Title:What Would You Ask the Machine Learning Model? Identification of User Needs for Model Explanations Based on Human-Model Conversations

Abstract:Recently we see a rising number of methods in the field of eXplainable Artificial Intelligence. To our surprise, their development is driven by model developers rather than a study of needs for human end users. To answer the question "What would a human operator like to ask the ML model?" we propose a conversational system explaining decisions of the predictive model. In this experiment, we implement a chatbot called dr_ant and train a model predicting survival odds on Titanic. People can talk to dr_ant about the model to understand the rationale behind its predictions. Having collected a corpus of 1000+ dialogues, we analyse the most common types of questions that users would like to ask. To our knowledge, it is the first study of needs for human operators in the context of conversations with an ML model. It is also a first study which uses a conversational system for interactive exploration of a predictive model trained on tabular data.