Agent Culture

Human-agent interaction in A Multicultural World

Sabine Payr, Robert Trappl

• Publisher: CRC Press
• Year: 2004
• ISBN:
  9781135617271
• Edition: 1
• Topic: Computer Science
• Subtopic: Human-Computer Interaction
• Index: Computer Science

This volume began with a workshop of the Austrian Research Institute for Artificial Intelligence held in 2001. Concerned with embodied agents as cultural objects and subjects, the book is divided into three parts. It begins by drawing attention to the cultural embeddedness of technology in general and agent design in particular, as a reminder that

PART I

CULTURE(S) AND AGENT TECHNOLOGY

CHAPTER 1

The Agents of McDonaldization

Phoebe Sengers

Cornell University

I used to build agents. In the old days, I built interactive computer characters, or animated creatures with idiosyncratic personality, who can express that personality in interaction with human users. To do so I used a programming framework that focused on expressing the agent author’s vision of personality through an ad hoc and flexible organization of complex, hand-written behaviors (Loyall & Bates, 1991). This approach excited me because of its openness: It did not impose a rigid theory or structure on the personalities that could be expressed. This minimum-commitment model made possible the creation of arbitrarily complex and idiosyncratic personality. In this approach, agent design was thought of as craftwork: Authors could manipulate the materials of electronic code to express their unique personal vision of personality and life.

One evening I had a horrible nightmare. In my dream, 10 years had passed. I had graduated and the field of agents with personality had boomed. In this brave new world, I had found a job at a large fast-food corporation, working on the software assembly line building intelligent, fran-chised characters to be packaged with Happy Meals. My dream of intelligent characters who represented the full breadth, beauty, and complexity of life had been reduced to Ronald McDonald and the Hamburglar. I woke with a shudder. As dawn broke, I laughed at my imagination, went back to sleep, and forgot about the dream.

Fast forward 10 years. Agent-building has, in fact, exploded. My research group at the time has formed a start-up company, Zoesis, whose current product is intelligent, trademarked characters to support branding at company Web sites. Agents are being developed for use as virtual salespeople in virtual malls. Virtual people, like Lara Croft, are branded and marketed as mega-stars. Just as I had subconsciously feared, agents have been reincorporated to create the kind of personalities with which corporate America feels most comfortable: standardized, optimized, and under control.

In this chapter, I argue that this convergence between agenthood and American consumer culture is no accident. The concept of agency, as well as major trends in agent design today, is deeply rooted in Western, industrialized culture. Agents are an example of and propagate the trends to McDonaldization that Ritzer (1993) identified as emblematic of postindustrial consumer culture: efficiency, quantifiability, predictability, and control. Agents are profoundly cultural; their very definition incorporates assumptions of Western, industrial culture. Because of this, the notion of culturally flexible agents is an oxymoron.

To make this argument, I take you on a trip back in time. We go back to the beginning of Western industrialization to understand how the shift from craftwork to industrialized labor on the assembly line changed our understanding of work and, eventually, human behavior. We see how human behavior became something to be rationalized, quantified, and mechanized. As we move forward in time, we recognize how contemporary trends in consumer culture are rooted in the mechanisms and philosophies of the assembly line. All along the way, we find cultural shifts that are oddly familiar to those building agents; they make the notion of agenthood possible and shape the way we think of and build agents today.

AGENTS: THE VERY IDEA

Various definitions of agent have been put forward by the agent research community, none of which has acquired universality. Here I use the term agent in its broadest sense. By agent I mean a piece of soft- and/or hardware that is intended to represent a complete person, animal, or personality. An agent may be an animated figure, a robot that may or may not be human- or animallike in form, or a nonvisualized piece of code. By using the term agent, an agent’s designers implicitly state that the code is best thought of as a self-contained, semi-intelligent unit corresponding in a vague sense to what we think of as a conscious being.

This definition of agent is intentionally quite broad, encompassing a huge range of potential technologies, applications, and motivations for agent-building. A wide range of agent applications is currently being used and developed. These include animated characters for video games; interface agents that provide a humanlike presence on Web sites, databases, and other programs; sales agents that engage in retail activity with human users; robotic toys such as the robotic dog Aibo and the hyperrealistic doll, My Real Baby; telephone-answering agents who can help users sort through the complexities of voice-mail systems; and so on.

