LLM-Powered GUI Agents: Shaping AI Behavior with Human Values
Analysis
This research explores a fascinating new dimension in the design of Generative AI Agents: incorporating human values and preferences to guide their decision-making. The study's controlled environment allows for systematic analysis, providing invaluable insights into how these factors shape agent behavior in everyday online tasks, making them more aligned to human intentions. It's an exciting step towards more personalized and trustworthy AI experiences!
Key Takeaways
- •User preferences, injected as "personas," successfully influenced the decision-making of LLM-powered Agents.
- •Agents exhibited an efficiency Bias when no user preference or value guidance was provided, often taking the shortest path.
- •Dominant interface cues, like discounts, could sometimes override preference-based behavior, impacting outcome consistency.
Reference / Citation
View Original"Our results show that preference and value-infused prompts consistently guided agents toward outcomes that reflected these preferences and values."
A
ArXiv HCIJan 26, 2026 05:00
* Cited for critical analysis under Article 32.