How to Use Brainstorming (Brainstorming) with ChatGPT

Why Read This from an HCI Perspective

This article helps you see AI not as a ‘smart tool,’ but as an interaction problem—one that includes how users understand, trust, and intervene in the system. For HCI and UX practitioners, it shifts attention away from model performance alone and encourages you to rethink why interface design—such as guidance text, status indicators, and failure handling—is so important. It’s especially useful for connecting how AI features are actually used inside a product.

CIT's Commentary

The core message of this piece is that, more than AI performance, we should prioritize ‘when users choose to trust it, when they stop, and how they can intervene.’ In other words, an interface that helps users read the system’s state can be a bigger safety net than simply having a model that answers well. For example, it becomes more important to understand what evidence the AI agent used and where the user can roll back, rather than what happens the moment the agent makes a recommendation. In Korea’s mobile and messenger environment—where screens are short, flows are fast, and immediacy is high—designing clear control points like this is particularly necessary. At the same time, it’s interesting to see an approach that uses LLMs to build UX measurement tools, but the key is to separate criteria so that convenience increases without sacrificing the rigor of measurement.

Questions to Consider While Reading

• Q.At what level should screens where AI only suggests differ from screens where it actually performs tasks on the user’s behalf?
• Q.What signals are most effective at helping users easily understand the system’s status?
• Q.When using LLMs as UX evaluation tools, how can we preserve both convenience and measurement rigor?