What these agents generally share is a basic process of agent design. To build an agent, generally speaking, you begin with an idea of a kind of agent you wish to build. This agent may be a natural agent, such as an insect, dog, or person, or an imagined agent, such as a component of a large software application or a personality one wishes to embody in code. Once the concept of the agent is chosen, you break the behavior of the agent into units, dividing each of the units up into steps, implementing each of the steps and units as a mechanical process, and assembling them into the final agent. The kind of unit you choose varies depending on the kind of agent technology one is using and the aspect of the agent you are trying to model. Typically, agents are split into functionalities—problem-solving units that give agents specific capabilities, such as the ability to see, the ability to understand English-language texts, or the ability to understand basic rules of social interaction—and/or behaviors—integrated collections of functionalities that work together to engage in some coordinated real-world activity, such as hunting, finding and sorting trash, or searching for and greeting other agents. Although we have the technology to build subtle, complex, and powerful functionalities and behaviors, it is typically a great challenge to knit together the various functionalities or behaviors into an agent whose overall behavior is coordinated and coherent. What we see first is that this process of agent design has its roots in early industrialized restructuring of human labor.

INDUSTRIALIZATION AS AN EARLY FORM OF AI

In a sense, the mechanical intelligence prouided by computers is the quintessential phenomenon of capitalism. To replace human judgement with mechanical judgement—to record and codify the logic by which rational, profit-maximizing decisions are made—manifests the process that distinguishes capitalism: the rationalization and mechanization of productive processes in the pursuit of profit. (Kennedy, 1989, p. 6)

The history of the Industrial Revolution is often told as follows: In the beginning, there were craftspeople who owned their own tools, manufactured articles in their own idiosyncratic ways, and whose work was largely integrated with their way of living. As the Industrial Revolution begins, these workers are collected into factories, where they work together using the owner’s tools. This owner, in an attempt to make work more efficient, begins to streamline the production process. Instead of having each worker build a piece from beginning to end, the production line is developed, where each worker works on some small part of the final piece. Work is broken up into stages, each of which is accomplished by a single worker; each stage is standardized so that articles can move from stage to stage without breaking work rhythm. Once work is divided up into standard stages, some of the steps can be done by a machine. Instead of building an article from beginning to end, workers now tend machines, which are each doing small steps of the article’s production.

At each stage, work becomes more rationalized, predictable, and efficient. Workers on the assembly line can generate more articles, and the articles lack the idiosyncratic variation of normal craftwork. Instead of doing whatever he or she wants in a haphazard order, a worker has a fixed set of steps in which he or she engages. The intelligence of the worker, which he or she previously needed to monitor what he or she was doing and make active decisions about how work should proceed, is now embodied in the structure of the assembly line. Workers no longer need to think; the factory machinery does the thinking for them. Even before computers, industrialization takes the first baby steps of AI.

These traces of industrialization can be seen in the way we build agents today. The AI researcher building an agent follows the same basic line as the factory manager designing new production processes. Just as the factory manager attempts to design and reproduce a preexisting work process, the AI researcher would like to copy a natural process—an agent or idea of an agent. The factory manager breaks this process into logical steps, figuring out which steps should happen and in which order. Similarly, AI researchers analyze the agent’s behavior to categorize its activity into typical behaviors, enumerate the conditions under which those behaviors are appropriate, or identify functionalities and develop well-defined algorithms by which those functionalities can be implemented.

Just as the factory manager embodies each step in machinery that can run with a minimum of human supervision, agent designers implement a mechanical version of each behavior, hooking them together so that they largely reproduce the imagined or real behavioral dynamics of the original creature. The early industrialist and the AI researcher are engaged in the same project: We analyze, rationalize, and reproduce natural behavior.

If agent design is like factory management, the nature of agent behavior may share characteristics with industrialized labor. In fact postindustrial work is radically different from preindustrial work, in both the qualities of the articles produced and the human experience of engaging in that work. The act of embodying work in the production line changes the nature of work. Work becomes more rationalized and less personal; workers are more dependable and more bored; the articles produced become more standardized and less individual. Several special qualities of postindustrial work and life have been identified by cultural theorists and industrial historians.

Reification. Things that were once thought of as ineffable or abstract become thought of as concrete. Labor, for example, which was once not strictly separated from the rest of life, becomes something that is measured and sold per piece or per hour. Once things are reified, they can be sold, becoming commodified, to be exchanged for particular sums of money.

