Blog Post

Embodied Agents

Have you met Anna from Ikea?  I have noticed a new trend of automated computer agents greeting me at commerical sites, graciously offering their assitance and procuring a chat window to personally assist my shopping experience. "Anna" is one such agent represented on the IKEA site by a smiling brown-haired, fair skinned avatar wearing a headset and IKEA shirt. The proliferation of these agents in online shopping sites, search engines, and online education practices has led me to explore the notion of anthropomorphizing computer agents and unpack the values systems and cultural frameworks that underlie such representations.  This blog post is a brief probing of agents and anthropomophization.

Brenda Laurel (1997) claims that an agents is “any person or institution who is empowered by us to take action on our behalf (p. 212).”   She identifies four kinds of computer-related tasks where agents may appropriately “provide expertise, skill, and labor (p. 212).”  The first of these tasks is information, generally comprising navigation and browsing functions, information retrieval, and the sorting, organizing and filtering of data.  Agents are also adept at performing work functions such as reminding, programming, scheduling, advising.  Learning is the third skill-set appropriate for agents and includes coaching, tutoring, providing.  Lastly, agents are used in entertainment and are found in gaming situations performing, playing with and against human users.  According to Laurel, agents’ work is appropriate in these instances because these tasks either require complex algorithmic solutions or “complete parametric specification by the human user (p. 212).”   

Laurel draws the distinction between the Aristotelian view of the agency (taking action) and the socio-legal understanding of agency as “empowerment.”  The Oxford English Dictionary (OED) presents three meanings of “empower” that demonstrate differing positions of control and authority.  The first meaning, and the one that Laurel is drawing on, is “to invest legally or formally with power or authority.”  This definition is akin to the definition of “proxy,” which the OED defines as “the agency of a person appointed to act in place of another; the action of a substitute or deputy; by or through a substitute; not in person.”  Here the power is not located in agent itself, but rather is wielded through them on behalf of someone else (a human).  The second given meaning of empower is “to impart or bestow power to an end or for a purpose; to enable, permit,” which remains close to the first definition.  The third definition, however, more forcefully articulates the inequality of the agent’s position: “to gain or assume power over.” 

I linger on these definitions to emphasize an inherent system of power already implicit in the idea of agents, be they human or computer.  When these agent computer programs are anthropomorphized into embodied representations, a second system of power (that of the material body) is inherently invoked.  While the power structure of human versus machine is (often) visible, the power structure of the signified body is largely rendered invisible because of cultural value systems that normalize the institutionalization of the body as a raced, gendered, and sexed object.  It is important to understand these two power structures as operating simultaneously but in intertwined concert as the computer agents are designed, marketed, and used.

 Though there are computer-related tasks that lend themselves to being performed by agents, it is not a given that agents need to take on anthropomorphized or fully embodied characteristics.  Instead, anthropomorphization is a strategy that provides psychologically and functionally advantageous scenarios for human users (Laurel, p. 211).  According to Laurel, “the kinds of tasks that computers perform for (and with) us require that they express two distinctly anthropomorphic qualities: responsiveness and the capacity to perform actions (p. 210).”  Laurel goes on to explain that humans are adept at relating to and communicating with other humans, and indeed have developed skill sets for doing so.  She is careful to recognize that “anthropomorphization is not the same thing as relating to other people, but is rather the application of metaphor with all its concomitant selectivity (p. 210).”  To paraphrase Lakoff and Johnson (1980), the anthropomorphic metaphor provides us not only a specific way of thinking about a topic, but also a way of acting towards it (p. 34).  In other words, anthropomorphization is a tool that repurposes human skill sets for sociability for the translation of otherwise foreign interactions with computer agents into more familiar social interactions.

Metaphors are not without limitations.  They operate through the drawing of parallels and emphasizing similarities in the broad strokes of a comparison—powerful tools to be sure.  However, subtleties, complexities, and nuances are necessarily sacrificed in these parallels to make them convincing.  This means that metaphors often map onto common stereotypes and tropes, forms that also gloss over complexities for the sake of “universality.”  When applying the anthropomorphic metaphor to technological agents, it is important we understand the kinds of obfuscations and stereotypes we are unwittingly participating in by the very nature of the activity.

Technological agents are designed to have particular anthropomorphic traits.  The following assertions highlight the link between physical inscriptions and value systems as determined by socio-cultural norms:

"External traits like diction and appearance must be shaped to suggest those internal traits (values, heuristics, etc.) which determine how a character will make choices and perform actions (Laurel, p. 217)."

"As in drama, traits can be represented directly through appearance, sound, communication style (external traits), which in turn cause us to infer traits on the level of knowledge and thought (internal traits) (Laurel, p. 210-211)."

These also serve as a reminder that technological agents are designed into being and thus are reflective of the designer’s own social reality, in addition to the norms that govern the wider social context.  In a sense, what is being designed is a stereotypical performance, a metaphor, of a personality.  As Butler (1993) describes, performativity is “not a singular ‘act,’ for it is always the reiteration of a norm or set of norms, and to the extent that it acquires an act-like-status in the present, it conceals or dissimulates the conventions of which it is a repetition (p. 12).”  Butler’s assertion points to the possibilities for performance, itself, as a tool for normativity.  The performance of a metaphor in the form of a technological agent, therefore, presents a powerful platform for norms and values to be both stereotypical and invisible. 



Lakoff, G. and Johnson, M. (1980). Metaphors we live by. Chicago, IL: The University if Chicago Press.

Laurel, B. (1997) Interface agents: metaphors with character. In B. Friedman (Ed.), Human Values and the Design of Computer Technology (pp. 207-219). Cambridge, MA: Cambridge University Press.


No comments