Specifying and Interpreting Reinforcement Learning Policies through Simulatable Machine Learning

Human-AI collaborative policy synthesis is a procedure in which (1) a human initializes an autonomous agent’s behavior, (2) Reinforcement Learning improves the human specified behavior, and (3) the agent can explain the final optimized policy to the user. This paradigm leverages human expertise and facilitates a greater insight into the learned behaviors of an agent. Existing approaches to enabling collaborative policy specification are unintelligible and are not catered towards making the autonomous system accessible to a non-expert end-user. These approaches involve black box methods which are uninterpretable to humans and are not flexible to providing multiple means of interacting with an autonomous agent. In this paper, we develop a novel collaborative framework to enable humans to initialize and interpret an autonomous agent’s behavior, rooted in principles of human-centered design. Through our framework, we enable humans to specify an initial behavior model in the form of unstructured, non-technical, natural language, which we then convert to lexical decision trees. Next, we are able to leverage these human-specified policies, in the form of lexical decision trees, to warm-start reinforcement learning and further allow the agent to optimize the policies through reinforcement learning. Finally, to close the loop on human-specification, we produce explanations of the final learned policy, in multiple modalities (e.g., Tree-, Language-, and Program-based), to provide the user a final depiction about the learned policy of the autonomous agent. We validate our approach by showing that our model is able to produce >80% translation accuracy, and that policies initialized by a human are able to successfully warm-start RL. We then conduct a novel human-subjects study quantifying the relative subjective and objective benefits of varying XAI modalities for explaining learned policies to end-users, in terms of usability and interpretability and identify the circumstances that influence these measures. Our quantitative findings and qualitative analysis emphasize the need for personalized explainable systems that can facilitate user-centric policy explanations for a variety of end-users.

[1]  Pierre Sermanet,et al.  Grounding Language in Play , 2020, ArXiv.

[2]  Chitta Baral,et al.  Language-Conditioned Imitation Learning for Robot Manipulation Tasks , 2020, NeurIPS.

[3]  Shen Li,et al.  Bayesian Inference of Temporal Task Specifications from Demonstrations , 2018, NeurIPS.

[4]  Demis Hassabis,et al.  Grounded Language Learning in a Simulated 3D World , 2017, ArXiv.

[5]  Anna Goldenberg,et al.  What Clinicians Want: Contextualizing Explainable Machine Learning for Clinical End Use , 2019, MLHC.

[6]  Sorin Grigorescu,et al.  A Survey of Deep Learning Techniques for Autonomous Driving , 2020, J. Field Robotics.

[7]  Percy Liang,et al.  Data Recombination for Neural Semantic Parsing , 2016, ACL.

[8]  R. Mayer,et al.  Three Facets of Visual and Verbal Learners: Cognitive Ability, Cognitive Style, and Learning Preference. , 2003 .

[9]  Shie Mannor,et al.  Graying the black box: Understanding DQNs , 2016, ICML.

[10]  Fred D. Davis Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology , 1989, MIS Q..

[11]  B. Jones BOUNDED RATIONALITY , 1999 .

[12]  Mirella Lapata,et al.  Language to Logical Form with Neural Attention , 2016, ACL.

[13]  Andrew Zisserman,et al.  Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps , 2013, ICLR.

[14]  Andrea Lockerd Thomaz,et al.  Robot Learning from Human Teachers , 2014, Robot Learning from Human Teachers.

[15]  Luke S. Zettlemoyer,et al.  Weakly Supervised Learning of Semantic Parsers for Mapping Instructions to Actions , 2013, TACL.

[16]  H. Friedrich,et al.  In: Probramming by Demonstration vs. Learning from Examples Workshop at Ml'95 Obtaining Good Performance from a Bad Teacher , 1995 .

[17]  Sheila A. McIlraith,et al.  Using Reward Machines for High-Level Task Specification and Decomposition in Reinforcement Learning , 2018, ICML.

[18]  Paul Smolensky,et al.  Connectionist AI, symbolic AI, and the brain , 1987, Artificial Intelligence Review.

[19]  Yunyao Li,et al.  Who needs to know what, when?: Broadening the Explainable AI (XAI) Design Space by Looking at Explanations Across the AI Lifecycle , 2021, Conference on Designing Interactive Systems.

[20]  Yoshua Bengio,et al.  Neural Machine Translation by Jointly Learning to Align and Translate , 2014, ICLR.

[21]  Yunfeng Zhang,et al.  Effect of confidence and explanation on accuracy and trust calibration in AI-assisted decision making , 2020, FAT*.

[22]  C. Flavián,et al.  Integrating trust and personal values into the Technology Acceptance Model: The case of e-government services adoption , 2012 .

[23]  Matthew Gombolay,et al.  Learning from Suboptimal Demonstration via Self-Supervised Reward Regression , 2020, ArXiv.

[24]  Quoc V. Le,et al.  Sequence to Sequence Learning with Neural Networks , 2014, NIPS.

[25]  Q. Liao,et al.  Questioning the AI: Informing Design Practices for Explainable AI User Experiences , 2020, CHI.

[26]  Matthew R. Walter,et al.  Understanding Natural Language Commands for Robotic Navigation and Mobile Manipulation , 2011, AAAI.

