Dissertation Abstract

The core problem of human-robot interaction (HRI) is to understand and shape the encounters between humans and robots. This mission is supported by two main activities for researchers and practitioners: evaluate the capabilities of both humans and robots, or design the technology and strategy to produce desirable interactions. Both evaluation and design are influenced by how humans perceive their robot counterparts.

Conventional wisdom suggests a possible definition of robot include such characteristics as physicality, manipulation, and automation. Human perception of robots, however, extends any technical definition with presumptions of task quality, benefit, and societal status. One traditional understanding of robots depicts them as servants, doing the vital things we humans no longer want to do. We also see them as tools, caretakers, peers, or predators. Adopting any of these perspectives brings with it a value-driven consequence for human-robot relationships. Every choice is also a constraint.

This research examines the dynamics of human-robot interaction by revisiting the field of HRI and exploring the potential impact of adopting particular notions of robot. Interaction arises from a combination of five factors—autonomy, exchange of information, team dynamics, adaptation, and shaping tasks—that present themselves differently as context shifts. In terms of both the job being performed and its outcome, desirability is a useful tension for exploring how changes in context impacts the value of a robot.

Four design themes emerge that might lead to a different way to approach robotic design and implementation. Need precedes suggests greater care be given to grounding design decisions in human need specific to the context of implementation. Ambiguity engages argues that co-created meaning leads to more personal investment. Ceremony maintains stresses the importance of phatics in opening and maintaining communication channels. Relationships move reminds designers that social affordances are not static properties.

Robots clearly hold certain benefits for humanity, but there is a disconnection in how grounded are the decisions to design and implement robotics interventions. This line of inquiry hopes to address the root question—Why a robot?—as it speaks to the state of current and future robotics development.

Teaching with Robots

According to a post on Engadget, robots will become a presence in the classroom early this decade. South Korea’s continued national investment in the robotics economy will put teacher robots in 8,000 pre-schools and kindergartens by 2013. The “R-Learning” program will give robots the ability to tell stories and allow a telepresence for parents who want to check in on their kids during the day.

More Sensitive Skin

The MIT Media Lab has commissioned Peratech Limited to create a sensitive material that can be used to skin robotic forms. Peratech, a UK company partnering with Durham University, manufactures news types of electrically conductive material called Quantum Tunneling Composite (QTC) that is proportionally sensitive to pressure.

Walls That Move

Clemson University’s Intelligent Materials and Systems for Architecture program unveiled a robotic wall, Animated Work Environment . AWE is “a user-programmable robotic work environment that dynamically shapes and supports the working life of architects, designers and their new and more traditional collaborators.” There are six configurations of the wall.

Robot Movers

Crate and Barrel announced this month that is was joining Zappos, Staples, and Walgreens in adopting mobile-robotic solutions by Kiva Systems to help with order fulfillment in their warehouses. Rather than sending human workers out onto the floor to hunt down a bunch of divergent items, these robots find shelves filled with the products and queue them near the human’s workstation. There is a demonstration video available on the company’s site, but Be Kind To Robots also found a video of the robots dancing to the Nutcracker.

As director of the Media Lab’s Personal Robots Group, Breazeal is interested in developing technologies that engage people in human terms. Her doctoral work, Kismet, is an anthropomorphic robotic head capable of conveying emotions through facial expressions. Turkle is the director of the MIT Initiative on Technology and Self. For the past two decades, she has studied how people relate to the digital world. Recently, Turkle has focused on the study of robots, digital pets, and simulated creatures. Mixing these two perspectives brought up a few interesting themes about designing social robots.

Robots are Also for Reflection and Communication

Two of the robot types Breazeal highlighted while sharing her work stood out as new additions to my previous list: Research robots, and Robot-mediated communication. The former exists not to perform some function in a social setting, but rather to allow humans to study themselves. By creating a mechanical mirror of our behavior, we can test our understanding of what makes use human. With the Huggable—a remote-controlled teddy bear robot—Breazeal also suggests robots can be considered a new medium for communication with others. The bear is meant to relay situational information to a remote operator, who can then control the bear’s expression, gestures and focus to create a physical avatar for interpersonal relationships.

Living Rooms are the Final Frontier

Breazeal begins most discussions of robots noting that, because they are creatures that live in the physical world, they are subject to the laws of physics. Because people live in that world, too, robots are also governed by human minds that interpret and act. The navigation of physical spaces is only part of the problem. A robot placed amongst humans is defined by and must respond to human perception. “The final frontier is your living room,” says Breazeal. “That is the most complicated environment you can possibly imagine.”

Long-Term is Too Short

When Breazeal talked about a project involving robots and weight loss, she indicated that it was a long-term study. In robotics, however, “long-term” is anything lasting from 2-6 weeks. The novelty of having a robot available to interact with you every day apparently wears off before it embeds itself into a personal routine. The research claims the robotic form helped prolong engagement, but there are other factors that might be considered (relative effectiveness of the tested devices, conversational intelligence, etc.).

While it is certainly true that that technology needs to advance further before it will achieve the developer’s ideal, a designer might argue that anything that fails to be adopted into a person’s routine probably isn’t meeting a real need. This revelation reinforces my suspicion that a more human-centered path of inquiry could lead to a more engaging robots.

Human Connection Doesn’t Need Human Form

Referencing a 2004 ad for the Apple iMac G4 and an early Pixar short film, Breazeal suggested that people can share non-verbal communication even with everyday artifacts. This is evident in projects like RoCo—a robot computer designed to invite and encourage beneficial movement during use—and the flexible AWE robot wall. This could support the notion that we don’t need to go out of our way to reflect the human form in order to communicate in a human manner.

Any Difference Between Perception & Programming?

A big challenge for robotics research is imbuing robots with a means to perceive non-verbal cues. This is important if there is to be a real connective communication between human and robot. Breazeal’s group has explored this question with Leonardo, an expressive learning robot. Likening the early training of artificial intelligence to teaching a toddler, she described tests where Leo was able to track where a doll was placed and associate value judgments with different objects. While clearly impressive feats of engineering, it also has an air of verbal programming. In the process of raising three kids of my own, the demonstrations with Leo lacked a certain authenticity as it obediently responded to cues from authority.

Robot Relationships are Something New

Both Breazeal and Turkle implied that while there are many similarities with other forms of relationships, humans will connect with robots in different way than they do appliance, pets, or other humans. Turkle applies the term “technological promiscuity” to convey the perhaps inevitable willingness of people to form meaningful connections with embodied tech. It isn’t a question of mistaking a robot for a human; it is accepting the simulated human behavior as an acceptable substitute.

“Nurturers turns out to be the killer app in robotics,” said Turkle. The fantasy of human-robot relationships, though, is in the expectation that because you nurture them, they will reciprocate. This comes about because people focus on a robot’s performance, not it’s inner state. We can understand that a robot doesn’t love as we feel loved through its behavior.

Human Expectations Change

Turkle suggests that we are in the midst of an evolution in how we perceive communication itself, brought on by the lightweight tools designed to reflect status. She calls this state of readiness to form meaningful relationships with the inanimate the “robotic moment.”

Instant Messaging and microblogging are not intended to open dialogue about our complexity, but we can still feel enough of a sense of connection with people that the hypothetical Dunbar Number becomes more of a fallacy. The trade-off is, perhaps, a fundamental change in what is required to be in a relationship. As our expectations lower, our relationships can thrive with superficiality. It opens the door for human-robot connection, even if “we don’t always know when we’ve been close, when we’re alone.”

Authenticity is a Contextual Value

Turkle told a great story about a trip to a museum where the children saw an exhibit featuring live turtles from the Gallapagos Islands. Upon seeing the lack of mobility of one of the turtles, one kid suggested that the museum would have been better served with a robot. This prompted a research discussion about aliveness, with most of the kids in favor of a robot. A father listening to this conversation countered that being real is the whole point. This is part of Turkle’s robotic moment, where in certain contexts authenticity is not needed to create a valued experience.

On the flip side is a second anecdote about an elderly man in a care facility. He talks about how, though lack of skill or meanness, the humans in charge hurt him. “A robot would never pull you by the hair or tubes,” he admits. However, instead of embracing the perceived safety of the robot caretaker, the man prefers the hurtful humans. People have a story. Even the unpleasant person made him feel more alive, because their authenticity provided a connection to the human narrative.

According to Turkle, kids are referring to things as “sort of alive,” as if it were a new category of being.

Experience Changes What is Uncanny

All discussions of the aesthetics of robots eventually turn to the Uncanny Valley, a hypothesis that claims people start to revolt from robotic forms as they approach human. Despite not being supported by research, this thinking pervades robotics and even influences design (such as is the case with Asimo). Turkle reminded us that the origin of the word come from Freudian notions of familiar things seeming foreign, an uncomfortable but not necessarily frightening experience. In some cases, the technology advances to find a sweet spot in human reaction (look at what happened in CGI as it grew up from the Tin Toy baby to Shrek). Mostly, there might be some universal aversion when first encountering robots with near-human features.

