AI Agents Should Teach Users, Not Just Ask Questions
AI agents often assume users know exactly what they want. New research suggests these tools should help people learn and form preferences instead. This could make AI assistants much more helpful for everyday tasks.
Researchers from arXiv cs.AI published a study arguing that AI agents should do more than just ask clarifying questions. These tools typically assume users have well-defined preferences, but most people don't. The study suggests AI should help users learn about their options, making it easier to form preferences.
Imagine trying to pick a new phone. An AI assistant shouldn't just ask, 'What features matter most?' Instead, it could explain differences in cameras, batteries, and processors, helping you decide. This approach would make AI tools more useful for people who aren't experts in every topic.
The paper, titled "Beyond expert users: agents should help users construct preferences, not just elicit them," argues that the common assumption of an "expert user" is unrealistic. Users often lack the domain knowledge needed to have fully formed preferences. When an agent asks about a specific feature, the user may be unable to answer without first learning some domain knowledge—for example, through examples or explanations. The research formalizes this principle, suggesting that agents should help users construct preferences rather than simply eliciting them.
You can try this idea today with AI tools like Claude or ChatGPT. Ask them to explain a complex topic, like choosing a laptop, and then have them help you compare options. This way, you'll learn as you go, making better decisions with AI's help.