UX for Machine Learning

A designer’s view on integrating data and intelligence

Machine learning is quickly being embedded in numerous products and services across a wide range of industries. It will soon be impossible to make any product without considering machine learning.

Machine learning is influencing how people make smarter decisions about their finances, how suppliers manage inventory more efficiently, how diseases are diagnosed, how people have access to mental health care, and how energy providers more efficiently distribute power. These are only a handful of its many applications today.

Engineers are getting proficient at building advanced algorithms. Data-scientists are getting more proficient at synthesizing and analyzing meaning from huge data sets. More product managers are generating requirements and developing market strategies with machine learning in mind. But what about designers? We’ve always wanted to take advantage of the data that machine learning gives us. But now that we can, how do we actually do that?

What are the implications, the possibilities, the responsibilities of designing for this new world?

At Punchcut, we regularly consider machine learning in our solutions. We’re even starting to apply it internally within our own design processes. The intent of this post is to share our insights and to provide a starting point for designers to become comfortable with designing for machine learning.

1 / Get comfortable with quantitative data.

Machines communicate with numbers. When designing for a system that learns, we will encounter numbers and probabilities that will guide it’s decisions. As designers, we will need to translate these numbers in a way that supports our user’s goals throughout their journey.

Let’s take an example of an application we built at Punchcut. RecycleBot visually analyzes a waste item and tells us which bin it should go to. Showing RecycleBot a coffee cup will give us something like this. “The probability of this item belonging to the compost bin is 0.892.” Here is where design comes in. What does this mean for the user? Should we tell the user to put it in the compost bin? What should be the minimum probability to give users a direction?

We don’t need to understand everything about a machine’s algorithms but we should be familiar with how they communicate so we can design the best experience. Enlisting the help of data scientists, engineers and subject-matter experts can help us refine our understanding of the machine and ultimately serve users better.

2 / Use user data to create hyper-personalized experiences.

For decades, designing truly personalized experiences have always been relegated to “blue sky” concepts that inevitably meet the realities of system limitations. But not anymore. Machine learning gives designers superpowers. With such an abundance of user information at our disposal, we can finally design hyper-contextual and hyper-personalized experiences. We still have a ways to go but some services are already going in that direction.

Let’s take Netflix as an example. Netflix’s goal is to build a personalized recommendation system able to “get the right titles in front each of our members at the right time.” When browsing Netflix, not everyone sees the same artwork. Instead, artwork is tailored based on user’s preferences and behaviors. For example, when showing cover art for Pulp Fiction a user who has watched many movies with Uma Thurman would see Uma on the cover. By contrast, a user who has watched more movies with John Travolta would see John on the cover. This makes the already personalized experience even more personalized.

Different artworks of Pulp Fiction are shown depending on user’s preference for cast members. Source

We’ve entered a world in which people expect to interact with experiences highly personalized to them. In the past, designers were limited to what systems can do. Now systems are limited to what designers can imagine.

3 / Design when and how to collect data.

As the name “machine learning” implies, machines need a ton of high quality data to train itself with. How do we get this data? First, we need to think about what kind of data we need to collect from the beginning. Second, designers need to design ways of collecting data that feel like a natural part of the user’s flow.

For example, Google Clips strives to automatically capture candid life moments, freeing users to enjoy moments rather than take pictures. However, their designers decided to still bring users into the process by allowing them to decide how the final curation should look. Users are presented with many moments taken by Clips and are asked to delete the ones they don’t want to keep. This helps the algorithms learn what users consider as memorable moments.

Users can select which moments they want to keep which helps improve the AI. Source

Gathering good data is the core of machine learning systems. Designers should have a hypothesis for helpful data to collect and have a plan for how and when to collect this data.

4 / Design for imperfection.

Machine learning is powerful but far from perfect. Human behavior is complicated and hard to predict. Take AI agents like Amazon Alexa and Google Home as examples. Although they are making strides in understanding natural language, anyone who has used these agents will tell you that they still have a hard time understanding the nuances in the way we speak. Designers need to understand these limitations and provide an option for users to escape choices made by machines.

Here’s another example. Let’s say we’re using machine learning to analyze a user’s listening patterns to recommend new music. What if the user decides to use her account to cast music at a party and allows others to change the music? What if she plays lullaby songs every night to help put her baby to sleep? These songs will count toward the algorithms that determine what music to recommend even though they were only played in very specific use cases.

Eventually, machines will learn to automatically account for these contexts. Still, designers should be aware of these use cases and design for ways users can opt-out of learning or modify the data.

5 / Be transparent about the data being used.

When was the last time you looked at a product online and then see that product on your Instagram feed, Facebook feed, and as ads on other sites. We know that products “follow” us based on our search history and browsing behavior. However, concerns around privacy can be amplified if users see a relevant ad but can’t figure out how the machine knew it would interest them. People form speculations. Is the machine listening to my conversations? Is the camera on my laptop watching me? Are my private messages being scrubbed?

Designers should be transparent and let users know what data is being collected and how it is used. Leave a trail that helps lead users to think, “This may be why it’s suggesting this.” A common example of this is the tried and true e-commerce design on sites like Amazon clearly labeling why certain items are suggested (e.g. “Customers who bought this item also bought”).

Users will formulate assumptions whether brands like it or not. Better to be transparent so that their assumptions are correct.

UX Design + Machine Learning

As products and services continue to integrate machine learning, designers will have a crucial role in adding value to the users’ experience. Machine learning systems are rapidly evolving and offer new opportunities to enhance humanity. Designers must embrace quantitative data and strive to apply data science principles in ways to infuse more meaningful moments in people’s lives. We’re super excited about continuing to master this new “superpower” in our designer tool belt.

A Punchcut Perspective
Contributors: Eunsol Byun, Reggie Wirjadi, Prooshat Saberi
© Punchcut LLC, All rights reserved.