Recently, the question of how we attribute gender to asexual entities like robots and AI has been something I have been pondering more and more. How we understand engage with entities like Alexa from Amazon, Siri from Apple, and Cortana from Microsoft is fascinating, as well as how we think about robots like BB-8 or Wall-E. My partner, Christopher, got me interested in this topic through his work on Adventure Time’s BMO, as well as his work on vocaloids like Hatsune Miku. I have written about it in relationship to the artificially intelligent operating system in Her. With these artificial agents increasingly becoming a part of our everyday lives, understanding how we make sense of them as “people” is of importance to communication scholars, as indicated by the upcoming International Communication Association preconference on Communicating with Machines.

But I had completely forgotten about this 2005 research proposal indicating my interested in this topic a decade ago, even though I did some work on it back then. I am glad to find this as I start to consider how to create a reception study that explores how people make sense of the gendered nature of artificial agents.

Interacting with an androgynous computer entity: Impact of gender ambiguity on perceptions and stereotypical categorization.

CarrieLynn Reinhard
The Ohio State University

Research Proposal March 16, 2005

Introduction

How we perceive the people around us will impact how we interact with them everyday. Commonly, what we can expect from an individual is primarily based on the person’s appearance, as visual cues can trigger certain stereotypic or experiential information stored in our memory or schemas. The perceptions triggered by visual cues can be modified by subsequent information from the individual’s behavior or situational cues surrounding the individual. Whether we like it or not, research has consistently shown that humans are prone to judge a book by its cover. One main piece of information humans gather to facilitate their interaction is based upon how they perceive the other person’s gender, as gender is a main schema with a variety of psychological traits and behaviors attached to it (Bem, 1981; Katz, Silvern & Coulter, 1990). If we have an idea as to the gender of the person we are interacting with, then we can better predict how they will respond to what we say or do.

Research into human-computer interactions has found a similar pattern of visual information impacting perceptions (Reeves & Nass, 2002). If a computer user perceives the computer or some entity on the computer (such as an agent or avatar) to be male or female, then that user will react in ways similar to if the perceived other was a real human being with which they were interacting. Typically, this reaction is based upon stereotypical perceptions of gender, and could thus have implications for a user interacting with the technology. In fact, studies have shown that women are less likely to interact with computer technology, and one possible reason for this is their perception of computers as male-oriented (Colley & Comber, 2003; Martinson, online). Oosterwegal, Littleton and Light (2004) found that as perceptions of computers moved away from falling along gendered lines (i.e. computers are just for boys), then attitudes towards computers became more positive. Could users’ interactions with computers be impacted by gendered reactions to how they perceive computers and computer entities?

The purpose of this study is to investigate the potential for computers and computer entities to embody a gender rare in the real human world, and to examine if this different embodiment could facilitate users’ interactions with the technology. In particular, this study will explore the concepts of gender ambiguity and gender androgyny as a means of creating positive interactions. Research has shown that people perceive androgynous individuals positively, as being more adaptable and psychologically adjusted, than either masculine or feminine individuals. Thus, the goal is to understand if creating technology with the ability to be perceived as androgynous would also create this positive perception. Also, as research has indicated the preference to create computer entities that are visually anthropomorphic in nature as a means of improving computer interaction (Parise, Kiesler, Sproull & Waters, 1999), the idea of creating an entity to appear gender ambiguous may serve as the means for creating an androgynous perception. Thus the research questions driving this study are:

RQ1: Does gender ambiguity, as a visually cued perception, relate to attributions
of psychological androgynous traits to computer entities?

RQ2: Does gender ambiguity, as a visually cued perception, relate to male or
female gender stereotypes?

RQ3: Does a gender ambiguous entity provoke more willingness to engage with
the entity than either a male or female entity?