What happens when the new experience becomes old? Is it really that a creepy looking fake baby always looks creepy, or could prolonged experience allow it’s behavioral performance to dominate? Appearances certainly help open the door for human relationships, but people who appear repulsive to some can still win other over with their minds and hearts. “It’s about our human vulnerability as much as the robot’s competence,” said Turkle.

The talks also pointed to a few additions to my reading list, including:

• Ahn, H., Teeters, A., Wang, A., Breazeal, C., and Picard, R. (2007). Stoop to conquer: Posture and affect interact to influence computer users’ persistence. In: Proceedings of the 2nd international Conference on Affective Computing and intelligent interaction (Lisbon, Portugal, September 12 – 14, 2007). A. C. Paiva, R. Prada, and R. W. Picard, Eds. Lecture Notes In Computer Science, vol. 4738, pp. 582-593.
• Heider, F., and Simmel, M. (1944). An experimental study of apparent behaviour. American Journal of Psychology, 13. (video)
• Levy, D. (2007). Love and sex with robots. HarperCollins.
• Riskind, J. (1984). They stoop to conquer: Guiding and self-regulatory functions of physical posture after success and failure. Journal of Personality and Social Psychology, 47, pp. 479-493.
• Turkle, S. (2008). Falling for science: Objects in mind. MIT Press.
• Vogeley, K., Bussfeld, P., Newen, A., Herrmann, S., Happe, F., Falkai, P., Maier, W., Shah, N.J., Fink, G.R., and Zilles, K. (2001). Mind reading: Neural mechanisms of theory of mind and self-perspective. Neuroimage, 14(1), pp. 170-181. (PDF)

A robot is a physical machine manipulated to automatically perform an undesirable work function that supports a desired human outcome.

This generated a constructive conversation with my advisor about the underlying value statements about both the technology and the humans who might do some of the “undesirable work.”

At the core of this simple definition is a complex assumption: People sometimes don’t want to do the work that produces benefit. At its worst, the interpretation is either laziness or superiority. In the best light, it is a call to appreciate those who do such jobs. This valuation of job function isn’t universal—one person’s chore is another’s passion—but it is still a value judgment on the task at hand and anyone in the situation to perform that task.

Undesired Work with Desired Outcomes

This idea of desirability as it relates to both work and outcome is interesting. We can describe all of the properties and functionality that a robot might entail, but the root question remains: Why a robot?

To explore this concept of desirability, there are a few key questions to ask and answer.

What are the boundaries of desirability?

In my two-dimensional world where benefit is measured in terms of affinity to work and outcome, there should be some point along each dimension where the perception moves from undesired to desired. Are there universally accepted thresholds of desirability? I want to understand the nature of these binary tipping points and how individual perspectives manifest themselves in the design domain. In other words, everyone’s assessment varies of whether or not work is attractive or outcomes are beneficial, and it is unclear to me how much that range overcomes the aggregate. This is a structural question that helps define the environment for robotic design.

How do humans cross boundaries of desirability?

Even before robots are introduced to the dynamics of interaction, humans have to navigate this space, moving from undesired to desired outcomes, or desired to undesired work. Where does playing World of Warcraft or mowing the lawn fit into this environment? How do people’s attitudes change over time and circumstance? Particularly with social robotics, where long-term interaction changes the relationship between humans and robots, understanding the dynamics of judgment might better inform design of the useful life cycle of such robots.

What is the relational impact of human-robot collaboration in each bounded quadrant?

Humans are our main concern here, mainly because the robots are new creatures shaped by humans. When you insert a robot or two into a situation common to one of the four quadrants of desirability, it will affect not just the human relationships with the machines, but also the human-human interaction that surrounds it. Giving a child a robot pet, for example, could have an adverse effect on how that child interacts with parents or friends. There are at least two dimensions at work here: affinity and self-interest. This question deals with the dynamics of agents in the system.

What is a robot?

Ultimately, I have to return to operationalization. My initial attempt at answering this question lacks input from the people who make robots and self-apply that label to their projects. The definition might also be compared to cultural artifacts that contrast what people think they believe and what they express. Although I am doubtful that a unified definition is achievable, the process of attempting to thoroughly answer this question should reveal a nice scaffolding for how to group robotics projects by their properties, function, design motivation and public perception.

Which design spaces would benefit from robotic collaboration?

Once the structure, environment and agent dynamics are understood, we still need to match these insights with real-world scenarios of need. It may prove easier to answer the question of which design spaces would not benefit from robots. From a designer’s perspective, this is a key question whose answers will result in robot imagination, or the optimal use of the medium to help address a need.

What I hope to address with this line of inquiry is that root question—Why a robot?—as it speaks to the state of current and future robotics development from a designer’s perspective. Even without articulating the detail, we can accept that robots exist and hold certain benefits. The disconnect appears to be in grounding the decision of when these benefits really come into play.

• Operator—directly influences or controls another’s behavior
• Mechanic—adjusting technical parts of a whole
• Supervisor—overseeing the coordination of subordinate workers
• Mentor—teaching or leading others
• Teammate (Peer)—cooperating with others
• Bystander (Peer)—witnessing behavior and experiencing indirect effects
• Information Consumer—benefits from information being broadcast without direct control of behavior or mechanics

Although all of these roles were meant to describe a human interacting with a robot, I have intentionally generalized them to include the reciprocal notion of a robot interacting with a human. It is also important to note that these roles do not necessarily define the robot’s purpose. There may be situations where the interaction shifts from operator to information consumer between the same human and robot.

The distinction between these two lists highlights the role of perspective in setting expectations for robot performance. People come into a relationship with a robot with some preconceived notion of its purpose, and that may or may not match what the designer had in mind. A human’s orientation toward a robot impacts how they interaction [1]. This should be true across the functional types of robots—Search and rescue, Assistive and educational robots, Entertainment, Military and Police, Space exploration, and Unmanned Air/Underwater Vehicles—although both human and robot are oriented toward each other in different ways.

In the field, roboticists rely on multi-disciplinary methods to figure this out. On the technical side, a mix of physical and simulated interaction can help inform development. Simulated users can provide insights [3], but robotics relies on the creation of real systems perhaps moreso than other computer interaction. There are some standard metrics used for evaluation [5], but they may not apply to many of the emergent dynamics that will arise out of long-term use. Figuring out how to test human-robot interaction and what to look for is still an open question.

Goodrich and Shultz suggest that from the problems motivating HRI research (such as: requirements on autonomy, information sharing, and evaluation) emerge several solution themes:

Interaction Complexity. Most interesting HRI applications deal with exchanges in dynamic environments. Interactions and the behaviors that result must be able to function within the complexity of this environment as well as adapt to changing dynamics.

Telepresence and Information Fusion. Remote control operation of robots is the oldest form of HRI, but humans continue to be disconnected from the sensor processing and communication conducted through the robot. With more attention paid to the interfaces used and information exchanged during remote control sessions, robots may become more of an extension of the human and therefore enjoy more success in operation.

Cognitive Modeling. When humans interact with each other, effective communication and collaboration is build from shared experiences and cultures. This common ground creates realistic expectations. It is the same for a robot. By developing models of human reasoning and behavior, roboticists hope their creations are capable of recognizing and adjusting to human communication.

Team Organizations and Dynamics. More and more, the 1:1 human-robot interaction is evolving into many:many relationships. Multiple robots and managed by multiple humans. These groups are subject to the same human dynamics studied by communication and sociology researchers. Roboticists can help shape these dynamics by establishing organizational structures, protocols, and support tools.

Interactive Learning. With complex interactions, it is impossible to anticipate all problems with scripted responses Robots and humans collaborate to incrementally improve perception, autonomy, and interaction.

These five solution themes all deal with the evolution of human-robot relationships from perception to understanding.

References:

1. Dautenhahn, K., Woods, S., Kaouri, C., Walters, M., Koay, K. L., and Werry, I. (2005). What is a robot companion—friend, assistant, or butler? In: IEEE/Robotics Society of Japan International Conference on Intelligent Robots and Systems.
2. Goodrich, M.A., and Schultz, A.C. (2007). Human-robot interaction: A Survey. Foundations and Trends in Human-Computer Interaction, 1(3), pp 203-275.
3. Ritter, F. E., Van-Rooy, D., and St. Amant, R. (2005). Using a simulated user to explore human robot interfaces. In: ACT-R Workshop.
4. Scholtz, J., Theofanos, M., and Antonishek, B. (2002). Theory and evaluation of human robot interactions. In: 36th International Conference on Systems Sciences, Hawaii: IEEE.
5. Steinfeld, A., Fong, T., Kaber, D., Lewis, M., Scholtz, J., Schultz, A., and Goodrich, M. (2006). Common metrics for human-robot interaction. In: First ACM International Conference on Human Robot Interaction, Salt Lake City, UT, USA: ACM.

