“We recently published an article by Robert Schumacher and Gavin Lew—a frequent contributor to Ask UXmatters—titled ‘Artificial Intelligence, Supervised Learning, and User Experience,’ which explores some issues relating to this discussion in depth.”
“Another area in which UX skills can help is in combating bias,” suggests Adrian. “The problems that arise from biased training data are common and well known. Take a look at Karen Hao’s MIT Technology Review article ‘This Is How AI Bias Really Happens—and Why It’s So Hard to Fix.’ Our user-research practices and awareness of cognitive biases can sometimes be very useful in helping organizations to understand issues of bias and bring them to the surface within an organization.”
Setting Clear Goals for AI Applications
Even as development teams use AI in more and more applications, a common misunderstanding persists: That an AI technique that is successful in one arena can be applied in a different arena and result in similar benefits. Unfortunately, this is not true. Companies build AI systems with specific goals in mind—for example, to discover fraudulent transactions, filter out spam email messages, recommend products on a retail Web site, or recognize speech. The AI algorithms that find fraudulent transactions won’t be helpful in recognizing spoken instructions.
AI systems generate very focused results, so it is important to define the nuances of the system goals in detail. The first thing a UX designer can do to support the creation of an AI-assisted product is to ensure that a product team is building an AI system with the correct goals in mind. It is vital that the team has clear goals. Otherwise, the system might find the right answers to the wrong questions!
One way in which UX designers can assist in the creation of AI-assisted applications that leverage machine learning is to vet the training data or user feedback. In addition to building an AI system with the correct goals in mind, it is critical to provide good training data to the system. Garbage in, garbage out. If the training data does not match what the system would encounter in real life, it is unlikely that the system would perform well in use. UX designers can help AI designers to create or find effective training data by establishing a better understanding of the user, their scenarios, and their data, and thereby help AI-assisted applications to perform significantly better.
Design Best Practices for AI Applications
“Best practices for designing artificial-intelligence applications are still developing, so it is crucial for the UX community to share how we are leveraging and designing for these new capabilities,” replies Andrew. “Recently, I’ve been seeing more examples of using machine learning to enhance the user experience. For example, in Google Drive, the Quick Access feature suggests several documents that the user might want to open. By default, the user’s recent actions determine the order in which documents appear in the Quick Access section. But over time, the system might make suggestions based on frequency of use, time of day, the user’s location, or permissions that the user grants to the system.
“One of the most critical design heuristics for AI-driven applications is the necessity of being transparent with the user. We must keep the user informed of AI-driven decisions and allow the user to maintain control.”