While the main application and purpose of this study is to further explore the “mindless” reactions people have in their interactions with computer technology (Nass & Moon, 2000), as the research into “the media equation” is founded in findings from social psychology, the study of people’s interactions with gender ambiguous entities could illuminate certain aspects of various phenomena studied in that discipline, such as schema theory and expectancy violation theory. Thus there are possible application and theoretical implications from this study. First, could creating computer technology with gender ambiguous and gender androgynous perceptions improve user’s interactions with it? Second, what does a person’s perception of gender ambiguous individuals reveal about how such individuals are categorized and stored in memory, and how this perception then impacts attitudes and behaviors towards this individual?

Gender Androgyny

People often recognize others in terms of schemas, or mental representations of what others are supposed to be like and do, and these schemas contain information determined by culture, society and/or experience to be important in facilitating social interaction, even if such information falls into stereotypes or generalities that may not apply to individuals (Harper & Schoeman, 2003). Schemas, in the form of stereotypes, help people organize an overwhelming amount of social information so as to facilitate recall and processing of new information, which can then impact how this new information is routed and stored. Gender is one of the defining schemas people construct to differentiate people, as there are biologically only two sexes, which over time have been related to two genders, although the exact trait categories used to define each gender differs depending on society and culture (Bem, 1981).

Initially, gender was conceived as being dichotomous, as only male or female (Major, Carnevale & Deaux, 1981). Bem (1974), using her Bem Sex Role Inventory, reconceptualized this categorization into two dimensions, masculinity and femininity, in such as way that an individual can have differing levels of both dimensions. According to Bem, gender schematic individuals rate themselves with either many masculine or feminine traits, and are thus either masculine or feminine. An androgynous person, on the other hand, is aschematic as they rate themselves high on both masculine and feminine traits. An androgynous person could be described as being more masculine than a woman and more feminine than a man (Major, Carnevale & Deaux, 1981). Thus, about two decades ago, researchers began to investigate the psychological construct of androgyny.

Much of the interest in androgyny (and one may wonder if this was driven by the women’s movement for equality at the same time) centered around the idea that an androgynous person was the most well-adjusted psychologically because they could easily adapt to situations requiring them to activate either masculine or feminine traits and behaviors. Indeed, there have been studies showing this. Research into “tomboys,” or women with strong masculine tendencies, found that they felt more confident about their ability to succeed, and that masculine traits such as leaderships ability being tempered by feminine traits such as compassion were factors in this relationship (Hilgenkamp & Livingston, 2002).

Perceptions of androgynous people also support the contention that such psychological constructs are desirable. Major et al (1981) found that androgynous individuals were perceived as being more popular and interesting than masculine or feminine individuals, as well as being better adjusted, competent, intelligent and successful. Arkkelin and O’Connor (1992) found that androgynous profiles were more desirable than either masculine or feminine profiles across various occupational types that could be gender stereotyped.

However, this positive perception does not appear to be without restrictions. Heilbrun and Schwartz (1982) discussed how female children are given greater leeway in experimenting with counter-stereotypical activities (i.e. being a tomboy) while this flexibility does not translate to male children (i.e. being a sissy). Deviation from gender stereotypes by men could result in more negative appraisals. Thus androgyny for men may not be as adaptive as it is for women. Jose and Wong-McCarthy (1983) in studying androgyny in intragroup relations found that only masculine females were rated positively as having excellent ideas whereas feminine males were seen as having poor ideas. The masculine females were seen as successfully combining instrumental and expressive qualities within the group, beating out true androgynous persons, as measured by the BSRI.

This negative appraisal against men who deviate from gender stereotypes appears to stem largely from violating the expectation that the individual should behave in a stereotypical masculine way. Rudman and Fairchild (2004) investigated the possibility of a backlash effect, where violating gender stereotypical perceptions could result in social reprisals for the violator. This type of reaction is inline with expectancy violation theory, such that an expectation about an individual is based on their appearance, and if their subsequent behavior violates this expectation, the individual could be perceived in a negative way. Participants rated their partner, either a man or woman, who succeeded in either a masculine or feminine game. Those individuals who succeeded in the cross-gender situations were perceived more negatively than if they succeeded in the same-gender situation. Rajecki, Graaf-Kaser and Rasmussen (1992) found that pairing a counter-stereotypical description with the picture of a man or woman resulted in less favorable perceptions of the target in regards to a job choice situation. However, if the individual’s appearance does not have these stereotypical features, would there still be a backlash when the individual behaved in a non-stereotypical fashion?