Here, I condense their long paper into a framework applicable to my research interests in the design of social robotics. The result is a good perspective of the field of HRI and a short list of reading material on which to follow-up later.

Describing HRI

Although the foundation of the field is much older, Goodrich and Schultz describe a shift in robotics thinking during the 1990s. By 2001, a NSF- and DARPA-sponsored workshop on HRI was organized by Robin Murphy and Erica Rogers. It is considered by many to be the birth of the discipline. In 2006, an ACM conference for HRI was established, and a special SIG appearing for the first time the following year at CHI. (The current HRI conference is being held March 2-5 in Osaka.) The field is also influence by practical applications in the form of robotics competitions.

The authors distinguish between two general categories of HRI communication. Remote interaction is when the human and robot are separated spatially or temporally. Proximate interaction is when they are co-located in the same space and time. They further separate the field by whether the robot in question requires mobility, physical manipulation, or social interaction.

Three broad areas of robotics development include:

• Assisted search and rescue—humans remotely guide robots into dangerous areas to find victims of a disaster
• Assistive robots—technology giving physical, mental, or social support to benefit those in need
• Space exploration—robots functioning in extreme operating conditions, through both long-distance communication and co-located cooperation with astronauts

Challenges for assistive robots are largely issues of physicality (contact or environmental safety) and social (signals and emotive communication), with cultural influences due to long-term interaction.

I like theses distinction because it gives me a framework for my research to compare the effects of these requirements on human-robot relationships. My own work may well focus on just proximate communication of assistive robots.

A Problem of Interaction

The authors argue that the core HRI problem—to “understand and shape the interactions between one or more humans and one or more robots”—is supported of two main activities. Researchers and practitioners evaluate the capabilities of both humans and robots, or design the tech and strategy to produce desirable interactions. Interaction arises from a combination of five factors: autonomy, exchange of information, team dynamics, adaptation, and shaping tasks.

Autonomy

Tom Sheridan established a scale describing to what degree a robot can act on its own accord [21]. These levels of autonomy range from fully controlled by humans to computers ignoring humans entirely, with ever-increasing decision-making power given to the robot along the spectrum. As Miller and Parasuraman [11] noted, this continuum is most appropriate for individual tasks, not robots as an entity. In fact, metrics for autonomy—such as, neglect tolerance, or the amount of time a robot can be neglected—is only meaningful so far as it supports interaction.

These levels of autonomy present different issues for designers as one moves along the scale. When humans control most or all of a robot’s functions, the user needs involve reducing cognitive load. With robots making their own decisions, designers concentrate on natural and efficient communication behaviors. The choice of behavioral model also constrains design. Many researchers are leaning toward hybrid solutions that leverage sense-think-act models on top of behavior-based modules, where strategies emerge [12]. Reaction and reason are separate but work together. Other architectures include belief-desire-intention [24], joint intention theory [3], affect-based computing [1, 16], and temporal logics.

Exchange of Information

The authors identify two primary dimensions dictating the way information is exchanged between humans and robots: the communication medium, and its format. The media typically focus on three of the five human senses—seeing, hearing, and touching—which manifest in HRI as things like visual displays, gestures, audio alerts, and haptic interactions. There is a little goldmine of sources in this paper dealing just with the form of communication between humans and robots, as well.

The efficiency of information exchange between humans and robots is measured in different ways: time required to communicated to the robot [4]; the interaction workload [20], the change in situation awareness [5], and the common understanding shared between humans and robots [7, 9].

Team Dynamics

Because the primary concern for this field is interaction, the structure and function of human and robot team members is of great interest. Much of this work concentrates on the human perspective [19], specifically the upper bound on how many robots a single operator can manage (fan-out) [14, 15]. However, the issues of who has authority, which roles are played, and how conflict is resolved are not restricted to human agents. Similarly, there is a third entity—the interface software—that plays an important role in mediating the communication between team members.

Adaptation

Almost by definition, robots are learning machines. Adaptability, though, concerns both the robots and their human operators.

With humans, there is a tension between minimizing human training and demanding adaptive skills cultivated from knowledge of the robot’s domain. Current research is looking at the effects of prior experiences, such as instant messaging and online gaming, to help people form early mental models that positively influence human-robot interactions [17]. The “uncanny valley” theory is generally accepted among roboticists, but it is experimentally unproven [10]. Understanding how the presentation of a robot affects a human’s ability and willingness to adapt with it is a point of research interest.

Shaping Tasks

The designers themselves are not exempt from scrutiny. Designing a system is tantamount to making a hypothesis about how it will be used [23]. The choice of who to help and the core philosophies are embedded in any robot, making efforts to improve designers’ understanding of a work-practice domain vital to the success of a project.

Task-shaping is a term that examines the effect of technology on how and whether a task will be done. Some robots are specifically designed to invoke change in humans. The purpose of therapeutic and other social robots is to educate or train people through long-term interactions [8, 18, 22]. Learning takes place through both the design process [2, 13] and long-term interaction during operation.

This all leads the authors to a concept they call dynamic interaction. This emphasizes emergent interactions fed from previous interaction experiences by both humans and robots:

The scope of HRI research and design, therefore, includes all efforts at evaluating systems and interaction paradigms, designing autonomy algorithms in the context of HRI, designing interfaces and information exchange protocols, defining and switching roles, and influencing learning and training. This emphasis on dynamic interaction differs sharply from the historically static interactions of pure teleoperation and pure supervisory control.

A later post will revisit previous thoughts on roles and definition, as well and exploring possible methodological approaches to answering some early research questions.

References:

1. Breazeal, C. (2003). Emotion and social humanoid robots. International Journal of Human-Computer Studies, 59, pp. 119–155.
2. Brock, O., Fagy, A., Grupen, R., Platt, R., Rosenstein, M., and Sweeny, J. (2005). A framework for learning and control in intelligent humanoid robots. International Journal of Humanoid Robotics, 2(3).
3. Cohen, P. R., and Levesque, H. J. (1991). Teamwork. Nous, 25(4), pp. 487–512.
4. Crandall, J. W., Goodrich, M. A., Olsen Jr., D. R., and Nielsen, C. W. (2005). Validating human-robot interaction schemes in multi-tasking environments. IEEE Transactions on Systems, Man, and Cybernetics: Part A — Systems and Humans, 35(4), pp. 438–449.
5. Endsley, M. R., Bolt´e, B., and Jones, D. G. (2003). Designing for situation awareness: An approach to user-centered design. New York: Taylor and Francis.
6. Goodrich, M.A., and Schultz, A.C. (2007). Human-robot interaction: A Survey. Foundations and Trends in Human-Computer Interaction, 1(3), pp 203-275.
7. Johnston, J., Fiore, S. M., Paris, C., and Smith, C. A. P. (2006). Application of cognitive load theory to developing a measure of team decision efficiency. Military Psychology.
8. Kanda, T., Hirano, T., Eaton, D., and Ishiguro, H. (2004). Interactive robots as social partners and peer tutors for children: A field trial. Human-Computer Interaction, 19(1–2), pp. 61–84.
9. Klein, G., Feltovich, P. J., Bradshaw, J. M., and Woods, D. D. (2005). Common ground and coordination in joint activity. In: Organizational Simulation, (Rouse, W. B., and Boff, K. R., eds.), pp. 139–184, John Wiley and Sons.
10. MacDorman, K. F. (2006). Subjective ratings of robot video clips for human likeness, familiarity, and eeriness: An exploration of the uncanny valley. In: ICCS/CogSci, 2006 Long Symposium: Toward Social Mechanisms of Android Science.
11. Miller, C.A., and Parasuraman, R. (2003). Beyond levels of automation: An architecture for more flexible human automation and collaboration. In: Human Factors and Ergonomics Society 47th Annual Meeting.
12. Murphy, R. R. (2000). Introduction to AI Robotics. Cambridge, MA, USA: The MIT Press.
13. Nicolescu, M. N., and Mataric, M. J. (2001). Learning and interacting in human-robot domains. IEEE Transactions on Systems, Man, and Cybernetics, Part A, 31(5), pp. 419–430.
14. Olsen Jr., D. R., and Goodrich, M. A. (2003). Metrics for evaluating human-robot interactions. In: PERMIS, Gaithersburg, MD, USA.
15. Olsen Jr., D. R., and Wood, S. B. (2004). Fan-out: Measuring human control of multiple robots. In: Human Factors in Computing Systems, pp. 231–238, Vienna, Austria: ACM Press.
16. Picard, R. W. (1997). Affective computing. Cambridge, MA: MIT Press.
17. Reeves B., and Nass, C. (1996). The media equation: How people treat computers, television, and new media like real people and places. Cambridge University Press.
18. Robins, B., Dautenhahn, K., te Boerkhorst, R., and Billard, A. (2004). Effects of repeated exposure to a humanoid robot on children with autism. In: Designing a More Inclusive World, (Keates, S., Clarkson, J., Landong, P., and Robinson, P., eds.), pp. 225–236. London: Springer-Verlag.
19. Scholtz, J., Theofanos, M., and Antonishek, B. (2002). Theory and evaluation of human robot interactions. In: 36th International Conference on Systems Sciences, Hawaii: IEEE.
20. Sheridan, T. B. (2002). Humans and automation: System design and research issues. John Wiley and Sons.
21. Sheridan, T.B., and Verplank, W.L. (1987). Human and computer control for undersea teleoperators. MIT Man-Machine Systems Laboratory.
22. Werry, I., Dautenhahn, K., Ogden, B., and Harwin, W. (2001). Can social interaction skills be taught by a social agent? The role of a robotic mediator in autism therapy. In: Fourth International Conference on Cognitive Technology. Warwick, UK: Springer-Verlag.
23. Woods, D.D. (1998). Designs are hypotheses about how artifacts shape cognition and collaboration. Ergonomics, 41, pp. 168–173.
24. Wooldridge, M. (2000). Reasoning about rational agents. Cambridge, MA: MIT Press.

