, it will be out-of-date, irrelevant or worse (provide misinformation).
Balance predictability and serendipity. Any personalized AI adopts the prompter’s bias. This is great for tasks that require predictability, where you need consistently effective results. But for other tasks, it limits curiosity. It constrains options inside comfort zone. Part of being humans is following your intuition off the beaten path, even if it might lead nowhere. Tweak your algorithms to find the right balance, and maybe even design your interface to offer ways to escape the “filter bubble”.
Prototype with real data and fake AI. Using real user data for early prototypes helps build machine learning model on the right assumptions. Use the Wizard of Oz method to get the experience design right before actually building the AI.
Share your process and intentions. Explain how the data is gathered, handled and processed. Sharing process and intentions builds trust and goodwill.
AI is starting to impact most everyone. All of us should be part of the discussion of what we want AI to be. AI is shaped by the experiences and values of the people that make it. As we are still figuring out the foundations of AI design, collaboration is more important than ever. Data analysts, researchers, developers, marketers and designers all need to work together to build a cohesive product. Domain experts should be deeply involved in the design of both the ML model and the interface architecture and design. It’s our job to translate expertise into a shared understanding that our team can build on and create experiences that our customers love.
—
Ziburski, Lennart. UX of AI. https://uxofai.com