Gender Ambiguity

Ambiguity in semantics refers to any word or type of communication that can be “read”
with more than one meaning or interpretation. Thus, in the English language, the word “bank” can mean a place where money is stored or a place where the river meets the forest. Ambiguity is conceptually similar to androgyny, as both concern the incorporation of more than one state of being, creating a scenario where the perceiver could interpret this person in more than one way. However, ambiguity relates to communication or the message put forth by the object for the perceiver to perceive, and thus could be the external manifestation of an internal androgynous state. In terms of computer design, Goonetilleke, Shih, On and Fritsch (2001) describe ambiguity as any time an icon is perceived by a user has having multiple meanings at one time, thereby increasing response times to the icon and creating accuracy problems.

Ambiguity when perceiving people occurs when one individual has a number of traits that would otherwise be used separately to classify the individual into a specific social category. People attempt to identify gender based on facial cues as there are some general physiological differences between the two sexes, such as facial hair, skin smoothness and tone, and the size and shape of the nose, jaw, eyes and forehead (Zebrowitz, 1997). An individual having a predominance of facial features traditionally seen as either male or female would be more likely to be perceived as either masculine or feminine if no other social information, such as behavior, is available (Corneille, Brédart, Huart & Becquart, 2004). However, if the perceiver can receive clear and unambiguous information about a target’s counter-stereotypical traits, then the perceiver will be less likely to identify that target in a stereotypical way (Rudman & Fairchild, 2004). In times of ambiguity about visual stimuli, if the information is fuzzy or uncertain, individuals may rely on self-generated categorical information, especially in times when asked to recall the stimuli they just perceived (Corneille et al., 2004). Thus a person’s stereotypes may be employed to fill in any holes a person may have when attempting to perceive an ambiguous stimulus, such as an ethnic or gender ambiguous face where more than one category’s stereotypical features are present.

Some studies have approached this understanding of gender ambiguity by pairing an image of a man or woman with a description that contains counter-stereotypical traits or behaviors (Rajecki et al, 1992). Lobel (1994) showed films of boys playing either masculine, feminine or neutral games, and then asked elementary age boys to rate their gender traits and how willing they would be to engage with that boy in the video. The boys playing the feminine game (skip rope) was seen as more feminine, and only feminine participants liked this target. Overall, the boy playing skip rope was rated as the least popular. Katz, Silvern and Coulter (1990) used line drawings of male/female bodies, five drawings each, to represent a scale of less to more stereotypical (i.e. breasts, waist, hips, shoulders, posture), which were then combined with a feminine, masculine or neutral trait description (behaviors). Arthur and White (1996) showed children pictures of bears, void of physical gender characteristics but depicted as engaging in gender stereotypical activities, with one set doing neutral activities, such as a child bear walking towards an adult bear. The younger children perceived the gender of the bears according to their own gender, whereas the older children perceived them according to the behaviors in which they were engaging. Thus, all three studies agreed with Rudman and Fairchild (2004) by showing the importance of additional social cues, such as behaviors, on impacting perception, and also the importance of the social information interacting with the perceiver’s own gender schema.

