Weekly Highlights

The most striking commonality is that all five studies treat user experience not as a matter of convenience, but as a matter of distributing authority. When AI summarizes public opinion, it must be able to show which viewpoints were excluded. For mobile agents, the system needs to be finely tuned to decide when it should come to the foreground and when it should step back into the background. For AI companions, it must make clearer from the outset what kind of relationship it is presenting. In the same context, studies on autonomous driving interfaces and privacy consent screens show that what makes users trust the system is not a large amount of information or merely formal choice options, but a structure that users can understand quickly and that allows them to intervene when needed. Although the topics may look different on the surface, they all read as a broader effort to redesign how AI and digital systems replace or weaken human judgment.

The biggest trend these updates reveal is that the evaluation axes in HCI/UX are shifting from efficiency and accuracy toward representativeness, interpretability, and the safety of relationships. In the past, the focus was on whether the summary was good, whether tasks were faster, and whether the features were highly usable. Now, the key metrics are whether there are missing voices, what role the system plays in approaching users, and whether users can actually refuse or stop it. Especially as AI increasingly takes on not only delegated work but also emotional interaction, the interface becomes less a surface that merely displays results and more a control mechanism that positions intervention points and clarifies responsibility. Concepts such as participatory sources, timely visualization, role-based safeguards, and situation-aware explanations—all point in the same direction. In other words, user trust is not formed because the system looks perfect; it is formed when users can recognize the system’s imperfections and the possibility of bias and can correct them.

The key message for practitioners is that simply adding AI capabilities does not automatically create good UX. Elements such as representativeness checks, criteria for mode switching, boundaries in the relationship, and refusal paths should be built into product requirements from the beginning. For researchers, traditional measures like satisfaction or task success rates are no longer sufficient to fully explain today’s AI experiences—so there is a growing need for methodologies that measure the flow of understanding, trust, intervention, and recovery together. In particular, in Korea’s platform- and mobile-centered environment, factors such as notification overload, a culture of rapid experimentation, and high service integration can further amplify these issues. Rather than simply transplanting findings from overseas research, it is necessary to redesign representativeness criteria and safety mechanisms to fit domestic usage contexts. The point to watch going forward is whether interface principles for letting users know where to pause, verify, edit, and create distance—more than how naturally we make AI—become a competitive advantage across the industry.