Early concept for a robot theme park in Incheon Korea

The first of its kind, “Robot Land” would be located in Incheon and feature entertainment facilities, exhibition halls, R&D centers, education buildings, and other support facilities. Most of the $562 million investment money (87%) comes from private investors, with national and local governments splitting the remainder of the bill. Incheon beat out Masan for the first park, but other sites could join Robot Land in the future. Construction at Incheon will be completed in three years, though open to the public by 2012.

Among the other tidbits coming from the article:

• The robot theme parks are expected to generate 2.8 trillion won ($2.4M) and create 18,000 jobs.
• The Korean government is emphasizing development of service robots that help clean homes and provide entertainment.
• Korea ranks fifth in terms of revenue in the world’s existing manufacturing robotics market (led by Japan and the United States).
• The service robot market is expected to be worth between $17-50 billion by the time the park opens.
• The robotics industry will resolve the issue of labor shortage caused by an aging Korean population.

The motivation and support of the Korean government is clearly tied to strengthening the national interest in robotics as a way to help economic interests. The Scooby Doo aficionado in me, though, recalls the last time a robot ran a theme park.

Thanks to @bekindtorobots for the tip.

Marshall began with Paul Dourish’s attempt to herd embodiment into one theory. In Where the Action Is, Dourish examines three aspects of embodiment [1] (as paraphrased by me):

• Intentionality—the motivating purpose for an action (presumed meaning)
• Coupling—the meaning that is attached to an action
• Intersubjectivity—the meaning that is negotiated between people around an action

I use action instead of object, even though in many instances there is some kind of social artifact that acts as the focal point for the interaction. Meaning, though, isn’t found in the object, but rather in past interactive experiences.

Marshall pointed to a few criticisms that peck at Dourish’s goal of unification. These include: an emphasis of social interaction over emotion; the removal of bodies from embodiment; and, absence of a transition from focus to background as part of the discussion. He also expanded the domain of embodiment to include felt experience, somatics, embodied facilitation, image schemas, and the extended mind. Jumping down that rabbit hole took me deep into Google Scholar, but for now I want to latch on to the most important concept of embodiment: when does the physical distinction go away?

Learning a technology involves something akin to Heidegger’s transition from present-at-hand to ready-to-hand. In other words, when we are first exposed to a technology, we have to focus on the object and think about its function. Over time, those thoughts move into the background as we simply use the tech as an extension of ourselves to do something. The object and its function is embedded in the choice to use it but otherwise is unobstrusive as we use it to act.

This is a key topic for Marshall’s work with tactile vision systems (TVSS), one of the E-Sense projects at Open University. Based on work done in the 1960s by Paul Bach-y-Rita, this experiment attempts to replace vision with vibrations across a matrix, converting electronic vision to touch-sensitivity. When the participants in these studies wear the special belts that push tactile information to a person’s stomach, the result is an ability to “see” objects without use of eyes. For this to happen, there has to be some leap where the participant no longer thinks of the technical contraption and just uses it to sense things in the world.

Running counter to that (perhaps) is the idea that one’s connection with a device is influenced by a personal investment in its construction. In a 2009 study by the CHIMe Lab at Stanford, researchers looked at the effect participation has on one’s affinity to a robot [2]. 56 people were asked to construct little robots that looked like humans or cars. The car robots were preferred overall, but there were also preferences for robots they assembled themselves.

Put these two ideas together, and we have an interesting tension relevant to social robotics. Is there a point when the robot becomes such an extension of the human mind that it becomes transparent in use? Or is effective social robotics built on the idea that transparency is detrimental? Is it a long-developing relationship and witness to growth that allows a person to accept a robot, and are such relationships every transparent?

References:

1. Dourish, P. (2001). Where the action is: The foundations of embodied interaction. Cambridge: MIT Press.
2. Groom, V., Takayama, L., Ochi, P., and Nass, C. (2009). I am my robot: The Impact of robot-building and robot form on operators. HRI ’09, pp 31-36.

As many people have told me (including my own mirror), I have a problem with focus. Lots of things are interesting, and part of my daily joy is contemplating how disparate ideas connect to each other. That is a limitless pursuit, though, as the matrix of “lots of things” by “lots of things” would take several lifetimes to conquer. Focus, for me, was just learning to be OK saying no. I tried on projects in sports networks, pedagogy, sustainability, education, communication theory, political wikis, tangible computing, social media services, and design theory. I believe there are common threads running through my perspective on all of those, but taking the long way home hasn’t gotten me very far.

Enter social robotics.

Why I Chose Robots

The upside to having no clear research focus is that there isn’t much committing me to a particular path. I chose a completely new dissertation topic for three important reasons:

The local landscape has changed

The School of Informatics & Computing has taken steps over the past year not only to provide additional physical and financial resources to the pursuit of robotics research, but there are also new human resources that make this area compelling. The program now has expertise in robot culture and technology to augment the current faculty and research taking place at the school.

My career plans have changed

Much of my fretting over dissertation topic was tied to a reinforced belief that what I did for my Ph.D. research would shape the kinds of jobs I could (or could not) get in the future … as an academic. While I intended to go become a professor somewhere, there was a lot of internalized pressure to make this decision of topic deliberate. Having seeing the strings at the puppet show, though, my interest in pursuing some tenure track position has diminished. If I want to head into the world of startups and community design, the specific topic becomes much less important to me than an opportunity to polish my process.

My expectations for the future have changed

Whether they are called robots or not, the trajectory for technology that embeds constructed intelligence and places interactions in the physical world makes the area of social robotics meaningful. Regardless of my work after graduation, I think I can be changed by my experience researching robots in the same way I already have being exposed to HCI design and complex systems perspectives.

Once I accepted those insights, it has become much easier to see the academic world in a more practical way.

Using Touchstones

I still let my interests wander, of course. The difference is that I feel I have just three lenses through which to view new information:

• How does it impact understanding of how humans and robots relate (Dissertation)
• How does it help explain and shape the dynamics of organizations? (SociaLens)
• How does it guide the design of online connection through expressive activity? (pixSmix)

All of these have common themes of complexity, connection, community, and embodiment. Pursuing any one of them has often resulted in new insights for the others. I don’t think about those connections, though, when I’m watching someone present or reading a paper. I summarize and uncover the insights of the new information, and then apply those insights to the above questions.

I don’t know whether this is actually effective, but I certainly feel more productive in advancing each of these projects.

Still a Spinning-Wheels Danger

That isn’t to say everything is getting processed in a timely fashion. Paul Marshall came in to town yesterday to talk about various applications of different notions of embodiment. This came a couple hours after another weekly session with my dissertation advisor. My mind still races with all the new stuff, and sometimes it is too much to turn into a concise insight quickly.

This post is a response to not being able to adequately process the new information from yesterday. I feel compelled to attempt to publish something every day (even though I am forgiving of myself when i don’t) but I also want it to be more than just a reflection of someone else’s work. I want to be able to answer the question of relevancy with my dissertation by making connections between different sources. It was taking too long, and the longer it took the more I felt like Grady Tripp from Wonderboys, doomed to keep writing without closure.

This post is mainly for the closure.

The goal of Asimo is to create a bipedal humanoid robot that can be helpful in a person’s home. After decades of researching and attempting to mimic how humans walk, Asimo has achieved toddler status and a top speed of almost 4 mph. The current results are enough to inspire improvements in Honda’s traditional areas of business (for example, anti-lock brakes have improved) and create a WOW factor for anyone who gets to see this line of robots in action.