A rare study that studied only physical gender ambiguity, Karniol, Reichman and Fund (2000) investigated how children use gender as a primary category for perceiving animal and cartoon characters. Their study elaborated on the work of Hodge and Tripp (1986) where children indicated they could not tell the intended gender of some characters until they have specific gender characteristics like long hair and dresses. Karniol et al. (2000) showed children images of cartoon animals with or without gender specific characteristics, based on hairstyle and clothes. How the children perceived those characters neutral in terms of these visual cues was dependent upon the children’s own gender orientation. “Masculinity and femininity interacted in such a way that androgynous, undifferentiated, and masculine children all perceived the gender-ambiguous characters as relatively more male, whereas feminine children perceived the gender-ambiguous animal characters as relatively more female. That is, high femininity led to perceptions of the animal characters as more clearly female.” (p. 388). Perceiving the female or ambiguous characters as such did not impact how well they were liked, but perceiving the male characters as male did positively predict liking the male characters. However, boys liked the male characters more than the female or ambiguous characters, whereas girls did not make such a distinction.

Thus there are studies indicating how gender ambiguity can be linked or affected by gender stereotypes, and there are studies showing how gender ambiguity and gender schema can affect attraction to the ambiguous target. Approaching the attractiveness issue from a reverse angle, Jackson (1982) found that physically attractive males were perceived as being more masculine than feminine, while physically attractive women were more feminine than masculine. At the same time, people perceived the androgynous females, whose pictures were paired with an androgynous trait profile, as less masculine but more feminine than the androgynous males. While androgynous individuals were rated as being more likeable and better adjusted, they were not perceived as attractive as feminine women or masculine men, although there is no information as to how the androgynous compared against the feminine men and the masculine women.

It would appear people are attracted to individuals that represent the prototypical gender stereotypes for their specific gender, which would mean a gender ambiguous appearing individual, if perceived as being androgynous, might not be as attractive to the perceiver, and this level of attraction could impact their subsequent interaction. However, Perrett, Lee, Penton-Voak, Rowland, Yoshikawa, Burt, Henzi, Castles and Akamatsu (1998) found that both Japanese and Caucasian men and women, when given the ability to modify male facial features until they create an attractive face, would create faces with a large ratio of feminine to masculine features (Zebrowitz, 1997). Later in the study, the participants were asked to rate these feminized male faces on gender traits. The results indicate that such faces had a high number of feminine traits as well as masculine traits. Hence, these faces, being a mixture of feminine and masculine features and thus gender ambiguous, are being perceived as androgynous, with the more socially likable traits associated with it. So Jackson (1982), by investigating only psychological traits and not physical traits did not capture a perception of physical attractiveness the same as did Perrett et al (1998).

Gendered HCI

The field of human-computer interaction is conducive to studying these questions, chiefly because the manipulation of representations of computer entities is far easier to accomplish in such a setting than it would be in the real world or even in a social psychology lab using real people. Second, similar studies as to how people respond to computers in gender stereotypical terms have been conducted for the past decade. Studies have indicated that if computers use a voice stereotypically masculine or feminine, then the computer will be perceived along these gender-stereotyped categorizations (Reeves & Nass, 2002; Lee, 2003). Nass, Moon and Green (1997) found that a male-voiced computer was perceived as more knowledge about technical issues while the female-voiced computer was rated as knowing more about relationships. Reeves and Nass (2002) cite other studies showing similar results: men and women more influenced by the praise from a male-voiced computer; participants evaluating the male-voiced computer as friendlier. There is also evidence that individuals, when they listen to voices synthesized so as to be neither male nor female, would ascribe a gender to it (as discussed in Reeves & Nass, 2002).

Lee (2003), following up on the studies regarding gender perceptions of computers based on voices, sought to understand if the same perceptions would be made if the computer entity was graphically represented as either a woman or a man. Indeed, using cartoon characters, she found that people made different gender-role attributions to the computer based on which gendered entity they were interacting with on a task. However, she only created figures using features that unambiguously categorized the characters as either masculine or feminine. The potential of the computer is its ability to synthesize human qualities to the point of removing gendered qualities. As research has indicated with voices, such ambiguity is not well received, but Reeves and Nash (2002) say this failing may be due to the quality of the synthesized voice. People may have lower tolerance for ambiguous voices than they do ambiguous faces; or they may in fact react with even more intense cognitive processing to categorize an ambiguous face as either male or female. The reaction to an ambiguous computer entity has not yet been studied, and the possible relationship generates a number of hypotheses.

