Amy and Klara is a robotic system composed of two synthetic speech robots designed by Marc Bohlen (2006a).20 According to Bohlen, the purpose of this system is to explore the expectation, construction, and maintenance of norms of public speech and the ways that those norms and our experiences with variations to those norms are affected when transferred to nonhuman entities that are mimicking humans (2006a, 2006b, 2008). This topic of exploration is not unusual in the context of human-robot interaction research. In that field, one can easily imagine social science experiments grounded in some notion of productive communication or cooperation, conducted under controlled circumstances with clear variables, resulting in empirical findings and design guidelines. Bohlen 's work is differently situated, however. Amy and Klara is an experimental art and engineering project, and as such, the practices and agendas of conventional human-robot interaction research do not need to be upheld. With Amy and Klara, expectations of robot communication are explored not in the common form of social interaction but at one of the limits of human communication—cursing. By this exploration of the limits of the social qualities of robots and human-robot interaction, Amy and Klara is akin to Blendie and Omo in challenging our assumptions of human-robot relations, and it provides yet another example of an encounter with social robots that is made agonistic through the design of its embodiment.
The two synthetic speech robots Amy and Klara are physically instantiated as stationary boxes that are painted hot pink, equipped with speakers, and that curse at and argue with each other (figures 3.4 and 3.5). One of them (Klara) speaks with a German accent. Bohlen's choice of speech and cursing as the basis for these robots presents a critical perspective on computational speech recognition, speech generation, and adaptive dialog. As he explains (Bohlen 2006a),
It is not only the disconnect between a human voice and a box that produces it that can make one feel uncomfortable. It is also what these voices have to say to us. The language of synthetic speech recognition and synthesis systems is a highly selective subset of the full, rich and messy body of linguistic corpora that comprise our oral and written languages. Exclamations are absent, questions are rare and the vocabulary is generally optimized for commerce.
Here the charged world of foul language is under investigation. Swearing offers several interesting conduits into a critique of the under-exposed normative tendencies in automated language representation and social robotics. Why are most smart gadgets and toys friendly and playful, why are they usually modeled as pets or servants? Machines that curse and pick a fight might offer a more realistic preparation for a shared future between machines and humans.
Amy and Klara is agonistic in multiple ways. As is made clear in the preceding quote from Bohlen, the purpose of these robots is to question
Marc Bohlen, Amy and Klara (2006a)
foundational assumptions of human-robot communication and language- based robot expression. As with Omo and Blendie, the social character of social robots is being explored—what kinds of communication are assumed proper and privileged and what kinds of communication are not and are thus left out of social robot design. The remainder in this case is cursing or what would commonly be considered abusive, juvenile, petty, or dysfunctional communication. Similar to Omo, the remainder with Amy and Klara is a mode of expression and interaction that falls outside of the rational and productive directives that tend to drive mainstream social robotics. More than simply documenting and representing issues of social robot design, Amy and Klara is a demonstrative, interactive instantiation of these issues.
The design of Amy and Klara leverages and explores computational text- to-speech, automated speech recognition, and aspects of computational vision as they figure into the construction of social robots. With Amy and Klara the substance of the robots ' speech is produced by software that
A view of the electronics of Amy and Klara, Marc Bohlen (2006a)
accesses and reads online lifestyle magazines. The content from these Web sites is parsed and becomes the basis for the construction of ontologies, or computational models of the world, possessed by the robots. As these linguistic models develop, certain words are given weight by virtue of their frequency. These words become the speech that the robots exchange. But sooner or later, there is a misunderstanding between the robots. When one robot fails to understand the other robot, it might be because it has developed a different ontology and thus is speaking a word the other robot does not know, or it might be because of the distortion inherent in the microphones and speakers, which is exacerbated by the German accent of Klara. Whatever the reason for the misunderstanding, "dissent arises and they begin to call each other names" (Bohlen 2008, 211). When one robot detects the foul language of the other, it responds in kind with an utterance containing foul language. These quarrelsome exchanges rapidly increase in intensity. As Bohlen (2008, 212) describes the basic procedural structure of the software regulating the social exchange:
Repeated use of a curse word from scale n leads to the selection of a curse word from scale n + 1, provided the following word is recognized as a curse word within a given time frame (otherwise the aggression levels recede). Since recognition and utterance occur in quick succession, both a low level exchange (when recognition results are poor) as well as a heated escalating fight, if recognition results are positive, are possible.
In addition to linguistic modes of exchange, the design of the robots ' embodiment employs a computational vision system that regulates their interaction. Each robot is equipped with a camera that is pointed at the other robot, and the camera 's field of view includes parts of the surrounding room. Once each robot has developed its corpus of words, and they have been catalogued and compared, the verbal exchange state is triggered by this vision system, which initiates the dialog when it identifies the color pink (that is, when it registers the presence of the other robot). The vision system of both robots is also able detect the presence of a human, as identified by overall shape. When a human is detected by the vision system, the robots change their behavior by first lowering their voices, curtailing their dialog with one another, and finally asking the people present to leave. So these social robots are designed to be social with each another but asocial with people.
The color pink plays an important role in the design of Amy and Klara. On the surface, the color pink serves as a gendering device. Together with the voice and the names, it works to construe the robots as prototypically feminine. But the color pink has another purpose beyond serving as a signifier for human understanding. In robotics, the color pink is often used as a test color for computational vision research. Because of the distinctive chromatic qualities of pink, specifically hot pink, it is commonly used for location and targeting purposes in computational vision. There are, for example, robot demos and competitions in which robots search for pink- colored tags in the environment or on people. The color pink thus does dual duty in the design of Amy and Klara. It genders the robot in human terms and is simultaneously a fundamental aspect of a robot-centric embodiment, enabling its vision in a manner that reflects the robot 's distinctive sensing capacities.
