Human-AI Collaboration: Towards Socially-Guided Machine Learning Q. Vera Liao IBM Research AI vera.liao@ibm.com Rachel K.E. Bellamy IBM Research AI rachel@us.ibm.com Michael Muller IBM Research AI michael_muller@us.ibm.com Heloisa Candello IBM Research hcandello@br.ibm.com ABSTRACT 1 As Machine Learning (ML) systems become increasingly ubiquitous, capable and autonomous, it has become essential to take a human-centered view to consider how people’s interactions with ML systems, including the effort to develop and evolve ML systems, impact their work practices, wellbeing and the social-organizational environment. Built on our work on human-agent collaboration, we suggest a change of perspective, by considering human(s) and the ML model(s) they interact with as a team engaging in collaborative work. With that, we can apply metaphoric thinking based on team collaboration to inform the design of human-model interactions and rethink the collective goals to be embedded in computational models. Based on pillars of Computer-Supported Cooperative Work (CSCW) research, we point to three areas for future research into technologies to support human-model interaction as collaborative work: (1) model training as knowledge sharing; (2) interactions as communication actions; and (3) coordination for better collaboration and construction of trust. INTRODUCTION Human interactions with Machine Learning systems are growing to be ubiquitous as driven by two trends. One is the current movement for “democratizing ML” by lowering the barrier of entry to ML