WHAT’S THE BEST ROLE FOR A ROBOT?

Cybernetic Models of Existing and Proposed Human-Robot Interaction Structures

Victoria Groom

Department of Communication, Stanford University, 450 Serra Mall, Stanford, U.S.A.

Keywords: Cybernetics, human-robot interaction, robot teams, social robotics.

Abstract: Robots intended for human-robot interaction are currently designed to fill simple roles, such as task

completer or tool. The design emphasis remains on the robot and not the interaction, as designers have

failed to recognize the influence of robots on human behavior. Cybernetic models are used to critique

existing models and provide revised models of interaction that delineate the paths of social feedback

generated by the robot. Proposed robot roles are modeled and evaluated. Features that need to be developed

for robots to succeed in these roles are identified and the challenges of developing these features are

discussed.

1 INTRODUCTION

Human-robot interaction (HRI) is the study of

humans’ interactions with robots. While the field of

robotics focuses primarily on the technological

development of robots, HRI focuses not just on the

robot, but on the broader experience of a single or

group of humans interacting with robots.

Researchers have long sought to deploy robots

alongside humans as human-like partners,

minimizing humans’ involvement in dangerous or

dull tasks. While robots have demonstrated some

promise as coordination partners, in practice they

contribute little to achieving humans’ goals, often

requiring more attention and maintenance and

eliciting more frustration than their contributions are

worth. Through these failures, it has become clear

that not only must robots’ technical abilities be

improved; so must their abilities to interact with

humans.

Humans prefer that all interaction partners that

exhibit social identity cues display role-specific,

socially-appropriate behavior (Nass & Brave, 2005;

Reeves & Nass, 1996). A robot must cater to this

human need to facilitate a successful interaction, but

designers of robots are rarely attuned to human

psychological processes.

Discounting human needs and expectations has

led HRI researchers to propose design goals for

robots that fail to fully consider the needs of

humans. Creating a “robot teammate” has become a

guiding goal of the HRI community, even though the

needs and expectations of humans intended to team

with robots have not been properly considered

(Groom & Nass, 2007). Because HRI has yet to

become a fully-established field, putting careful

thought into the goals of HRI now is essential for its

future success.

Cybernetics--the study of complex systems,

particularly those that feature self-regulation--places

a strong emphasis on the value of modelling

interactions and provides an established framework

for understanding and talking about systems,

something much needed in HRI. While HRI

researchers often model systems within a robot, little

attention has been paid to modelling the interaction

between a human and a robot.

Cybernetic models featuring a goal, comparator,

actuator, and sensor clearly delineate the relationship

between systems and their environments. The

system’s goal is to affect the environment in some

manner within some parameters. The system’s

comparator determines if the goal has been achieved

and transmits this information to the actuator, which

takes some action on the environment. A sensor then

detects some feature of the environment, and this

information is passed to the comparator. With

cybernetic models, systems continually influence

and are influenced by their environments and other

systems.

In this paper, I draw on cybernetics to represent

the models shaping the design of robots intended for

323

Groom V. (2008).

WHAT’S THE BEST ROLE FOR A ROBOT? - Cybernetic Models of Existing and Proposed Human-Robot Interaction Structures.

In Proceedings of the Fifth International Conference on Informatics in Control, Automation and Robotics - RA, pages 323-328

DOI: 10.5220/0001507103230328

 SciTePress

close human interaction. I critique these models and

offer revised models that include the human, the

robot, and the interaction between them. I also

model the conversational abilities required of

teammates, identifying those features that must be

developed in robots for humans to accept them. The

difficulty of meeting these requirements raises

questions as to whether the field of HRI is pursuing

optimal goals.

2 HRI DESIGN TODAY

Today’s robots are not yet capable of serving in

roles like teammate that require sophisticated social

capabilities. While designers are working on

creating robots capable of filling these roles, the

majority of existing robots fill less demanding roles.

These roles have lower requirements for autonomy,

intentional action, and socially-appropriate behavior,

and are similar to those roles filled by other

advanced technologies such as computers.

