Automatic Discovery of Interpretable Planning Strategies

Abstract

When making decisions, people often overlook critical information or are overly swayed by irrelevant information. A common approach to mitigate these biases is to provide decision-makers, especially professionals such as medical doctors, with decision aids, such as decision trees and flowcharts. Designing effective decision aids is a difficult problem. We propose that recently developed reinforcement learning methods for discovering clever heuristics for good decision-making can be partially leveraged to assist human experts in this design process. One of the biggest remaining obstacles to leveraging the aforementioned methods is that the policies they learn are opaque to people. To solve this problem, we introduce AI-Interpret: a general method for transforming idiosyncratic policies into simple and interpretable descriptions. Our algorithm combines recent advances in imitation learning and program induction with a new clustering method for identifying a large subset of demonstrations that can be accurately described by a simple, high-performing decision rule. We evaluate our new algorithm and employ it to translate information-acquisition policies discovered through metalevel reinforcement learning. The results of large behavioral experiments showed that prividing the decision rules generated by AI-Interpret as flowcharts significantly improved people's planning strategies and decisions across three diferent classes of sequential decision problems. Moreover, another experiment revealed that this approach is significantly more effective than training people by giving them performance feedback. Finally, a series of ablation studies confirmed that AI-Interpret is critical to the discovery of interpretable decision rules. We conclude that the methods and findings presented herein are an important step towards leveraging automatic strategy discovery to improve human decision-making.

arXiv commentSubmitted to the Special Issue on Reinforcement Learning for Real Life in Machine Learning Journal (2021). Code available at https://github.com/RationalityEnhancement/InterpretableStrategyDiscovery

References in corpus (16)

• Scikit-learn: Machine Learning in Python
• An Algorithmic Perspective on Imitation Learning
• Visualizing and Understanding Atari Agents
• Verifiable Reinforcement Learning via Policy Extraction
• Programmatically Interpretable Reinforcement Learning
• Learning Task Specifications from Demonstrations
• Selecting Computations: Theory and Applications
• Explain Your Move: Understanding Agent Actions Using Specific and Relevant Feature Attribution
• Exploratory Not Explanatory: Counterfactual Analysis of Saliency Maps for Deep Reinforcement Learning
• Towards Interpretable Reinforcement Learning Using Attention Augmented Agents
• Learning to select computations
• What Did You Think Would Happen? Explaining Agent Behaviour Through Intended Outcomes
• Imitation-Projected Programmatic Reinforcement Learning
• Identifying Reasoning Flaws in Planning-Based RL Using Tree Explanations
• Towards Mixed Optimization for Reinforcement Learning with Program Synthesis
• Few-Shot Bayesian Imitation Learning with Logical Program Policies

Cited by in corpus (2)

• Explainable Deep Reinforcement Learning: State of the Art and Challenges
• Reinforcement Learning in the Wild with Maximum Likelihood-based Model Transfer