A Bandit Model for Human-Machine Decision Making with Private\n Information and Opacity

2020/07/09 by Sebastian Bordt, Bordt, Sebastian, Ulrike von Luxburg +1
Computer Science · Decision Sciences · #Advanced Bandit Algorithms Research #Auction Theory and Applications #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Machine Learning and Algorithms #Optimization and Search Problems

paper · pdf · doi:10.48550/arxiv.2007.04800

openalex publication_date 2020/07/09 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28

Abstract

Applications of machine learning inform human decision makers in a broad\nrange of tasks. The resulting problem is usually formulated in terms of a\nsingle decision maker. We argue that it should rather be described as a\ntwo-player learning problem where one player is the machine and the other the\nhuman. While both players try to optimize the final decision, the setup is\noften characterized by (1) the presence of private information and (2) opacity,\nthat is imperfect understanding between the decision makers. We prove that both\nproperties can complicate decision making considerably. A lower bound\nquantifies the worst-case hardness of optimally advising a decision maker who\nis opaque or has access to private information. An upper bound shows that a\nsimple coordination strategy is nearly minimax optimal. More efficient learning\nis possible under certain assumptions on the problem, for example that both\nplayers learn to take actions independently. Such assumptions are implicit in\nexisting literature, for example in medical applications of machine learning,\nbut have not been described or justified theoretically.\n

Citations

• The Nonstochastic Multiarmed Bandit Problem
• Contextual Bandit Algorithms with Supervised Learning Guarantees
• Sanity Checks for Saliency Maps
• Predict Responsibly: Improving Fairness and Accuracy by Learning to Defer
• What Clinicians Want: Contextualizing Explainable Machine Learning for Clinical End Use
• Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps
• Decentralized Cooperative Stochastic Bandits
• On the Utility of Learning about Humans for Human-AI Coordination
• On the ethics of algorithmic decision-making in healthcare
• Consistent Estimators for Learning to Defer to an Expert
• Algorithm aversion: People erroneously avoid algorithms after seeing them err.

Related

• Persuasion, Delegation, and Private Information in Algorithm-Assisted Decisions
• Offline Reinforcement Learning for Human-Guided Human-Machine Interaction with Private Information
• Decision-aid or Controller? Steering Human Decision Makers with Algorithms
• Bayesian Elicitation
• Decision Making in Changing Environments: Robustness, Query-Based Learning, and Differential Privacy