Paper Reading #5: Ripple Effects of an Embedded Social Agent: A Field Study of a Social Robot in the Workplace

Ripple Effects of an Embedded Social Agent:
A Field Study of a Social Robot in the Workplace

Min Kyung Lee(1), Sara Kiesler(1), Jodi Forlizzi(1), Paul Rybski(2)

Human-Computer Interaction Institute1, Robotics Institute2

Carnegie Mellon University

5000 Forbes Ave, Pittsburgh, PA 15213 USA

{mklee, kiesler, forlizzi, rybski}@cs.cmu.edu

Author Bios:

Min Kyung Lee

• Ph.D. student in Human-Computer Interaction Institute at Carnegie Mellon University.

• Sara Kiesler is her Ph.D. adviser.

Sara Kiesler

• Hillman Professor of Computer Science and Human-Computer Interaction at Carnegie Mellon University.

• Social and behavioral aspects of computers and computer-based communication technologies.

Jodi Forlizzi

• Associate Professor, School of Design and Human-Computer Interaction Institute,

• The A. Nico Habermannn Chair in the School of Computer Science

Paul Rybski

• Systems Scientist – RI, ECE Carnegie Mellon

• PhD, 2003 Computer Science and Engineering UMN

Summary:

A social robot called Snackbot delivered snacks to people in an office and engaged the recipients with small talk as it delivered their snacks.  The authors of the paper wanted to see how people interact with and perceive Snackbot. There has not been a study on if robotic agent's social skills begin to annoy people interacting with the agent for an extended amount of time.  Snackbot was 4.5 feet tall and had a face that went through many tests to be as pleasing to people as possible.  Snackbot was remotely controlled by an operator on a laptop and moved autonomously and had a speech recognition system. Unless an issue occurred, the operator never took control of the robot.  The most common times the operator had to step in was if a participant asked a question that Snackbot did not know how to answer.  Interactions started with Snackbot identifying the participant, engaging in small talk, giving them their snack and then leaving in a socially acceptable manner.  Snackbot was given categories and topics to engage the participant in based on the unique participant, what snack they chose, seasonal or apologies for breaking down since it happened often.   

Related work not referenced in the paper:

• Involving Users in the Design of a Mobile Office Robot
• The robot in this paper was made to assist in fetching objects for physically impaired people in the work place.  The authors were interested in how people in an office would perceive a robot over an extended period of time exactly like with Snackbot. It was not autonomous nor did it have speech recognition like Snackbot.
• The USUS Evaluation Framework for Human-Robot Interaction 
• The authors identify what a robot must be and be capable of doing in order for it to have a seamless integration into the work place. It pretty much lists out the same topics that Snackbot was meant to test, but they do not provide any field study data.
• Studying the acceptance of a robotic agent by elderly users
• The point of this paper is to do a filed study to see if a robot could be socially accepted among elderly people as it helps their day to day lives.  It may have a more focused audience of study than Snackbot, but the idea is exactly the same. Snackbot is not at all a novel idea.
• Human–Robot Interaction in Rescue Robotics
• This paper is about creating more environmentally aware rescue robots that do not take many people to control and to also be socially accepted as a robotic agent.  
• Whose Job Is It Anyway? A Study of Human–Robot Interaction in a Collaborative Task
• This paper explains how the time for collaboration with humanoid like robots in the work place is not far in the future. It describes exactly what all these other papers are doing, such as Snackbot, which is slowly perfecting the art of creating a robot so life like that it can easily be socially accepted by humans.
• Final Report for the DARPA/NSF Interdisciplinary Study on Human–Robot Interaction
• This paper explains how roboticists collaborated with psychologists, sociologists, cognitive scientists, communication experts and HCI specialist to setup a common ground for them all to work together in making a more humanoid robot agent to interact on a daily basis with people in many different environments. 
• A Methodological Variation for Acceptance Evaluation of HumanRobot Interaction in Public Places  
• This paper has the exact same idea that the experiment involving Snackbot was in tended to do only for public uses.  It does not have any field data though.
• A Social Informatics Approach to Human–Robot Interaction With a Service Social Robot
• The authors describe experiments that have been done involving the benefits of emotional and social robots against emotionless robots in work environments like an autonomous cherry picker. They describe how they believe robots should approach social intelligence in order to be successful in society.
•  Teamwork-Centered Autonomy for Extended Human-Agent Interaction in Space Applications
• Astronaut's jobs working in space is incredibly dangerous and would greatly improve if they could work with autonomous robots that can intelligently do jobs that are too dangerous or slow for humans to do in space.
• Enjoyment, Intention to Use And Actual Use of a Conversational Robot by Elderly People
• Not all social agents have to be used strictly as workers.  They will also be used as potential company for people like the elderly who can't be left alone too long, but can't afford a full time nurse either.  This robotic agents helping, a single nurse can take care of many elderly people at once.

Evaluation:

Snackbot delivered snacks to 16 offices on one floor of a building at a US university.  There were 21 participants, 8 women and 13 men ranging from 22-51. There were 11 graduate students, eight staff, one doctor and one faculty member. They were all part of the computer science department, but half had no programming experience.  Snackbot delivered snacks between 2:30-4 p.m. on Mondays, Wednesdays and Fridays for four months.  All data was subjective since the experiment was to see if the participants would accept Snackbot socially. 175 interactions were recorded and 161 interactions were videoed. Participants interacted with Snackbot an average of 9 times over their two months of deliveries. Participants were not told that they would be given snacks by a robot when they signed up. There was mostly qualitative data since the impressions Snackbot had on the participants were mostly collected in end experiment interviews. Five participants saw Snackbot as a failed person because it broke down every once on a while and sometimes interacted inhumanly such as asking a door if it could pass or pausing for extended periods during a conversation.  These breakdowns helped the participants keep the sense that Snackbot is in fact a robot.  From the interviews it was found that most of the participants got used to Snackbot and would sometimes go out of their way to make sure they were in their office when it arrived.  They would call days Snackbot would come around "Snackbot day".  When participants were busy and did not want to talk to Snackbot, they said they later felt bad about getting irritated with Snackbot. The majority of the participants felt emotions for Snackbot whenever it "humiliated" itself when it would breakdown.  Some participants got jealous of other participants because they appeared to get more attention from Snackbot when in actuality they were not.  An unintended effect of Snackbot called the ripple effect where people that were not participating, but watched Snackbot's interactions built both negative and positive behaviors about Snackbot.  Over time the ripple effect grew and was variables like office culture were a factor and needed to be considered when deploying a social agent like Snackbot in a work environment.  Over all 75% of participants grew to enjoy interacting with Snackbot.

Discussion:

I believe that as robotics and the technology to build them increases we will begin to see more and more robotic agents in work places everywhere. This experiment was not novel and has been done before, but not at this level of computing power.  It was building off of so many other experiments that all were better than the last.  The participants were not a representation of a normal work environment considering it was a computer science department in a university and they could easily have known the authors of this paper.  Their opinions on the matter were probably far from being unbiased.  Bias is impossible to get rid of in an experiment like this, but more diverse participants are a must.