2.1 Robot Roles

One role that robots are often designed to fill is task

completer. In this role, robots complete a task

designated by a human. Many military robots, such

as bomb-detecting and bomb-defusing robots, are

modelled in this role. In some cases the robot’s

system may be non-cybernetic and in others it may

be cybernetic. With non-cybernetic task-completer

robots, the human sets the goal of the robot and the

robot affects the environment in a manner intended

to achieve the goal. In the case of a bomb-detecting

robot, the robot may run tests on a potential bomb

and send data back to distantly-located humans. The

process terminates at this point, as the system lacks a

sensor, comparator, or both. The process used by the

human to select the goal is not modelled, nor is there

any indication that the robot’s behavior affects the

humans’ goals.

A cybernetic task completer is generally more

robust and capable of more complex tasks than a

non-cybernetic task completer. The Roomba is a

popular example of a cybernetic task-completer

robot. As with non-cybernetic task completers, the

goal of a cybernetic task completer is set by a

human. Unlike non-cybernetic task completers, the

Roomba features sensors and a comparator that

partly comprise a cybernetic system, which enables

the Roomba to navigate obstacles. As with the non-

cybernetic task-completer, the human is considered

only peripherally in the design process. In the case

of the Roomba, the human is modelled as having

little interaction with the robot. The human provides

the robot power, maintains and cleans it, and

initiates its activities by turning it on.

The tool is another model commonly used for

the design of robots. A tool extends humans’

influence on the environment or grants humans

power over the environment that they do not

normally possess. Because a robot tool is much like

an extension of the self, attention is paid to the

human operating the robot: the goals and processes

of the humans are often considered in the design of

the robot. The robot is designed to help a human

complete a task or range of tasks. As a tool, the

robot is outside the human system, acting within the

environment on the environment.

Search and rescue robots often take the form

of a tool. One reason robot tools are useful in search

and rescue situations is because they enable people

to examine and influence areas that are inaccessible

or too dangerous for humans to access (Casper &

Murphy, 2003). The model of the robot tool differs

from models of the robot task-completer in that the

influence of the robot on the human is

acknowledged. However, the influence of the robot

is indirect, as the human senses only the

environment which contains the robot. Additionally,

the influence of the robot on the human is limited to

the humans’ selection of the best means to

implement a task strategy. The design of the robot as

tool does not model the robot as influencing the

human directly nor directly affecting the human’s

higher level goals, such as selecting a task strategy.

2.2 Social Feedback

The existing models of robots as task completors

and tools fail to delineate the powerful direct

influence of the robot on the human. Most designers

of robots, even those within the HRI community, fail

to fully recognize the social feedback that robots

generate. The behaviors of humans that interact with

bomb-detecting and defusing robots, Roomba, or

search and rescue robots indicate that they are

receiving information from the robot beyond that

which is intentionally designed.

An ethnographic study of the use of the Roomba

in family homes found that half of all families

studied developed social relationships with it

(Forlizzi & DiSalvo, 2006). These families named

the robot, spoke to it, described social relationships

between it and pets, and even arranged “play dates”

for multiple Roombas to clean together. In addition,

the Roomba affected the cleaning strategies of

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

324

household members, with males assuming a greater

involvement in house-cleaning. Anecdotal evidence

suggests that soldiers who interact closely with

robots in high stakes situations, like bomb diffusion

and search and rescue, form close emotional bonds

with robots, giving them names and grieving when

the robots sustains serious injuries.

In these cases, humans are responding to social

information generated by the robot. Computers as

Social Actors theory (CASA) was developed by

Nass (Reeves & Nass, 1996). CASA posits that that

even when technologies lack explicit social cues,

people respond to them as social entities. Research

performed under this paradigm has shown that even

computer experts are polite to computers (Nass,

Moon, & Carney, 1999), apply gender stereotypes to

computers (Lee, Nass, & Brave, 2000), and are

motivated by feelings of moral obligation toward

computers (Fogg & Nass, 1997). Even unintentional

cues of social identity elicit powerful attitudinal and

behavioral responses from humans.