[27]  Stefanie Tellex,et al.  Accurately and Efficiently Interpreting Human-Robot Instructions of Varying Granularities , 2017, Robotics: Science and Systems.

[28]  Hod Lipson,et al.  Understanding Neural Networks Through Deep Visualization , 2015, ArXiv.

[29]  Giulio Sandini,et al.  Humanizing Human-Robot Interaction: On the Importance of Mutual Understanding , 2018, IEEE Technology and Society Magazine.

[30]  Ross A. Knepper,et al.  Following High-level Navigation Instructions on a Simulated Quadcopter with Imitation Learning , 2018, Robotics: Science and Systems.

[31]  Mark O. Riedl,et al.  Automated rationale generation: a technique for explainable AI and its effects on human perceptions , 2019, IUI.

[32]  Petter Nilsson,et al.  Toward Specification-Guided Active Mars Exploration for Cooperative Robot Teams , 2018, Robotics: Science and Systems.

[33]  Been Kim,et al.  Towards A Rigorous Science of Interpretable Machine Learning , 2017, 1702.08608.

[34]  Andrew Bennett,et al.  Mapping Instructions to Actions in 3D Environments with Visual Goal Prediction , 2018, EMNLP.

[35]  Dan Klein,et al.  Modular Multitask Reinforcement Learning with Policy Sketches , 2016, ICML.

[36]  Li Wang,et al.  The Robotarium: A remotely accessible swarm robotics research testbed , 2016, 2017 IEEE International Conference on Robotics and Automation (ICRA).

[37]  Eric Horvitz,et al.  Updates in Human-AI Teams: Understanding and Addressing the Performance/Compatibility Tradeoff , 2019, AAAI.

[38]  Alberto Suárez,et al.  Globally Optimal Fuzzy Decision Trees for Classification and Regression , 1999, IEEE Trans. Pattern Anal. Mach. Intell..

[39]  Hadas Kress-Gazit,et al.  Translating Structured English to Robot Controllers , 2008, Adv. Robotics.

[40]  Zachary Chase Lipton The mythos of model interpretability , 2016, ACM Queue.

[41]  Demetra Evangelou,et al.  Orientations and motivations: Are you a “people person,” a “thing person,” or both? , 2012 .

[42]  Sergey Levine,et al.  Learning Complex Dexterous Manipulation with Deep Reinforcement Learning and Demonstrations , 2017, Robotics: Science and Systems.

[43]  J. L. Peterson,et al.  Deep Neural Network Initialization With Decision Trees , 2017, IEEE Transactions on Neural Networks and Learning Systems.

[44]  Alvin Cheung,et al.  Learning a Neural Semantic Parser from User Feedback , 2017, ACL.

[45]  Christian Muise,et al.  Evaluating the Interpretability of the Knowledge Compilation Map: Communicating Logical Statements Effectively , 2019, IJCAI.

[46]  Diyi Yang,et al.  Hierarchical Attention Networks for Document Classification , 2016, NAACL.

[47]  Luke S. Zettlemoyer,et al.  Learning to Parse Natural Language Commands to a Robot Control System , 2012, ISER.

[48]  Peter Stone,et al.  Improving Grounded Natural Language Understanding through Human-Robot Dialog , 2019, 2019 International Conference on Robotics and Automation (ICRA).

[49]  Sung-Hyun Son,et al.  Optimization Methods for Interpretable Differentiable Decision Trees Applied to Reinforcement Learning , 2020, AISTATS.

[50]  Pushmeet Kohli,et al.  Learning to Understand Goal Specifications by Modelling Reward , 2018, ICLR.

[51]  Stefanie Tellex,et al.  Sequence-to-Sequence Language Grounding of Non-Markovian Task Specifications , 2018, Robotics: Science and Systems.

[52]  Laura G. Militello,et al.  Macrocognition, Mental Models, and Cognitive Task Analysis Methodology , 2017 .

[53]  Matthew C. Gombolay,et al.  ProLoNets: Neural-encoding Human Experts' Domain Knowledge to Warm Start Reinforcement Learning , 2019, ArXiv.

[54]  Ross A. Knepper,et al.  Learning to Map Natural Language Instructions to Physical Quadcopter Control using Simulated Flight , 2019, CoRL.

[55]  Tim Miller,et al.  Explainable Reinforcement Learning Through a Causal Lens , 2019, AAAI.

[56]  Pieter Abbeel,et al.  Apprenticeship learning via inverse reinforcement learning , 2004, ICML.

[57]  Abhishek Das,et al.  Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization , 2016, 2017 IEEE International Conference on Computer Vision (ICCV).

[58]  Wojciech Samek,et al.  Methods for interpreting and understanding deep neural networks , 2017, Digit. Signal Process..

[59]  Hadas Kress-Gazit,et al.  Robot-Initiated Specification Repair through Grounded Language Interaction , 2017, ArXiv.

[60]  Maya Cakmak,et al.  Power to the People: The Role of Humans in Interactive Machine Learning , 2014, AI Mag..

[61]  Thomas G. Dietterich Hierarchical Reinforcement Learning with the MAXQ Value Function Decomposition , 1999, J. Artif. Intell. Res..

[62]  Sheila A. McIlraith,et al.  Teaching Multiple Tasks to an RL Agent using LTL , 2018, AAMAS.

[63]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.