ChatGPT for Research: How to Find Papers and Study Smarter

Why Read This from an HCI Perspective

This article helps you see AI not as a ‘high-performance tool,’ but as an interaction problem—how people learn to trust, verify, and correct it. For HCI/UX practitioners and researchers, this matters because it goes beyond simple feature comparisons and pushes you to think about how to design users’ intervention paths and failure situations. This is especially meaningful when applying AI agents to safety-critical services or agentic AI.

CIT's Commentary

The most dangerous moment when working with AI agents isn’t when the model is wrong—it’s when users can’t easily notice that it’s wrong. That’s why the interface shouldn’t be a window that simply looks nice; it should be more like a dashboard that lets users read the system’s state. For example, it should be clear whether the AI is confident, what evidence it used to answer, and where the user can pause and re-instruct the system. In industry, speed and automation are important, but that also makes the research questions about failure modes and intervention pathways even sharper. In service contexts like those of Naver or Kakao, rather than transplanting global paper frameworks as-is, you need to consider mobile-first usage patterns and interactions with a high social density. Korean users often expect fast responses and smoothness, so transparency can feel like ‘inconvenience.’ How to resolve that tension is a key design challenge.

Questions to Consider While Reading

• Q.What core information would help users understand an AI agent’s state at a glance?
• Q.As the level of automation increases, where and in what form should user intervention be possible?
• Q.In the context of Korea’s mobile and social services, how should the balance between transparency and smoothness be designed differently?