AI-infused applications and recommender systems are much more prolific in recent years, following technological advances in machine learning and artificial intelligence. This partially explains the increased interest of the HCI research community to investigate how to design systems that will optimise Human-AI collaboration. However, most studies predominately focus on the work environment in which domain experts interact with AI systems to solve relatively well-defined tasks. In this paper, we explore Human AI collaboration involving non-domain experts in the everyday task of waste sorting for recycling. We conducted a study in which 35 participants completed a waste shorting task using Waste Wizard, an AI-infused prototype that we developed. We present the results of our qualitative analysis based on follow-up interviews aiming to understand how our participants perceived the role of the AI system during their collaboration. Our results show four distinct roles (Mirror, Assistant, Guide, and Oracle) that our participants ascribed to the AI system.