Successful ML solutions start with UX
Prioritizing human-centered design for better outcomes
We need to reevaluate the role and involvement of user experience (UX) teams in machine learning (ML) projects. Research has indicated that UX teams rarely have the opportunity to come up with entirely new ways to implement ML solutions. In a survey of 27 UX practitioners, only 25% reported the UX team playing a role in generating novel design concepts.
The problem is that UX teams are often engaged in ML projects toward the end of the development process, after models have already been trained and functional decisions have already been made. UX designers are often expected to make existing ML products more visually appealing and user-friendly, without addressing underlying shortcomings or historically poor user adoption. At times, we are simply putting lipstick on the pig.
By involving UX teams early in the creation of ML solutions, we can foster innovation, identify user needs more accurately and ensure we’re not only getting the design right but also designing the right thing.
Involve UX Early and Often
Involving UX designers early in the product development process will help ensure your ML products are user-friendly, effective and successful. Benefits of early UX involvement include avoiding costly mistakes and aligning the ML project with actual user needs.
ML projects require substantial time, money and resources, and limitations in prototyping technology mean that ML development demands an extraordinary level of commitment. There’s little room for “failing fast and often.” It’s important to have confidence in the desirability, feasibility and viability of a proposed ML solution before diving into costly development. If there is no desire for your product, the feasibility and viability of its development become moot. By bringing UX designers on board from the outset, you can ensure that your ML solutions are designed with user needs at their core.
Case Studies: The Impact of Human-Centered Design
To further emphasize the importance of involving UX teams early, let’s look at some case studies that demonstrate the impact of human-centered design.
Caste Study 1: Clinical Decision Support Tool
In Unremarkable AI: Fitting Intelligent Decision Support into Critical, Clinical Decision-Making Processes, a UX team discusses their process in designing a clinical decision support (CDS) tool. CDS tools are meant to enhance decision-making in the clinical workflow, usually through computerized alerts and reminders. In this case, the tool aimed to aid cardiologists in the decision to implant an artificial heart in end-stage heart failure patients.
The UX team ultimately discovered that the clinicians responsible for the implant decision did not typically find this decision challenging to make on their own. Moreover, decision-making seldom occurs in front of a computer, where clinicians typically interact with CDS tools.
So here was an initial proposal for a tool that would help users make a decision that they don’t need help making, via a computer that they wouldn’t be near while making that decision.
Case Study 2: Co-Designing a Classroom Orchestration Tool
Another case study, Co-Designing a Real-Time Classroom Orchestration Tool to Support Teacher–AI Complementarity, reinforces the idea of starting with the user rather than the technical solution or data. In designing a learning analytics (LA) tool for K-12 teachers in the United States, the design team didn’t even craft a hypothesis statement before beginning their ethnographic research. Based on existing knowledge of the complexities of LA systems, the team decided to take an investigative and collaborative approach by co-designing the tool with real end users (teachers).
A co-design approach not only enables deeper understanding of user needs and values but can also lead to unexpected innovations and real-time realizations.
“To encourage teachers to talk freely about challenges they face in AI-enhanced classrooms, without feeling constrained to those for which they believed a technical solution was currently possible, we initially avoided asking direct questions about ‘learning analytics.’ Instead, […] we asked, ‘If you could have any superpowers you wanted, to help you do your job, what would they be?’”
The team discusses the fact that beginning with and designing for user needs occasionally led them away from common LA design patterns and visualizations and instead manifested the need for novel approaches. In fact, there were moments when teachers responded positively to particular visualizations but then changed their minds when prompted to explain how they might actually use these visualizations in the classroom context.
These case studies highlight the significance of proximity to end users and cultivation of a deep understanding of their workflows and routines, which reinforces the importance of involving UX teams early in the development process to create more successful ML products.
Successful ML Solutions Start With UX
The recommendation for organizations working with ML technologies is to actively involve UX designers from the very beginning, fostering open communication between designers and end users.
The case studies demonstrate that employing human-centered design and directly observing users in their natural environments can reveal essential insights and prevent costly missteps in ML development. Engaging UX teams at the beginning of the process not only ensures that ML products are functional but also genuinely cater to user needs, resulting in saved time, money and resources.
So long as we’re creating these intelligent machines, the secret to making them better is prioritizing the human experience over the artificial intelligence. By embracing user-centered design from the start, we can create ML products that real people really want to use.
A Fusion publication. We are employees of UHG and these views are our own and not those of the company nor its affiliates.