Research indicates that some of the reasons that

people respond to computers socially is because

computers exhibit key human characteristics (Nass,

Steuer, Henriksen, & Dryer, 1994), including using

natural language (Turkle, 1984) and interacting in

real time (Rafaeli, 1990). Robots generally

demonstrate even more human characteristics than

robots. Some robots, such as Asimo or Robosapien,

feature a humanoid form. Many robots, such as

Nursebot or Roomba, feature some form of

locomotion, an indicator of agency. In addition,

robots often exhibit at least some autonomous action

and appear to humans to sense their environments,

make judgments, and act on their environments. The

very nature of robots make them appear even more

like social entities than most other existing

technologies and elicit an even more powerful social

response. But only when one of the primary design

goals is to foster a social relationship, as with

entertainment robots like Aibo or Robosapien, is the

social influence on the robot considered.

As indicated in Figure 1, the robot’s behavior has

a powerful influence on operators’ higher-level

goals. People have a high-level goal of recognizing

and evaluating social information transmitted by

others. Humans wish to respond to the behavior of

others in a socially-appropriate manner (Reeves &

Nass, 1996). While designers may have intended for

the robot to be an invisible tool, it is in fact sending

powerful cues indicating that it is a social entity. The

robots behavior may affect humans’ task strategies,

either through direct feedback or by influencing

humans’ higher level goals to act socially

appropriately.

Recognizing the influence of social feedback on

Figure 1: Model of a robot tool. Social feedback indicated with dotted arrows. Component boxes contain examples.

WHAT’S THE BEST ROLE FOR A ROBOT? - Cybernetic Models of Existing and Proposed Human-Robot Interaction

Structures

325

humans interacting with robots has important

implications for the design of robots. Designers are

more likely to consider which aspects of their design

are likely to generate a social response from humans.

Designers may be more inclined to create intentional

cues to foster a social relationship or to elicit the

desired social response. For example, it has been

demonstrated that humans apply gender stereotypes

to voices—even those that are obviously synthetic

(Nass & Brave, 2005). Awareness of this effect may

lead designers to choose robot voices not only based

on the clarity of the robot’s voice, but also based on

the desired social response.

Considering social feedback when designing

robots plays a key role in setting humans’

expectations of robots. The fewer and weaker the

cues of social identity, the lesser the likelihood is

that a robot will elicit a social response. Robotocist

Masahito Mori (1970) coined the term the “Uncanny

Valley” to describe humans’ responses of discomfort

when a robot’s visual or behavioral realism becomes

so great that humans’ expectations of human-like

behavior are set too high for the robot to meet. When

a robot is less realistic, humans have lower

expectations and are able to tolerate non-humanlike

behavior. As visual and behavioral indicators of

humanness increase and human-like behaviour

doesn’t, people become negative. Only when the

humanness of robots’ behaviors catch up to their

highly human-like appearance will robots emerge

from the valley of uncanniness. When designing

robots for interaction with humans, recognizing the

role of social information in setting user

expectations will enable designers to manage social

cues and set expectations that the robot is capable of

satisfying.

3 FUTURE OF HRI DESIGN

The roles that robots are successfully filling today,

such as task completer and tool, fail to take

advantage of robots’ full potential. Computers also

succeed in these roles, but robots have features that

computers do not. Robots have the potential to move

about their environments, sensing the world around

them, and either transmitting that information to

distantly-located humans or making decisions and

acting on the environment directly.

The ultimate goal for designers involved with

HRI is to create a robot capable of serving as a

member of a human team. Few researchers have

sought to define “team” or “team member” or

identify the requirements for creating a robot team

member. The robot team member has been generally

accepted as a lofty but worthy and attainable goal.

(For a summary and criticism of the “robot as

teammate” model, see Groom & Nass, 2007).

A well-established body of research is dedicated

to the study of teams. Successful teammates must

share a common goal (Cohen & Levesque, 1991),

share mental models (Bettenhausen, 1991),

subjugate individual needs for group needs (Klein,

Woods, Bradshaw, Hoffman, & Feltovich, 2004),

view interdependence as positive (Gully,

Incalcaterra, Joshi, & Beaubien, 2002), know and

fulfill their roles (Hackman, 1987), and trust each

other (Jones & George, 1998). If a human or robot

does not meet these requirements, they may never be

accepted into a team or may be rejected from the

team when problems arise (Jones & George, 1998).