Hypotheses

While not a study of gender ambiguity, Corneille, Brédart, Huart and Becquart’s (2004)
study of ethnic ambiguity did focus on facial information. In their study, they found that respondents used the most predominant feature of the ambiguous face in their perception and categorization of the image. Thus, when faced with ambiguity, the participants responded by making a decision as to a particular ethnicity by relying on the most salient social information. Forced attribution could be related to the individual’s need to know about the ambiguous person in order to be able to predict and control how they should interact with this other person (Fiske, 2004). Thus, in times of ambiguity without additional information that could be used to make a decision, perceivers may simply make a judgment call, which would mean relying on schematic data, gender-based or not, to gather the information necessary to make a decision. Along with the research on ambiguous synthesized voices, this result leads to the first hypothesis:

H1a: Entities visually representing a mixture of both masculine and feminine
features will be perceived as either male or female but not gender ambiguous.

H1b: However, entities visually representing a mixture of both masculine and
feminine features will have a slower response time in how they are classified as
either male or female.

Katz, Silvern and Coulter (1990) pointed out the need to understand the interaction of person’s gender schema and the visual cues of the gendered object in order to understand how a perceiver makes sense of another person. According to Harper and Schoeman (2003), gender schematic individuals hold gender stereotypes and prototypes as more important in how they understand and organize the world. “Gender thus becomes a primary means of resolving social ambiguity.” (p. 518). Thus, one would expect when faced with a gender ambiguous face, without additional social information, a person would rely on the information stored in their own gender schema, such that the following hypothesis is proposed:

H1c: Participants will perceive the gender ambiguous entity in terms of how they
view themselves, such that a masculine will label it as male, feminine will label it as female and an androgynous will label it as androgynous.

Zebrowitz (1997) discussed how gender schematic individuals were more likely to be sensitive to and thus use physical features in determining how they would respond to people, whereas androgynous individuals would be less likely to make judgments based on appearance alone. Lindner, Ryckman, Gold and Stone (1995) found that gender aschematics rated their ideal men and women as androgynous (having both gender stereotypes of instrumental and expressive) while gender schematics rated their ideal men and women based on their related gender stereotypes (men being instrumental and women being expressive). There is thus an indication that how the gender ambiguous face is perceived will be linked to how it is stereotyped as a gendered entity:

H2: If an entity is perceived as either male or female, then that entity will be
categorized using gender stereotypical traits.

There is confusion over exactly how positive it is to be perceived as androgynous (Rajecki, Graaf-Kaser & Rasmussen, 1992). As discussed, numerous studies show that being androgynous can result in better psychological adjustment was well as being perceived as adaptable and likeable. Little, Penton-Vook, Burt and Perrett (2002) discussed how both men and women prefer men that have more feminine features, calling these feminized men more attractive. One possible reason for this is that such men are perceived as being both dominant and cooperative, thus possessing a combination of masculine and feminine traits. Hence, this feminized appearance amongst males may be an indication of androgynous traits that are more socially desirable and acceptable. There is the possibility that an ambiguous face will be perceived as attractive due to this mixture of features and the attribution of an androgynous disposition. However, the backlash effect and expectation violations theory may become important if the face is perceived that it should be male or female, and thus there is upset if it is not. As Jackson (1983) cited, the more a person is prototypical for their gender, the more attractive they are perceived. But if a person is not perceived as being prototypical, would an individual react negatively towards this image because it could not be easily categorized? How attractive a person will perceive the gender ambiguous computer entity results in conflicting hypotheses to be tested in this study:

H3a: Entities visually representing a mixture of both masculine and feminine features will be perceived as more attractive due to a perception of androgynous psychological traits.

H3b: Entities visually representing a mixture of both masculine and feminine features will be perceived as less attractive due to a violation of gendered expectations.

The proposed method for this study can be found in a subsequent post, here…