Directed by Joe Berlinger (Metallica’s Some Kind of Monster) and produced @radical.media, Living With Robots includes interviews with the Honda employees responsible for evolving Asimo from a pair of mechanical legs to an amazing bipedal creature about the height of a child. The chats also include other scientists, philosophers, and writers. Some of the highlights include:

• “People and machines need a trusting relationship, just like the trust between people.”
  Jun Ashihara, Honda
• “In educating people about robots, one thing we notice … if it moves, you think it’s alive. And many people think robots are simply creepy.”
  Robin Murphy, CRASAR
• “I don’t think robots can coexist with people if you can’t overcome that issue [with being creepy].”
  Yasuhisa Arai, Honda
• “Some people obviously have some preconceptions about robots, maybe from what they’ve seen in movies or whatnot. I think over time, as people become familiar with technology in all areas of their life, they are going to come to understand what kind of benefits this technology can provide people.”
  David Iida, Honda
• “It’s kind of like a mirror we hold up to ourselves. Whatever robots turn out to be will largely be a function of the decisions we make.”
  Mark Rowlands, University of Miami

The pattern here is the relationship between humans and robots, and how appearance and experience play a role.

The opening interview in the film talks about how nearly all robots were developed as tools to extend human faculties. Robots like Asimo, however, may be service-oriented, but they are not pictured by their creators as “tools.” One of the early insights that arose from Japanese robotics came from scientist Masahiro Mori when he realized that there is a danger spot between human-like and human where the creep factor lives. Mori proposed the Uncanny Valley as a way of suggesting that humans feel more and more comfortable with a robot as it begins to resemble a person, but once it gets too close, there is a significant drop-off in comfort level.

Asimo’s design recognizes this by incorporating a space helmet-shaped design for a head—shielding any specific facial features—and downsizing the robot to be a less intimidating height. These choices were made to help Asimo to the top of the first peak of acceptance without dipping into the valley of creepiness.

What intrigues me most about Asimo is the mix of a technology-centrist process and the spiritual Eastern philosophy driving the project. Asimo clearly looks like a response to the question, “Can we make a machine walk like a man?” with all the assumptions about why that might be a good thing, but it also follows the Karakuri tradition comes from a tradition that imbues all things with spirit. In that sense, the robot is built as a technology but draws from its own kind of humanity.

Edit: I completely messed up my understanding of Karakuri tradition. I was probably looking for something more like Shinto.

• Laborer—robots are meant for jobs that are “dirty, dangerous, and dull.”
• Caretaker—robots are granted responsibility over humans and human success
• Peers—robots are the equal of humans, but with different attributes and weaknesses

I would add to that list:

• Tools—robots are soul-less machines, to be used when needed
• Predators—robots are ultimately at odds with humanity, waiting for their skills to improve before making a move

I’m sure there are more, but these five groups seem to cover most of the conventional perceptions of a robot. All of them are laden with value judgment.

Judging Robots

Adopting any of these perspectives about robots brings with it a value-driven consequence for human-robot relationships.

If a robot is a laborer—the original root is Slavic for “drudgery”—there is an implication of a lower class. The work it does is clearly deemed beneficial, or there wouldn’t be a need to do it in the first place. The robot’s relation to humanity, however, is serving as a new bottom rung on the societal ladder, regardless of whether the humans are appreciative.

If a robot is a caretaker, it is granted some degree of trust. The robot is expected to be present, available, and attentive to its human charge. It presumably does so at the expense of free will (It was made to care). This turns a robot into a servant or slave, and once again a class structure is implied. More so than the laborer robot, a caretaker is a relational creature that likely elicits a human bond.

If a robot is a tool, there is no relationship or soul behind it function. It waits until needed, is used as dictated, and then put back on a shelf. We value tools as property, extending ourselves to do things easier and with more precision.

If a robot is a predator, we fear it. There is no trust. We assume the nature of the beast is to attack us, at some point. We build defenses and prisons to keep it at bay, restricting rights and access to resources. The robot is allowed to exist only so long as the outcome of its work helps us.

We can look to how societies treat humans in these roles. We pay them less, demand less education, or lock them up.

Robots As Peers

Even when we see robots as the equals of humans, the value judgments are present, but they shift to us. A robot can be viewed as competition, which plays on our insecurities about our own skills and future. The presence of a robot peer is also theological, potentially elevating us to gods for creating them in the first place. In both cases, we may ultimately re-situate ourselves into positions where we again see the robot as a threat or subordinate, and so the peer relationship vanishes.

However, unlike with the other archetypes—where the adoption of a perspective brings with it certain societal constraints—viewing robots as peers offers more opportunity to leverage the dynamics of relationships. Our understanding of how robots work with humans can evolve.

In the Stanford study, 250 people responded to a public survey that matched robots with 812 job types, culled from the U.S. Department of Labor. The methodology attempted to tease out how different ways of asking the questions might bias the results.

There was a clear distinction between whether people thought of the jobs as collaborative or competitive. People were more positive about robots working with people instead of replacing them. This might be more of a reflection of societal sensitivities to job loss than the robot form, but it still provides support for an acceptance of robots as collaborators, if not peers. The methodology also revealed that people may not understand the capabilities of robots enough to distinguish between what robots could do and what they should do.

The study concluded that robots are best matched with jobs that require memorization, keen senses, and service. Humans, on the other hand, are built for art, evaluation, judgment, and diplomacy.

What this suggests is that there are characteristics of a robot that distinguish it from a human and make it more ideally suited to certain tasks. Similarly, humans offer the robot an external cognition most people don’t believe are part of its skill set. When trying to define a robot, our understanding should incorporate an opportunity to move beyond early archetypes where robots help humans and into a space where we help each other.

Reference:

1. Takayama, L., Ju, W., and Nass, C. (2008). Beyond dirty, dangerous and dull: What everyday people think robots should do. HRI 2008, pp. 25-32.

A tip from Science Friday leads to robot actors

Science Friday tweeted this little gem today about an NPR interview with Texas A&M graduate student Kevin Pratt. He explains how his search-and-rescue flying robots became part of a production of Shakespeare’s A Midsummer Night’s Dream.

This collaboration between the theater department and the Center for Robot-Assisted Search and Rescue (CRASAR) was helpful to further robotic research as well as provide a post-modern interpretation of a classic play. Disaster response robots mixed with the human actors on stage for eight shows. The larger AirRobot (flown by Pratt) kept company with with Oberon, the King of Fairies, while several smaller robots circled the air near Titania’s crew.

Texas A&M mixed robot and human actors for Shakespeare

The stage doubled as a lab, where researchers from the robotics team observed how the human actors and audience members responded to the robots. This is important data that will inform the use of robots in disaster response

“They’re going to be interacting with people who have just gone through a disaster,” said Pratt, “and so you need to be able to communicate with them effectively and quickly.”

CRASAR is led by Director Robin Murphy and researchers the use of new technology for crisis response. In particular, the center is interested in unmanned systems used for humanitarian purposes around the world. Their robots were deployed during the aftermath of 9/11. CRASAR’s robots are meant to serve as assistants to the human rescue teams rushing to the disaster scenes.

The purpose of this work, inspired by Darwinian selection theories, is to illustrate how natural selection can lead to complex properties—in particular, adaptive behavior:

Just a few hundred generations of selection are sufficient to allow robots to evolve collision-free movement, homing, sophisticated predator versus prey strategies, coadaptation of brains and bodies, cooperation, and even altruism. In all cases this occurred via selection in robots controlled by a simple neural network, which mutated randomly.

One key insight from this work is that it highlights the perhaps flawed strategies in past attempts to make robots smarter. The authors point out that other techniques rely on vast prior knowledge of a particular environment in order to be successful. Failures occur when the robot shifts environments to find out a dependent resource is no longer there, or that it is incapable of detecting the information in the new setting that would be most helpful.

In a previous paper (Floreano, D., Husbands, P., and Nolfi, S. (2008). Evolutionary robotics. In: Siciliano, B., Khatib, O., editors. Springer handbook of robotics. Berlin: Springer Verlag. pp. 1423–1451.), Dario Floreano listed some of the ways engineers have attempted to address the problem of real-world failures with their robots:

• Measure the fitness of evolving individuals in several environments, varying dimensions in each
• Incorporate noise into the simulation models that moves the virtual environments away from the ideal
• Coevolve the robot and the key parameters of the simulation model, periodically testing with real robots to improve fitness estimations
• Add ontogenetic plasticity to the evolving individuals, so they can adapt to environmental changes over the course of their lifetime

I find this intriguing from a social robotics scale because of the dynamics of social interaction is often unpredictable. There may be some lessons to learn from this approach, even if I don’t end up building anything.

