LIMEADE: A General Framework for Explanation-Based Human Tuning of Opaque Machine Learners

Research in human-centered AI has shown the benefits of systems that can explain their predictions. Methods that allow humans to tune a model in response to the explanations are similarly useful. While both capabilities are well-developed for transparent learning models (e.g., linear models and GA2Ms), and recent techniques (e.g., LIME and SHAP) can generate explanations for opaque models, no method for tuning opaque models in response to explanations has been user-tested to date. This paper introduces LIMEADE, a general framework for tuning an arbitrary machine learning model based on an explanation of the model's prediction. We demonstrate the generality of our approach with two case studies. First, we successfully utilize LIMEADE for the human tuning of opaque image classifiers. Second, we apply our framework to a neural recommender system for scientific papers on a public website and report on a user study showing that our framework leads to significantly higher perceived user control, trust, and satisfaction. Analyzing 300 user logs from our publicly-deployed website, we uncover a tradeoff between canonical greedy explanations and diverse explanations that better facilitate human tuning.