Reproduce “From Occupations to Tasks: A New Perspective on Automatability Prediction Using BERT” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Dawei Xu, Haoran Yang, Marian-Andrei Rizoiu, Guandong Xu

Researchers from the University of Technology Sydney developed a BERT-based model to predict the automatability of individual tasks, classifying them as substitution, complementarity, or negligibility. This granular approach, validated against a comprehensive annotated dataset, reveals that 25.1% of O*NET occupations face a substantial risk of full substitution, while 61.8% are more likely to experience partial automation.

Key insights

  • Task automatability can be accurately apprehended by extracting contextual and semantic information from textual task statements using advanced NLP models like BERT.
  • Strategic data augmentation, even with large language models such as GPT-4, significantly improves the precision and robustness of deep learning models for classification tasks with inherent class imbalance.
  • The interpretability of transformer models through attention weights can visually corroborate theoretical understanding of automation bottlenecks (e.g., social interaction, creativity) by highlighting influential terms in task descriptions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.