One key requirement of teammates that underlies

all other requirements is the ability to engage in

conversation with other teammates. To be a

successful conversation partner, a robot teammate

must be able to both convey meaning in a way that

other teammates can understand and understand the

meaning intended in the communications of other

teammates. If a robot cannot do this, human

teammates can never be certain if the team shares a

common goal, which makes the human unable to

trust the robot in risky situations. Likewise, humans

would be uncertain if the robot was subjugating its

needs, viewing interdependence as positive and

knowing and fulfilling its role. Without

conversation, humans would feel certain that the

robot was incapable of sharing a mental model.

Figure 2 provides a model of conversation

between teammates that is derived from

cybernetician Gordon Pask’s (1975) Conversation

Theory (CT). One key element of this model of

conversation is the emphasis on both conversation

partners’ involvement in the communication.

Another related element is that both partners

construct the meaning of a message in their mind.

Meaning is not directly transmitted from one

conversation partner to the other, so each partner

must be capable of deriving meaning from a

message. A successful conversation requires that

each person not only ascribe their own meaning to

messages, but also infer the meaning of others and

compare the meaning of each partner to determine if

they are in agreement. While some robots are

capable of recognizing words or gestures and

responding appropriately, no robot has come close to

being able to fully engage in conversation.

Figure 2 highlights those features that must be

developed in robots for them to achieve the most

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

326

basic requirement of teammates: the ability to

engage in conversation. These requirements may be

broken down into three general categories: concepts,

knowledge, and systems. To communicate and

behave in a manner that allows humans to interpret

meaning, robots must demonstrate awareness of

basic concepts, including goals and motivation.

Robots lack humans’ complex hierarchy of goals.

Human teammates deployed in a high-stakes

situation like search and rescue maintain many goals

at once, including a goal to survive, a goal to protect

other teammates, and a goal to succeed at the task at

hand. Robots maintain a limited number of simple

goals that are always set at some point by a human.

One of the most important areas of knowledge

that robots lack is an understanding of common

human motivating factors and the relationship

between specific goals, motivations, and actions. In

order for robots to be useful in uncontrolled,

changing situations, they must posses a broad body

of knowledge. Robots’ lack of knowledge of

common goals, motivations, and actions also make

them difficult for humans to understand, eliciting

unintended negative responses from humans. While

a human’s motivation to avoid harm encourages

quick acts of self-protection, even a robot with a

goal of avoiding danger may, for example, enter and

remain in a dangerous environment, like a burning

void, and destroy itself. Human teammates are likely

to feel frustration, disappointment, and betrayal

when a robot acts in a manner that is self-destructive

and detrimental to the team.

In order for robots to be accepted by humans in

situations that rely on conversations and mutual

dependence, robots must exhibit behavior that

appears to humans to imply an underlying systems

much like the system used by humans to create and

use mental models. Human conversation partners

rely on their own mental models and their abilities to

create mental models of others’ mental models.

While it is possible that robots could successfully

fake mental models, they must rely on a system that

can serve a similar purpose to mental models and

appear to humans as a mental model. If robots are

unable to do this, humans will never feel certain they

share the same model of a goal.

Teams rely on a willingness of teammates to

subjugate their own personal goals for a team goal.

To do this, robots must demonstrate a sophisticated

goal hierarchy and effective communication skills.

Teams also depend on a high level of trust. Any

breakdowns in conversations may result in the

unraveling of the team. Maintaining trust requires

that teammates continually communicate about goals

and actions, and when trust is damaged, the

responsible party must acknowledge the violation

Figure 2: Model of a robot conversation partner. Robot must be capable to both communicate (left side) and interpret (right

side) meaning. Robot abilities that must be developed are indicated in each box. Dotted arrows indicate inferences. Dashe

arrows indicate checks of consistency.

WHAT’S THE BEST ROLE FOR A ROBOT? - Cybernetic Models of Existing and Proposed Human-Robot Interaction

Structures

327

and seek to repair the relationships (Jones & George,

1998). Even if a robot meets the basic requirements

of a conversation partner, its conversational abilities

will need to be further developed to meet the higher

expectations of teammates.

