Interactive Machine Learning: Managing Information Richness in Highly Anonymized Conversation Data

Abstract

This case study focuses on an experiment analysing textual conversation data using machine learning algorithms and shows that sharing data across organisational boundaries requires anonymisation that decreases that data’s information richness. Additionally, sharing data between organisations, conducting data analytics and collaborating to create new business insight requires inter-organisational collaboration. This study shows that analysing highly anonymised and professional conversation data challenges the capabilities of artificial intelligence. Machine learning algorithms alone cannot learn the internal connections and meanings of information cues. This experiment is therefore in line with prior research in interactive machine learning where data scientists, specialists and computational agents interact. This study reveals that, alongside humans, computational agents will be important actors in collaborative networks. Thus, humans are needed in several phases of the machine learning process for facilitating and training. This calls for collaborative working in multi-disciplinary teams of data scientists and substance experts interacting with computational agents.