Exposing Limitations of Language Model Agents in Sequential-Task Compositions on the Web

TL;DR

This paper introduces a new benchmark, CompWoB, to evaluate language model agents on complex web tasks, revealing significant performance gaps in compositional scenarios and proposing a new model that surpasses human performance on some benchmarks.

Contribution

The paper presents CompWoB, a challenging compositional web automation benchmark, and introduces HTML-T5++, a model that outperforms humans on MiniWoB and shows improved generalization.

Findings

Prompted LMAs' success drops from 94% to 24.9% on compositional tasks.

Transferred LMAs' success drops from 85.4% to 54.8%.

HTML-T5++ achieves 95.2% on MiniWoB and 61.5% zero-shot on CompWoB.

Abstract

Language model agents (LMA) recently emerged as a promising paradigm on muti-step decision making tasks, often outperforming humans and other reinforcement learning agents. Despite the promise, their performance on real-world applications that often involve combinations of tasks is still underexplored. In this work, we introduce a new benchmark, called CompWoB -- 50 new compositional web automation tasks reflecting more realistic assumptions. We show that while existing prompted LMAs (gpt-3.5-turbo or gpt-4) achieve 94.0% average success rate on base tasks, their performance degrades to 24.9% success rate on compositional tasks. On the other hand, transferred LMAs (finetuned only on base tasks) show less generalization gap, dropping from 85.4% to 54.8%. By balancing data distribution across tasks, we train a new model, HTML-T5++, that surpasses human-level performance (95.2%) on MiniWoB, and achieves the best zero-shot performance on CompWoB (61.5%). While these highlight the promise of small-scale finetuned and transferred models for task compositionality, their performance further degrades under different instruction compositions changing combinational order. In contrast to the recent remarkable success of LMA, our benchmark and detailed analysis emphasize the necessity of building LMAs that are robust and generalizable to task compositionality for real-world deployment.