To Ask or Not to Ask: Learning to Require Human Feedback

2025-10-09

AI Summary

Existing decision-support systems (e.g., Learning to Defer) treat human experts and predictive models as mutually exclusive decision-makers, permitting experts to provide only final predictions—thus precluding fine-grained human-AI collaboration. Method: We propose Learning to Ask (LtA), the first framework to systematically model two core questions: *when* to query an expert and *how* to incorporate their feedback. LtA departs from the exclusive-decision paradigm via a dual-module architecture supporting both sequential and joint training, and introduces a surrogate loss function with formal consistency guarantees. Contribution/Results: Extensive experiments on synthetic and real-world expert-annotated datasets demonstrate that LtA significantly improves collaborative accuracy and robustness over baselines. It enables more flexible, interpretable, and verifiable human-AI co-decision making for classification tasks—advancing beyond rigid deferral-based paradigms toward adaptive, feedback-driven collaboration.

Technology Category

• Humans and AI: Learning Human Values and Preferences
• Search and Optimization: Learning to Search
• Reasoning under Uncertainty: Sequential Decision Making

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Humans versus LLMs for data annotation and labeling
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Developing decision-support systems that complement human performance in classification tasks remains an open challenge. A popular approach, Learning to Defer (LtD), allows a Machine Learning (ML) model to pass difficult cases to a human expert. However, LtD treats humans and ML models as mutually exclusive decision-makers, restricting the expert contribution to mere predictions. To address this limitation, we propose Learning to Ask (LtA), a new framework that handles both when and how to incorporate expert input in an ML model. LtA is based on a two-part architecture: a standard ML model and an enriched model trained with additional expert human feedback, with a formally optimal strategy for selecting when to query the enriched model. We provide two practical implementations of LtA: a sequential approach, which trains the models in stages, and a joint approach, which optimises them simultaneously. For the latter, we design surrogate losses with realisable-consistency guarantees. Our experiments with synthetic and real expert data demonstrate that LtA provides a more flexible and powerful foundation for effective human-AI collaboration.

Problem

• Developing systems that complement human performance in classification tasks
• Addressing limitations of Learning to Defer by incorporating expert feedback
• Handling when and how to integrate human input into ML models

Innovation

• Proposes Learning to Ask framework for human-AI collaboration
• Uses two-part architecture with enriched model and feedback
• Implements sequential and joint training approaches with guarantees