In agent design, we too reify qualities of our agents. We sort ineffable behavior into well-defined categories, defining them cleanly and embodying them in code. We create rigid, formal definitions of categories of life, which are then said to drive behavior. In emotional modeling, for example, we define small sets of well-defined emotions, which are supposed to represent the rich complexity of human experience. Similarly, interaction between people and agents is reified; all too often interaction with agents is, at heart, simply a choice among some predefined options, although it may be masked by the appearance of more complex behavior.

Specializationl/Atomization. Workers no longer engage in the entire work process. Rather, they each perform some small function within the process as a whole. Without an overview of the process, workers no longer need or are able to adapt to one another; each part takes place without reference to the others. Without feedback among the pieces, each piece is built in isolation, the whole being merely the sum of each individual, separately designed atomic part.

The production process, which was once a wholistic attribute of individual workers, is broken into rationalized parts, each of which is embodied in pieces of machinery or in production rules that regulate how they interact. Workers, who were once thought of as individual humans deeply embedded in the context of their daily life, now become interchangeable parts of the production process, whose time is to be sold to the highest bidder. They move from factory to factory, no longer connected to their home place or even a particular manufacturer. Workers see themselves as free and atomic individuals bound by no human ties.

Agents’ functionalities or behaviors are similarly specialized, developed separately from one another and difficult to integrate. Agents are specialized for particular uses. Rather than representing the full breadth of human behavior, they are optimized for their use in particular situations. Interaction with them can be unnatural precisely because of their goal-directed, specialized design; unlike people, agents cannot handle behavior outside of their mandate. Like the real estate-selling agent REA’s small talk (Bickmore & Cassell, 1999), focused on making the sale, agents interpret all behavior within the limited confines of their purpose.

Standardization. The idiosyncrasies of craftwork mean that one can never be sure what the produced goods will be like. The factory owner, however, who consolidates craftwork and has promised broader distribution networks goods of a particular kind and quality, wants to have some guarantees that the factory will produce similar goods no matter which workers are present on a particular day. The idiosyncrasies of personal work are no longer valuable. Instead the owner introduces steps of production control to ensure that the output is always similar. The qualitative, human, individual dimension of work is eliminated and replaced by efficient, controlled, and standardized work processes.

Similarly, standardization is one of the goals of agent architecture: to provide one single technology that can support multitudinous agents, each variations on a theme. Some architectures, like Loyall and Bates’ (1991) Hap, allow for idiosyncratic variation according to the taste of the designer, but by and large such labor-intensive variations are considered a liability by the agent-building community. Variations in which one can alter agent behavior simply by changing the levels of a few parameters are preferred. A strong theme in agent research is the search for general rules behind behavior, which allow one to see behavior variations as parameter setting. In this way of thinking, all agents are fundamentally alike. Interactions between agents and people are standardized; the same technology is used in all cultures.

Formalization. The individual, material properties of workers and the material they operate on is ignored except insofar as it impinges on the production process. As the production process becomes more and more efficient, extrinsic considerations—whether social, spiritual, or physical—are left out. The production manager thinks of the production process in terms of abstract steps without reference to the particular identity of the worker or chunk of material involved. The factory is set up to enforce this abstract, impersonal view, which then seems to be an accurate reflection of the real. Individual differences become noise, unvalued and only reflected on to control their effects.

The individual, material properties of humans are also degraded in agent research. Generally speaking, agents are immaterial, presented as images on a screen. At best their physicality is simulated. There is some work on robotic agents, yet their physicality has little in common with human physicality, especially the aspects of physicality that are degraded or problematic in our culture. Few robotic agents bleed when cut and fewer still menstruate.

More generally, all aspects of being human that are considered inessential to the task at hand—whether this is a specific task the agent must fulfill or the more general problem of modeling an abstractly defined intelligence—are ignored. Individual differences are parameters in a general model, nothing fundamental—we are modeling intelligence, not a particular person at a particular time. Formal models of being human are developed and seen as explanations of the complex, difficult to define real.

With agents, human interactions are formalized and instantiated as code. Unlike telemarketers, agents can never break out of the script. For us this means unformalized interactions with them are no longer possible.

Mechanization. To maintain standard production, workers are given less and less leeway in decisions about their jobs. Rather than relying on the worker’s judgment, the factory manager uses standardized production rules to ascertain that the product is always made the same way. As the steps of the production process are more and more formalized, the worker’s intelligence becomes ...

Table of contents

• Cover Page
• Title Page
• Copyright Page
• Preface
• Introduction
• Part I: Culture(s) and Agent Technology
• Part II: Design for Cross-Cultural Believability
• Part III: Agents for Intercultural Communication
• Suggested Readings
• Contributors