Reference:

• Floreano, D., and Keller L. (2010). Evolution of adaptive behaviour in robots by means of Darwinian selection. PLoS Biol, 8(1)

There is a second notion of embodiment, however, that may be more interesting to consider for the domain of social robotics. It involves sharing responsibility for the value and function of a robot between the device and its human partners … keeping “embodied robot” from being redundant.

Actions are an Outcome

Human-computer interaction (HCI) has long been interested in the objects of interaction. In work that influenced Don Norman’s foundational contributions to the interaction design, J.J. Gibson described the relationship between the organism and the environment [7]. The observer constructs meaning from the individual properties and specific actions evident in the situation. Gibson stops short, however, at suggesting the same is true for the environment:

“The concept of affordance is derived from these concepts of valence, invitation, and demand but with a crucial difference. The affordance of something does not change as the need of the observer changes. The observer may or may not perceive or attend to the affordance, according to his needs, but the affordance, being invariant, is always there to be perceived.” [7, p138]

Conceptually, actions are often considered tangible. It is as if we can take that action, put it in a box representing the conditions of the environment, and have it act a certain way. If we put that action in a different box, or put a different action in the same box, new things happen. Either way, the action and box, or context, are separable.

Anthropologist Lucy Suchman describes action as an outcome rather than an intention. Actions arise from a context in which actors and resources interact. Any plan can only reflect what has happened and anticipate what is to come [10]. Outside of the moment, action doesn’t exist.

On the surface, Gibson’s perspective appears to match well with the attitude of many roboticists. Different humans may interact with robots in distinct ways, but the success of a human-robot interaction is dictated by the quality of design in the device. People are involved only to benefit from the robot’s skills. If human-robot interaction (HRI) adopts Suchman’s insights, robots could be designed to benefit from those interactions, too. The robot could be seen as both agent and environment.

Embodied Interaction

These conceptualizations of action echo two notions of embodiment. The first emphasizes physicality, an idea or action manifesting in the world. This definition arises from a reductionist perspective that wishes to objectify things in order to turn them into rational, wholly understandable entities. The second notion comes from a phenomenological perspective, arguing that categorization of the world is imposed rather than pre-existing. In this sense, embodiment is a phenomenon that occurs in real time, in real space [4,5].

Embodied agency—which typically takes the form of things like dialoguing avatars and robotic media—embraces the first interpretation. Embodied conversational agents (ECA) distinguish themselves from traditional forms of dialogue systems through the nonverbal behaviors they are able to convey [3]. The agent is not necessarily a co-creator of experience.

Paul Dourish represents the seminal thinking about the second interpretation. In Where the Action Is, Dourish suggests that not only are tangible and social computing the second-wave evolution of computer interaction, they are also two halves of the same coin:

“[Embodied interaction] is an approach to the design and analysis of interaction that takes embodiment to be central to, even constitutive of, the whole phenomenon” [5, p102]

In other words, it is not enough that something lives in the world. Embodiment in this sense demands that we recognize that its existence shapes the world around it, which in turns affects how it is manifested.

Facilitate the Co-creation of Meaning

It is a radical idea to suggest that not only should you holistically examine a system, you must do so. Embodied interaction changes the question from how can people do their work, to how can people create their own meanings and uses [4].

In HCI, this is evident in a transition to “third wave” methodologies [1]: exploration is now laced with ambiguity in the form of culturally embedded computing [9] and cultural probes [6]. These inquiries attempt to inform design decisions without absolutes. (Old habits die hard, however, as HCI often tries to transform these approaches back into concrete findings [2].) The design of systems dependent on community participation faces challenges both in development and evaluation [8].

The same is likely applicable for HRI, where design involves a more constrained form of IT. Roboticists could benefit from some of the techniques researchers use in HCI, but there are sure to be some means of inquiry that will be specific to the domain. An embodied robot must examine its environment and other agents in such a way that allows it participate in the creation of meaning, rather than just reflect a fixed understanding or reality.

References

1. Bødker, S. (2006). When second wave HCI meets third wave challenges. In: Proc. of NordCHI’06, pp. 1-8.
2. Boehner, K., Vertesi, J., Sengers, P., and Dourish, P. (2007). How HCI interprets the probes. In: Proceedings of CHI’07, pp. 1077-1086.
3. Cassell, J. (2000). Nudge nudge wink wink: Elements of face-to-face conversation for embodied conversational agents. Embodied Conversational Agents. MIT Press., pp. 2-28.
4. Dourish, P. (2004). What we talk about when we talk about context. Persuasive & Ubiquitous Computing, 8, pp. 19-30.
5. Dourish, P. (2001). Where the action is: The foundations of embodied interaction. Cambridge: MIT Press.
6. Gaver, G., Dunne, T., and Pacenti, E. (1999). Cultural Probes. Interactions, pp. 21-29.
7. Gibson, J.J. (1979). The ecological approach to visual perception. Boston: Houghton-Mifflin.
8. Ryan, W., Hazlewood, W.R., & Makice, K. (2008). Twitterspace: A co-developed display using Twitter to enhance community awareness. PDC ’08, pp. 230-234.
9. Sengers, P., Kaye, J., Boehner, K., Fairbank, J., Gay, G., Medynskiy, Y., and Wyche, S. (2004). Culturally embedded computing. IEEE Pervasive Computing, pp. 14-21.
10. Suchman, L. (1987). Plans and situated actions: The problem of human-machine communications. New York: Cambridge University Press, pp. 49-67.

As introduced to HCI in 2003 by Gaver et al, ambiguity falls into three broad classes: information (in the artifact), context (surrounding the artifact), and relationship (in the participant’s experiences) [5]. The authors suggested several actions the designer might consider when enhancing the ambiguous design of an experience, including:

• Over-interpret data to encourage speculation. (Information)
• Cast doubt on sources to provoke independent assessment. (Information)
• Implicate incompatible contexts to disrupt preconceptions. (Context)
• Offer unaccustomed roles to encourage imagination. (Relationship)

Ambiguity can be considered an asset to the design of interactive systems.

With pixSmix—where users are asked to make collages out of random photos—people actively construct a meaningful experience around the technology [10]. Individuals build social artifacts from this creative material based on the sense they make of the images they see. In turn, that artifact will be assigned different meaning by others who see it. This is a key characteristic of ambiguous design.

Examples of ambiguous design include:

• The Home Health Horoscope—a sensor-based system that collects information about activity in a home and turns that data into over-interpreted statements, modeled after horoscopes [6].
• AuralScapes—a project to bring arrhythmic sounds and overhead images into an enclosed internal room, changing the ambiance of a physical space [9].
• A field test conducted with networked cameras, where participants attempt to capture images revealing the context of their own lives, being appropriated into a broadcast tool for storytelling [8].
• Pangmangi—a flat-panel display installed on office doors to create awareness of the occupant’s availability, fostering some questions of trust [7].

There can also be ambiguity in purpose, to provoke behavior counter to the design. Digital deviance addresses the needs of the darker aspects of humanity through application design [4]. Sinister prompts can engender positive outcomes: the technology assumes a negative role, and humans in the group counter with noble behavior.

Ambiguity is not a fixed target, however. Over time, interpretations become convention, lessening the ambiguity and thus need to interpret [3]. Different meanings arise out of different contexts, and therefore, ambiguity must be designed as an interpretive space [1], rather than a discrete object that remains eternally ambiguous.

Ambiguity in Robots

If ambiguity is the sweet spot in between precision and art, much of robotics research focuses on the precision end of the spectrum. Attempts to make robots social, for example, have dealt primarily with accurately reproducing human movement and expression. Recently, ambiguity has been considered from the robot’s perspective, in attempting to improve it’s understanding of commands by appropriating a human context [11]. Researchers have also begun to leverage ambiguity in the way robotic faces are designed [2].

With ambiguity as a material, designers can theoretically increase connection between human and robot by creating a situation where the former has to interpret the latter. The human gets just enough from the robot’s behavior and appearance to be willing to fill in the rest with her own experience, imbuing the robot with more meaning than it was programmed to deliver. If the robot, too, is given the learning skills to cultivate meaning from interactions with those in its environment, each human-robot relationship can potentially grow into a unique experience.