Like Blendie and Omo, Amy and Klara evokes the uncanny. The primary way it does so is by the content and character of the dialog, which is at odds with the expressive capabilities of the robots. When people hear the exchanges between Amy and Klara, they do not mistake the dialog as occurring between humans. The prosidy of the speech is dull and dronelike. Due to the processing time involved for one robot to recognize the other 's utterance, the timing between the exchanges feels fractured. The German accent of Klara amplifies the incommensurate quality of the scenario. Such an accent is associated with sternness, but in this context, the attempt to imbue the robot 's voice with a greater amount of personhood only draws attention to the awkwardness of the technology in mimicking human dialog. The overall experience transforms the line between a human mode of communication and its replication into a fissure.21
In addition, Bohlen 's framing of the dialog as fundamentally antagonistic is striking. Unlike other social robots, Amy and Klara are not designed for companionship or therapy; they are designed to engage in verbal conflict with one another. So even the relatively benign exchange that Bohlen (2006a) provides as an example in the project documentation is disconcertingly at odds with our preconceptions of what and how robots might verbally communicate:
Rl: "Hey you."
R2: "Leave me alone." (synthetic German accent)
Rl: "What is wrong with you?"
R2: "Leave me alone please." (synthetic German accent)
Finally, the design of Amy and Klara also challenges assumptions concerning social robots in another, more fundamental manner: they appear to be robots that exist primarily in relation to one another, with only a peripheral connection to humans. In this way, they confound expectations of social robots as intelligent artifacts that are social with humans. The social qualities of social robots serve primarily as a mode of robot-to-robot interaction, and the robots are social with one another, to the exclusion of humans. This disassociation from humans is made all the more strange by the fact that robots are nonhuman entities that depend on a human mode of communication with each other: they speak to each other. One result of this odd scenario is a further troubling of the common we/they distinctions between people and machines. In the case of Amy and Klara, the we and the they are conflated as the robots take on decidedly human qualities.
The design of Amy and Klara exemplifies a variation on the tactic of reconfiguring the remainder that I call agonistic reification. In what might first appear as a paradox, reification—the process of objectifying a thing— can be employed agonistically in the design of robots to produce encounters that objectify human beings in ways that prompt critical reflection on the processes and effects of objectification. The linguistic content coupled with the ways in which language is handled in the design of Amy and Klara provides an example. The use of a German accent to connote sternness plays on stereotypes, which are means of socially objectifying people. But this objectification extends social construction also to include the technical construction of the robot. The design of the robot's embodiment—the computational rendering of a synthetic voice into a German accent— requires objectifying the synthetic voice itself, making it into an element that can be examined and manipulated. In its original form, the synthetic voice does not have a German accent. The German accent is constructed from what is considered to be a more neutral American accent by procedurally "swapping select vowels and consonants" in the text-to-speech software. For example, "Welcome" is transformed to "fvelk2:m."22 So Bohlen's exploration of the social qualities of speech (for example, expectations, norms, and stereotypes) is an exploration of the social and technical qualities of speech. Indeed, in the design of Amy and Klara and arguably all social robots, the exploration of the social depends on an exploration of the technical, in effect, melding these two into a single thing. Through such reciprocal plays on stereotypes, language content, and computational phoneme manipulation, the design of Amy and Klara fulfills both parts of the definition of objectify: it transforms a quality or condition into a unit of analysis, and it makes that quality or condition actual—that is, materially instantiated and experiential.
Reification in this case serves the agonistic purpose of highlighting assumptions concerning human qualities and relations and the way that those assumptions are imbued in computational systems. That is, reification is one way of bringing the remainder to the fore. Selecting and transforming a particular quality into unit of analysis provides a means to examine and manipulate that quality or unit.23 With Amy and Klara, dialogic interaction is reified, examined, and manipulated. This occurs through a nesting or layering of objectification. Dialogic interaction becomes reified as cursing, and that particular irrational mode of human speech exchange is reified into a series of stimuli and response mechanisms and procedures, which are reified through the discrete computational manipulation of speech.
As with the prior examples of engineering the uncanny, this agonistic encounter is achieved through the design of the robot' s embodiment. But contrary to the prior examples of Blendie and Omo in which embodiment was a quality of a robot that coupled it with humans, embodiment here is a quality of a robot that couples it to another robot. Even the pink coloring of the robot eludes simple explanation as a means for expressing human gender norms: it functions to enable coupling between the computational objects themselves. Thus Amy and Klara extend the machinecentric point of view witnessed in Blendie to the point of hyperbole. The design of the robot 's embodiment operationalizes the human as a base form of stimulus. Such agonistic reification can be cast as an ironic but still critical response to concerns about translating human qualities into machines. To use Suchman's term, the tactic of reification performs a kind of "retrenching." This retrenching or performance of reification, however, is done in a manner that is self-mocking and contradictory, as it reduces sociability to a single, bounded feature and then instantiates that feature in robot form to produce an exaggerated effect (or affect, as the case might be). In this case, the irony of Amy and Klara demonstrates in material and experiential form the problems of extending machine sociability in human terms, and Amy and Klara functions as incitement for reflection—to consider assumptions in designing sociability through robot embodiment. For example, one assumes that one' s social interactions with robots will be congenial. Moreover, one assumes that social robots will be social with people. Amy and Klara demonstrate that there is no essential basis to those assumptions. There is no technical or social reason why social robots must behave that way. In considering how we will comingle with intelligent systems and the role of design in shaping those experiences, Amy and Klara is a test case of a scenario in the extreme. The two robots counter saccharine modes of expression and interaction by privileging the abusive and petty, and in doing so they provide another vector along which to consider the possibilities and limitations of designing sociability into computational entities.