PLACO Framework for Cost-Effective Human-AI Team Performance
A new multi-stage framework called PLACO aims to improve performance in Human-AI teams while reducing costs. The framework addresses classification tasks where a human and a model combine outputs. Prior work used Bayes rule assuming conditional independence between human and model given ground truth, combining a deterministic human labeler with a probabilistic model. PLACO extends this by introducing a multi-stage approach that considers instance-level model confidence and class-level human expertise, optimizing for cost-effectiveness. The paper is published on arXiv under identifier 2605.08388.
Key facts
- PLACO is a multi-stage framework for Human-AI teams.
- It focuses on cost-effective performance in classification tasks.
- The framework combines a deterministic human labeler and a probabilistic model.
- Prior work used Bayes rule with conditional independence assumption.
- PLACO uses instance-level model confidence and class-level human expertise.
- The paper is available on arXiv with ID 2605.08388.
- Human-AI teams are important when neither alone achieves optimal performance.
- Generative AI has expanded Human-AI team tasks to writing and algorithm development.
Entities
Institutions
- arXiv