References:

1. Aoki, P., & Woodruff, A. (2005). Making space for stories: Ambiguity in the design of personal comm. systems. In: Proc. of CHI ’05, pp. 181-190.
2. Blow, M., Dautenhahn, K., Appleby, A., Nehaniv, C. L., and Lee, D. (2006.) The art of designing robot faces: Dimensions for human-robot interaction. In Proceedings of the 1st ACM SIGCHI/SIGART Conference on Human-Robot interaction (HRI 2006), pp. 331-332.
3. Boehner, K., & Hancock, J.T. (2006). Advancing ambiguity. In: Proceedings of CHI ’06, pp. 103-106
4. Foucault, B., Mentis, H.M., Sengers, P., & Welles, D. (2007). Provoking sociability. In: Proc. of CHI ’07, pp. 1557-1560.
5. Gaver, W.W., Beaver, J., & Benford, S. (2003). Ambiguity as a resource for design. In: Proc. of CHI ’03, pp 233-240.
6. Gaver, W., Sengers, P., Kerridge, T., Kaye, J., & Bowers, J. (2007). Enhancing ubiquitous computing with user interpretation: Field testing the home health horoscope. In: Proc. of CHI ’07, pp 537-540.
7. Huh, J., Ackerman, M., & Douglas, R. (2007). The use of aesthetics in HCI systems. In: Proc. of CHI ’07, pp. 2441-2444.
8. Makela, A., Giller, V., Tscheligi, M., & Sefelin, R. (2000). Joking, storytelling, artsharing, expressing affection: A field trial of how children and their social network communicate with digital images in leisure time. In: Proc. of CHI ’00, pp 548-555.
9. Mathew, A.P., & Taylor, J. (2008). AuralScapes: Engaging ludic ambiguity in the design of a spatial system. In: Proc. of CHI ’08, pp 2533-2540.
10. Sengers, P., Boehner, K., Gay, G., Kay, J.J., Mateas, M., Gaver, B., & Hook, K. (2004). Experience as Interpretation. In: Proc. of CHI ‘04 [workshop].
11. Trafton, J., Schultz, A., Bugajska, M., and Mintz, F. (2006). Perspective-taking with Robots: Experiments and models. IEEE International Workshop on Robot and Human Interactive Communication, 2005. ROMAN 2005, pp. 580-584.

Have you ever wondered what a robot really is? Over coffee the other day we were trying to find a sleek and simple one size fits all definition for all the robots we’ve covered on the show from molecular robots to smart houses, humanoids or flying crawling and jumping robots. However, for every definition we came up with there was a counter example that either didn’t fit the definition or did although it wasn’t really what we think of as a robot.

The show will dedicate some time in every episode to modify a living definition of robot, subjecting new entries to the “counter-example” test until satisfied with the result.

Since I have to operationalize the term for my dissertation, this part of the bi-weekly podcast will be of great interest.

In search of a definition

The task isn’t an easy one. Most robotics researchers seem to take an “I know one when I see one” approach to a definition. The origin of the word (robota) firmly ties it to hard labor and a sub-class of humanity. This is reflected in the cultural perception of robots as servants, doing the vital things we humans no longer want to do.

Several official definitions (ISO, RIA, Encyclopaedia Britannica, Merriam-Webster) agree that a robot is at least a manipulated machine that performs a work function. Excepting the low end of RIA’s spectrum of control, automation could also be part of the working definition.

Academics may not always articulate it, but a common definition may be implied whenever they self-associate a pet project with the term. A thorough review of the research literature is in order, but in the short-term, the low-hanging fruit is found in the Wikipedia’s article on the topic. It suggests the following considerations for refining the working criteria:

• electric powered
• reprogrammable
• automatically guided (able to do tasks on its own)
• aligning with cultural expectations (appearance and movement)
• an intent or agency of its own
• physical embodiment
• processes with environment (interacts with physical objects)
• human-inspired
• subordinate while superior

From the above list, the most compelling factors are the requirements to be situated in physical space and tied to humanity. A robot is not something virtual, operating only within a computer processor. It is material and interacts with material things, including people. Robots are also implicitly tied to their human counterparts, whether that be at the smallest scale (molecular) or the largest (societal). There is a philosophical question of superiority in play, but any good designer will argue that one can’t dictate use. The hierarchy of the human-robot relationship is one in flux.

That suggests my first attempt at a universal definition:

A robot is a physical machine manipulated to automatically perform an undesirable work function that supports a desired human outcome.

If you think you have a good answer to the question, “What is a robot?”, email a short definition to Robots Podcast with your phone number so that they can contact you on the air.

Panasonic’s Robot Bed

Without knowing much detail about how this contraption came into being, I can use it as a thought experiment to explore some of my initial thinking about social robotics.

Who needs this?

From a human-centered design perspective, this is perhaps the key question that should drive development. One of the perceived problems with robotics research is the focus on the technical challenges, often in the form of a “wouldn’t it be cool” view of the work. That isn’t to say that the prototypes and products scattered through the domain of robotics are toys or devoid of real value. However, are the decisions that go into making a design—including the choice of a robotic form—grounded in real need?

People stuck in beds for whatever reason likely have a desire not to be there all the time. In order to leave the bed, they need assistance transferring from the bed to a chair. Without assistance, they remain bedridden. Panasonic’s solution certainly will address the need to move, but does it also reinforce other problems this population deals with, such as social isolation? Without a robot bed, other people can facilitate not only motion but also relational connection, which is known to be tied to health.

Is this a robot?

Presumably, yes. The video implies a self-guided docking system, like a Roomba finding its base, that doesn’t require manual precision from a human operator. There are also sensors and collision avoidance intelligence that would also give it some of the technical properties commonly associated with robots.

Is this sufficient? The human can override the robot’s control. Does that act turn the robot into a mere vehicle? It is unclear if the Panasonic bed can interact with anything beyond it’s physical environment. Does it need to be aware of and respond to its operator to be considered a robot? Answering these kinds of questions for these and other systems is vital to the success of my research. In order to know whether a robot is an appropriate design choice, we must first understand what makes it robotic.

Is this social?

There is not direct dialoguing occurring between bed and operator. The robot in this case isn’t imbued with a personality that might allow someone to relate to it in some anthropomorphic manner. However, its operation has the ability to both enhance or deteriorate existing social connections between the operator and other humans in her social circle. It also has a potential effect on the issue of elder care, which has social and political ramifications. If a robot affects the social, is it not itself social?

As one of my advisors pointed out: perhaps a better question than “Is it social?” would be “What isn’t social?” We live in an embodied world, in the sense that meaning is co-created by participants and environment. There is no absolute or ideal truth. If all meaning is constructed, would those participating in that interaction be implicitly social?

Obviously, I want to avoid operationalizing my key terms to be too inclusive, but there is value is examining “social” in a way that might get beyond its traditional understanding of being tied to groups and conversation.

The main contributors to this resource include HCII professors Sara Kiesler and Jodi Forlizzi of CMU, Stanford professor Pamela Hinds, and recent CMU graduates Bilge Mutlu and Carl DiSalvo. There are a couple dozen academic papers listed on this site, feeding four major areas of interest:

How robots can help people

Research here includes studying the social aspects of movement (Companion), the positive and negative aspects of receiving help from robots, and how intelligent objects use sensors to infer when to help (Sensechair). Also under this heading is a more general study of the elder population and their use of products, largely driven by Forlizzi.

Fundamental mental models and communication

Here, the research is split between something quite technical (mimicking human gaze with robots) and something sociological (understanding mental models of robots). The implementation of robotic gaze is less interesting to me than the work done trying to figure out how people use gaze in communication. On the other hand, linking how people’s expectations based on a robot’s appearance influences how they interact is very relevant to my dissertation interests.

How service robots function in physical and social context

This gets a bit at the insight I tried to express in yesterday’s post, that whether or not something is social is based on interaction not just designer intent. The Roomba studies by Forlizzi try to get at this by examining how people imbued a service-focused robot with social properties.

Deployment of robots over the long term

Time is such a critical and often neglected aspect of both system adoption and quality of communication. The CMU crew is testing this through Snackbot, a a mobile, semi-autonomous robot whose primary job is to serve snacks. Snackbot is also a research platform for projects in robotics, design, and behavioral science.

The above areas of exploration have generated almost two dozen different academic papers with strong relevance to what I’m considering for my dissertation. (Thanks, Chloe!)

