3 Good UX Practices for Designing an Intuitive AI Applications
Last year I was designing product Merlon Intelligence, where we used Artificial Intelligence (AI) to deliver a core functionality. There is no simple pattern for designing for AI. However, over that time I found some UX practices more useful than others. Here are 3 to embrace when striving for an intuitive UX.
In this article, I am using the simplified term AI (Artificial Intelligence), covering terms Machine Learning, and Machine Learning Models.
1. Design a simplified conceptual model of AI
I once asked an engineer to explain to me how AI works. He started with straightforward terms, then suddenly he said: “…then imagine you have a 20 dimension of data”. I stared blankly.
Understanding AI is hard. It uses a complex model to produce outcomes, and it is nothing intuitive at all. People get Master degrees to grasp it completely. So how it is possible millions of people are using it daily?
“The design projects all the information needed to create a good conceptual model of the system, leading to understanding and feeling of control”
— Dan Norman, The design of Everyday Things
Every person using any tool or a machine builds a conceptual model of how it works. Most of the people using AI-powered search can’t grasp how 20 dimension space of data looks like; however, they have a certain understanding of how the keyword search works. And for the most of us its just enough…
On the UI communicate a simplified conceptual model, which is solid and people can easily follow.
2. Explain WHY with metadata
“Hey all that AI is cool, but how can I trust the AI decisions?”
Although AI innovation comes in areas with measurable higher accuracy than humans, users still may have certain objections using smart application.
Dig deeper into their question. You can use user roles to identify specific concerns people have. Some people may look for ways to improve the algorithm, other for accountability of its decisions. Identify the user concerns and show supporting metadata aligned with a conceptual model you have designed.
Overwhelming the users with details of the AI, often cause more frustration than usefulness.
If your AI process text sources, refer to it. Google Translate displays the translation frequency of appearance in public documents. Higher the frequency, better the result. Is it an exact answer? No. Does it explain its decision and make sense within the product? Yes.
3. Always keep people in the control
Self-driving cars have a steering wheel, until the technology advance to the point that having it there cause more harm. Whatever the future car may look like, it will always have an emergency brake at the hand of the passenger.
Take Spam email functionality as an example. You can perform any on email in the spam folder. You can also add a new mail to the spam folder, so the system knows similar messages are marked as spam as well. The intelligence used to identify unwanted messages is there, but you can always take action on top of it.
However your AI can be smart at the end of the day, you are building the tools for people to use. Once you present AI outcome on the user interface, allow people to interact with it as other items. Include Dismiss button, or report. So your AI can gather valuable feedback on its outcomes.
Would you like to read more articles on the design of AI products? Clap to let me know if you liked it and follow me to not miss the next article.