Online Media, Agency and

Responding to Online Agents

As Nass, Lombard, Henriksen and Steuer (1995) demonstrate, human beings have an innate tendency to anthropomorphize technologies, both physically and psychologically. An individual's tendency to anthropomorphize is shown to predict his or her acceptance of computers in routinized (e.g., telephone operator), interpretive (e.g., novelist), and personal (e.g., judges) roles. Sundar (2004) showed that the level of psychological anthropomorphism even predicted a tendency to be loyal toward particular computer terminals. Reeves and Nass (1996) argue that even when technological interfaces do not expressly involve anthropomorphic cues or representations, human beings are hardwired to respond to communication technologies in social ways. Sundar and Nass (2000) call this tendency source orientation and demonstrate that users make social attributions directly to machines as if they are autonomous information and communication sources, without needing to invoke an underlying human agency such as the programmer or networker.

Sundar and Nass (2001) proposed a typology of online media sources by suggesting not just that traditional visible senders or gatekeepers of information are considered sources, but that the medium (technological source) and even the receivers (audience as source and self as source) can be construed as an independent source. In their experiment, they showed that users' perceptions of news content attributed to these different types of online sources were significantly different, implying that people are able to psychologically distinguish between professional journalists as gatekeepers and algorithms (as in Google News) as selectors of our information, and between themselves as source (e.g., portal sites) and other users as a collective source (e.g., collaborative filtering systems such as those used by http://Amazon.com wherein others' behavior is conveyed as an input for decision making).

However, user responses to these sources are based not always on effortful processing of source characteristics but on mental shortcuts (cognitive heuristics), which can be triggered by simply invoking the name of the source. For example, Sundar, Knobloch-Westerwick, and Hastall (in press) show that the computer as a source can trigger the machine heuristic, emphasizing objectivity in information selection and absence of ideological bias. But the operation of such heuristics is largely dependent on the cues embedded in the interface as well as content (Sundar, 2006). If an online chat bot (short for chatterbot, a program designed to simulate intelligent conversation) has a visibly human presence on the interface and is Caucasian and female (e.g., Ramona on http://www.kurzweilai.net), then users are likely to factor her race and gender into their interactions with this bot and their perceptions of the content it delivers. In fact, anthropomorphic agents attempt to approximate real humans in their outward appearance, with the use of such design elements as animation, gesture, natural language communication, animation, and personality—ingredients that, according to Heckman and Wobbrock (2000), form the foundation of “a dangerous illusion,” with potential for powerful persuasion through unethical, if not illegal ways.

...