A Zero-Shot Language Agent for Computer Control with Structured Reflection
This addresses the problem of limited adaptability in computer control agents for AI researchers, offering a zero-shot approach that reduces reliance on expert data, though it is incremental as it builds on existing agent frameworks.
The paper tackles the challenge of enabling language agents to autonomously learn and improve computer control without expert traces, by introducing a zero-shot agent that uses structured reflection to identify and learn from mistakes. The result shows that on easy MiniWoB++ tasks, it often outperforms recent state-of-the-art models with more efficient reasoning, and on complex tasks, it performs on par with prior best models despite lacking expert traces or additional screen information.
Large language models (LLMs) have shown increasing capacity at planning and executing a high-level goal in a live computer environment (e.g. MiniWoB++). To perform a task, recent works often require a model to learn from trace examples of the task via either supervised learning or few/many-shot prompting. Without these trace examples, it remains a challenge how an agent can autonomously learn and improve its control on a computer, which limits the ability of an agent to perform a new task. We approach this problem with a zero-shot agent that requires no given expert traces. Our agent plans for executable actions on a partially observed environment, and iteratively progresses a task by identifying and learning from its mistakes via self-reflection and structured thought management. On the easy tasks of MiniWoB++, we show that our zero-shot agent often outperforms recent SoTAs, with more efficient reasoning. For tasks with more complexity, our reflective agent performs on par with prior best models, even though previous works had the advantages of accessing expert traces or additional screen information.