References:

• DiSalvo, C., Forlizzi, J., Zimmerman, J., Mutlu, B., and Hurst, A. (2005). The SenseChair: The lounge chair as an intelligent assistive device for elders. In: Proceedings of the Conference on Designing for User Experiences (DUX05) November, 2005, San Francisco, CA, USA.
• DiSalvo, C.F., Gemperle, F., Forlizzi, J., and Kiesler, S. (2002). All robots are not created equal: the design and perception of humanoid robot heads. In: Proceedings of DIS02: Designing Interactive Systems: Processes, Practices, Methods, &Techniques 2002. pp. 321-326.
• Forlizzi, J. (2007). How robotic products become social products: an ethnographic study of cleaning in the home. In: Proceedings of the ACM/IEEE International Conference on Human-Robot Interaction 2007. pp. 129-136.
• Forlizzi, J. (2005). Robotic Products to Assist the Aging Population. Interactions, 12(2), 16-18.
• Forlizzi, J., and DiSalvo, C. (2006). Service robots in the domestic environment: a study of the roomba vacuum in the home. In: Proceedings of the 1st ACM SIGCHI/SIGART Conference on Human-Robot Interaction 2006. pp. 258-265.
• Forlizzi, J., DiSalvo, C., and Gemperle, F. (2004). Assistive Robotics and an Ecology of Elders Living Independently in Their Homes. In Human-Computer Interaction, 19 (1) pp. 25-59
• Gemperle, F., DiSalvo, C., Forlizzi, J., and Yonkers, W. (2003). The Hug: a new form for communication. In: Proceedings of DUX03: Designing for User Experiences 2003. pp. 1-4.
• Gockley, R., Bruce, A., Forlizzi, J., Michalowski, M., Mundell, A., Rosenthal, S., Seller, B., Simmons, R., Snipes, K., Schultz, A., and Wang, J. (2005). Designing Robots for Long-Term Social Interaction. Proceedings of ICRA05. New York, NY: John Wiley and Sons, 2199-2204.
• Gockley, R., Forlizzi, J., and Simmons, R. (2007). Natural person-following behavior for social robots. In: Proceedings of the ACM/IEEE International Conference on Human-Robot Interaction 2007. pp. 17-24.
• Hirsch, T., Forlizzi, J., Hyder, E., Goetz, J., Stroback, J., and Kurtz, C. (2000). The ELDeR Project: Social, Emotional, and Environmental Factors in the Design of Eldercare Technologies. In: Proceedings of the 2000 ACM Conference on Universal Usability 2000. pp. 72-79.
• Kiesler, S., Powers, A., Fussell, S. R., and Torrey, C. (2008). Anthropomorphic interactions with a robot and a robot-like agent. Social Cognition, 26(2), 169-181.
• Kiesler, S. &Sproull, L. (1997). Social responses to “social”computers. In B. Friedman, Human values and the design of technology, CLSI, Publications.
• Kiesler, T., &Kiesler, S. (2004). My pet rock and me: An experimental exploration of the self extension concept. Advances in Consumer Research, Vol. XXXII (32).
• Lee, M.K., Forlizzi, J., Rybski, P.E., Crabbe, F., Chung, W., Finkle, J., Glaser, E., and Kiesler, S. (2009). The snackbot: documenting the design of a robot for long-term human-robot interaction. Proceedings of the 4th ACM/IEEE International Conference on Human Robot Interaction 2009. pp. 7-14.
• Lee, M. K., Kiesler, S., Forlizzi, S. (2010). Receptionist or Information Kiosk: How Do People Talk With a Robot? Proceedings of the Conference on Computer-Supported Cooperative Work. NY: ACM Press.
• Lee, S.-L., &Kiesler, S. (2005). Human mental models of humanoid robots, International Conference on Robotics and Automation, ICRA 2005, Barcelona, SP. April.
• Mutlu, B., and Forlizzi, J. (2008). Robots in Organizations: The Role of Workflow, Social, and Environmental Factors in Human-Robot Interaction. In: Proceedings in the Third International Conference on Human-Robot Interaction March 12-15, 2008, Amsterdam, The Netherlands.
• Mutlu, B., Hodgins, J., and Forlizzi, J. (2006). A Storytelling Robot: Modeling and Evaluation of Human-like Gaze Behavior. In: Proceedings 2006 IEEE-RAS International Conference on Humanoid Robots December 2006, Genova, Italy.
• Mutlu, B., Osman, S., Forlizzi, J., Hodgins, J. and Kiesler, S. (2006). Perceptions of ASIMO: An exploration on co-operation and competition with humans and humanoid robots. In: Extended Abstracts of the Human-Robot Interaction Conference (HRI06) March, 2006, Salt Lake City, UT, USA.
• Powers, A., Kiesler, S., Fussell, S. R., and Torrey, C. (2007). Comparing a computer agent with a humanoid robot. In Proceedings of the International Conference on Human-Robot Interaction. HRI ’07. ACM, New York, NY, 145-152.
• Torrey, C. (2008). Social and psychological reactions to receiving help from a robot. In CHI ’08 Extended Abstracts on Human Factors in Computing Systems. 2653-2656.
• Torrey, C., Fussell, S.R., and Kiesler, S. (2008). Trying to Help: Social Challenges for Smart Robots. In Proceedings of the Workshop on Robotic Helpers: User Interaction, Interfaces and Companions in Assistive and Therapy Robotics at HRI ’08.

[Robert] Williams died instantly in 1979 when the robot’s arm slammed him as he was gathering parts in a storage facility, where the robot also retrieved parts. Williams’ family was later awarded $10 million in damages. The jury agreed the robot struck him in the head because of a lack of safety measures, including one that would sound an alarm if the robot was near.

The industrial robot that served up the death blow wouldn’t be considered “social” in most domain interpretations. However, the outcome of this human-robot interaction certainly had social repercussions.

When I declare my interest in social robotics, the definition is being constructed from a decidedly designer’s perspective. These metal creatures are social because their creators want them to serve social functions. Perhaps that definition needs to be broadened to include the interaction perspective. If Robert Williams had lived, for instance, would he have described his accident more like being hit by a bus or a boxer?

P.W. Singer, author of the 2009 book Wired for War, gave a TED talk last February about the disturbing trends in use of robotics to automate armed conflict. In his presentation, Singer gave this prediction:

“When historians look at this period they’re going to conclude that we’re at a different type of revolution, a revolution in war, like the invention of the atomic bomb. But it may be even bigger than that, because our unmanned systems don’t just affect the ‘how’ of war fighting, they affect the ‘who’ of fighting at it’s most fundamental level.”

Singer goes on to recall what a roboticist at the Pentagon once told him: “There’s no real, social, ethical, moral issues with it comes to robots. That is, unless the machine kills the wrong people repeatedly. Then it’s just a product recall issue.”

While the main point of Singer’s talk is about the dangers of disconnecting ourselves from the operation of war, there is a corollary relevant to any discussion of social robotics. Namely, any technology that deals with humans may become inherently social.

Maybe this thought experiment leads down the wrong path for a dissertation, but I have to question whether human outcomes force a change in context, regardless of designer intent.

Autom, the weight-loss coach robot

Autom is a weight-loss coach that comes in the form of a small robot. This first product released by Intuitive Automata tracks dieting activity and cultivates a relationship with the person trying to become healthier.

The robot is an outcome of doctoral research done by Cory Kidd, founder and current CEO of Intuitive Automata. Kidd cut his academic teeth with Cynthia Breazeal’s Personal Robotics Group at MIT. His work with sociable robots—robots that understand something about how people interact with each other—focused on addressing real-world needs for long-term health care.

In this case, the problem is that the average diet is about 3-4 weeks long. We humans are good at making changes, but bad at sticking with them. According to Kidd, 95 percent of all diets fail, with any weight loss regained within a year. A key reason for success is that the dieter keeps track of what she is doing and is aware of how her day-to-day decisions affect long-term goals. Chances of succeeding increase with a personal trainer, an inaccessible resource for most people.

With Autom, the robot becomes the trainer. It conducts daily conversations that emphasize relationship over goals and provide affirming encouragement and advice. Each 3-4 minute session begins with small talk and data input via a touch screen. Autom only asks about weight once a week (not daily). As evidenced by the difficulty people had in giving the equipment back after testing, the robot engenders trust and attachment.

Why Does This Work?

In the podcast interview, Kidd described the foundational research that supports Autom’s design. He looked at human responses to computers and robots, generating two key insights from the differences:

1. Even accounting for an initial novelty effect, people engaged with the robot more quickly and enduringly than with a computer.
2. When considering the information being disseminated, people find a robot more trustworthy (“The robot is seen as the more immediate thing.”)

These important tidbits come as result of a 6-week experiment involving 45 overweight people who wanted to diet. They were divided into three groups: one got robots, and the other two got either a computer or pencil and paper. All of the weight-loss tracking was done through these tools. For those without Autom, the diet lasted 3-5 weeks. For those with Autom, diets lasted 8 weeks … longer than the study.

Kidd also discussed the properties of embodiment that may play an important role in this connection between human and robot. Having a physical presence is what makes the most difference. Otherwise, effective traints depend on the specific application and context of use. (A list of Cory Kidd’s academic publications can be found at the MIT Media Lab web site.)

While I don’t dispute the results of these inquiries, I wonder if the right question was asked. Is the engagement really a result of the robot form, or does it have more to deal with opportunity and accessibility? Is a combination of social robotics themes responsible for the human-robot connection, or is it simply because the placement of the computer screen embedded in Autom fits into a dieting routine better? A Wii Fit has some overlap with Autom’s role in weight-loss, for example, but it is only accessible in certain places and contexts. I can imagine this study being adapted for other design spaces with special attention paid to where the optimal moments of interaction are situated.

I admire the human-centric origins of Autom, but I am left wondering how much the success was attributed to the robot and how much to other non-robotic distinctions between the tracking methods.