4 CONCLUSIONS

If designers wish to place robots in roles that have

previously been filled only by humans, they must

design robots that demonstrate the social behavior

and communication skills that humans expect of

people in these roles. To create robot teammates,

robots’ concepts of goals, motivations, actions, and

the relations between them must become further

developed and nuanced. Achieving this requires the

development of systems so complex that they

generate behaviors that enable humans to infer the

existence of shared mental models. Once researchers

recognize that creating a robot teammate takes far

more than improving a robot’s performance and

introducing it into a human team, the HRI

community can weigh the challenges of developing

a robot teammate to determine if creating a robot

teammate is indeed the best goal to guide the

direction of HRI research.

ACKNOWLEDGEMENTS

The author thanks Hugh Dubberly, Paul Pangaro,

and Clifford Nass for their feedback and support.

REFERENCES

Bettenhausen, K. L. (1991). Five years of groups research:

What we have learned and what needs to be addressed.

Journal of Management, 17(2), 345-381.

Casper, J., & Murphy, R. R. (2003). Human-robot

interactions during the robot-assisted urban search and

rescue response at the World Trade Center. Systems,

Man and Cybernetics, Part B, IEEE Transactions on,

33(3), 367-385.

Cohen, P. R., & Levesque, H. J. (1991). Teamwork. Nous,

25(4), 487-512.

Fogg, B. J., & Nass, C. (1997). Do users reciprocate to

computers? Paper presented at the ACM CHI.

Forlizzi, J., & DiSalvo, C. (2006). Service robots in the

domestic environment: a study of the roomba vacuum

in the home. ACM SIGCHI/SIGART Human-Robot

Interaction, 258-265.

Groom, V., & Nass, C. (2007). Can robots be teammates?

Benchmarks in human–robot teams. Psychological

Benchmarks of Human–Robot Interaction: Special

issue of Interaction Studies, 8(3), 483-500.

Gully, S. M., Incalcaterra, K. A., Joshi, A., & Beaubien, J.

M. (2002). A meta-analysis of team-efficacy, potency,

and performance: Interdependence and level of

analysis as moderators of observed relationships.

Journal of Applied Psychology, 87(5), 819-832.

Hackman, J. R. (1987). The design of work teams. In J. W.

Lorsch (Ed.), Handbook of organizational behavior.

Englewood Cliffs, NJ: Prentice Hall.

Jones, G. R., & George, J. M. (1998). The experience and

evolution of trust: implications for cooperation and

teamwork. The Academy of Management Review,

23(3), 531-546.

Klein, G., Woods, D. D., Bradshaw, J. M., Hoffman, R.

R., & Feltovich, P. J. (2004). Ten challenges for

making automation a "team player" in joint human-

agent activity. IEEE Intelligent Systems 19 (6 ), 91-95.

Lee, E.-J., Nass, C., & Brave, S. (2000). Can computer-

generated speech have gender? An experimental test

of gender stereotypes. Paper presented at the CHI

2000, The Hague, The Netherlands.

Mori, M. (1970). The uncanny valley. Energy, 7(4), 33-35.

Nass, C., & Brave, S. B. (2005). Wired for speech: How

voice activates and enhances the human-computer

relationship. Cambridge, MA: MIT Press.

Nass, C., Moon, Y., & Carney, P. (1999). Are people

polite to computers? Responses to computer-based

interviewing systems. Journal of Applied Social

Psychology, 29(5), 1093-1110.

Nass, C., Steuer, J., Henriksen, L., & Dryer, D. C. (1994).

Machines, social attributions, and ethopoeia:

Performance assessments of computers subsequent to

"self-" or "other-" evaluations. International Journal of

Human-Computer Studies, 40(3), 543-559.

Pask, G. (1975). An Approach to Machine Intelligence. In

N. Negroponte (Ed.), Soft Architecture Machines:

MIT Press.

Rafaeli, S. (1990). Interacting with media: Para-social

interaction and real interaction. In B. D. Rubin & L.

Lievrouw (Eds.), Mediation, Information, and

Communication: Information and Behavior (Vol. 3,

pp. 125-181). New Brunswick, NJ: Transaction.

Reeves, B., & Nass, C. (1996). The media equation: how

people treat computers, television, and new media like

real people and places: Cambridge University Press

New York, NY, USA.

Turkle, S. (1984). The second self: Computers and the

human spirit. New York: Simon & Schuster.

ICINCO 2008 - International Conference on Informatics in Control, Automation and Robotics

328