Team Reactions to Voiced Agent Instructions in a Pervasive Game Stuart Moran, Nadia Pantidi, Khaled Bachour Joel E. Fischer, Martin Flintham, Tom Rodden Mixed Reality Lab University of Nottingham, UK {firstname.surname}@nottingham.ac.uk Simon Evans and Simon Johnson SlingShot Bristol, UK {firstname.surname}@slingshoteffect.co.uk ABSTRACT The assumed role of humans as controllers and instructors of machines is changing. As systems become more complex and incomprehensible to humans, it will be increasingly necessary for us to place confidence in intelligent interfaces and follow their instructions and recommendations. This type of relationship becomes particularly intricate when we consider significant numbers of humans and agents working together in collectives. While instruction-based interfaces and agents already exist, our understanding of them within the field of Human-Computer Interaction is still limited. As such, we developed a large-scale pervasive game called ‘Cargo’, where a semi-autonomous ruled-based agent distributes a number of text-to-speech instructions to multiple teams of players via their mobile phone as an interface. We describe how people received, negotiated and acted upon the instructions in the game both individually and as a team and how players initial plans and expectations shaped their understanding of the instructions. Author Keywords Human-agent interaction; in situ; instructions; pervasive games ACM Classification Keywords H.5.2 [Information Interfaces and Presentation]: User Interfaces - Interaction styles. INTRODUCTION Intelligent interfaces continue to revolutionize the relationship between humans and computers. A product of this is that the default and comfortable position of users instructing and controlling computers is changing. As we continue to offload complex tasks and responsibility to machines, we will more frequently experience systems and interfaces that deliver instructions for us to follow. This role reversal becomes increasingly important in highly complex, dynamic and demanding time critical circumstances such as disaster response. Such complexity will require groups of humans and intelligent interfaces to seamlessly exchange instructions in order to effectively co-operate. While expert, diagnostic, support and recommender systems have been successfully deployed for some time, to date, exploring and understanding the instruction of users in HCI is still limited [26]. Without this understanding, agent based instruction systems may function inefficiently, or at worst, fail to achieve compliance from humans. This would be of particular consequence in safety/time critical systems. In light of this, our goal in this paper is to probe a number of issues centered on how people respond to instructions from a semi-autonomous agent-based system. In particular, we are interested in investigating:  How do people make sense of unstructured or ambiguous information from a software agent?  How do groups of people work together while under agent instruction?  What makes people trust instructions, and what might cause them to not comply? To explore these issues, we have developed a large-scale, ‘in the wild’ pervasive game called Cargo. Within the game, a semi-autonomous rule-based agent distributes a number of voice instructions to multiple teams of players via their mobile phones as an interface. This approach allowed us to study the visceral reactions and interactions of a large number of human teams receiving instructions in situ, in real time and under pressure. The results highlight the process by which players interpret instructions, negotiate their meanings, as well as how these are intertwined with the relationship between the players and the agent. We conclude the paper with a number of emerging design recommendations to consider when creating agent systems which instruct teams of humans. BACKGROUND Within the field of Human-Computer Interaction, there is a wide range of literature on the interaction between humans and agents. However the tendency is to focus on a one-to- one relationship. The functionality and practical application of such collaboration is limited to specific types of problems and tasks. When we begin to consider increasingly large-scale complex environments and challenges, such relationships start to lose value. Hence, one growing field of research in human-agent interaction is Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. IUI’13, March 19–22, 2013, Santa Monica, CA, USA. Copyright © 2013 ACM 978-1-4503-1965-2/13/03...$15.00.